Product Strategy

Strategy answers "where does the product win in the market," not "what do we build next." It is the layer of market, product, and growth choices that the roadmap then executes. Strategy drives the roadmap, never the other way around.

Market, not marketplace

Plain language. A market is a group of customers who share a need your product addresses. A marketplace is the physical or digital place where buyers and sellers meet.

Amazon and Flipkart are marketplaces. "Time-poor urban professionals who want groceries delivered in 20 minutes" is a market. In AI product work, understanding the market is the strategy, because a capable AI tool aimed at the wrong market expectation still fails.

Types of markets

Product-Market Fit

PMF means the product solves a real problem for a clearly defined market, and that market is willing to pay, adopt, and retain. Investors back PMF, not features. Without it there is no sustainable growth.

Vision, Goals, Bets

The cleanest way to turn a vision into action without jumping straight to features.

• Vision. The aspirational outcome for users and the business. Directional and stable.
• Goals. Concrete, measurable outcomes derived from the vision. Quarterly or yearly, covering both user and business metrics.
• Bets. High-impact initiatives that carry risk and force explicit trade-offs. Bets are where you choose what to fund and what to drop.

The Two-Cycle Flywheel

Plain language. Building the product and marketing the product are not a one-way line. They are two loops that feed each other. Market feedback from the external loop flows back into the internal build loop.

Strategic frameworks

Four tools recur. SWOT diagnoses current reality. GE-McKinsey allocates capital across a portfolio. BCG maps growth against share. Ansoff brainstorms growth directions.

Strategy is not the roadmap

Market & Environment Scanning

Scanning is the continuous monitoring of the forces around a product so you spot opportunities and threats before they show up in your metrics. It runs across four levels, from macro forces down to individual customer sentiment, and AI accelerates it without replacing judgement.

Why scanning is continuous

Scanning answers what is reshaping customer expectations, who is entering from an unexpected direction, which macro trends fit your vision, and what regulation could break your model. Skip it and decisions become reactive.

PESTLE (macro)

Porter's Five Forces (micro)

Evaluates how attractive and how competitive an industry is, so you know where profit erodes and where to build a moat.

Competition runs four levels deep

Most PMs look only at Level 1, which is the classic mistake. Level 4 competitors often evolve into Level 1 threats. ChatGPT started as entertainment and now offers language learning.

Perceptual mapping and sentiment

Perceptual mapping plots competitors on two customer-relevant axes (for Duolingo, gamification vs depth of instruction) to reveal crowded zones and white space. It captures customer perception, not internal intent.

Sentiment analysis uses NLP to classify opinions from reviews, Reddit, and social posts as positive, negative, or neutral. It reveals the why behind churn that raw metrics hide. Duolingo sentiment shows gamification as a strength and shallow grammar depth as a weakness, which feeds straight into a SWOT.

AI in scanning

AI accelerates collection, trend detection, and synthesis at a scale manual methods cannot match. Agentic scanning chains specialised agents into one workflow.

Product Ideation

Ideation is a repeatable process, not a wait for inspiration. Structured sources, formal techniques, and value-innovation frameworks let you generate high-potential ideas on demand and shortlist them before you waste a build.

Seven sources of ideas

Instead of staring at a blank wall, scan these seven reliable wells.

• Customer pain points. Direct frustration signals from users.
• Unexpected occurrences. Anomalies, side effects, and failures.
• Incongruity. The gap between what is and what should be.
• Industry and market changes. New regulations, technologies, business models.
• Demographic changes. Shifts in age, lifestyle, population.
• Changes in perception and values. Evolving societal attitudes.
• New knowledge and technology. Advances like AI, VR, or scientific breakthroughs.

Formal techniques that beat brainstorming

SCAMPER

A structured way to re-imagine an existing product by running seven prompts against it.

Blue Ocean and the ERRC grid

Blue Ocean Strategy creates new market space instead of fighting for share in a crowded "red ocean." It pursues value innovation, higher value at lower cost, so competition becomes irrelevant. The ERRC grid is its working tool.

Three thinking lenses

• Analogical thinking. Borrow a solution from another industry facing the same problem.
• First-principles thinking. Break the problem to fundamental truths and rebuild, instead of optimising a copied template.
• Inversion and systems thinking. Ask "how could this fail," and view the product inside its larger ecosystem to surface hidden risks.

The human-AI collaboration model

AI accelerates the whole journey, from clustering pain points to drafting pitch decks, but the division of labour is fixed.

Beyond core teams, innovation accelerators (internal labs, incubators, internal startups, strategic investments) and open innovation (startup partnerships, academic collaboration, developer ecosystems, hackathons) keep ideas flowing without the inertia of the core business.

Developing Product Propositions

A proposition answers one question. Why should a customer choose this product over every alternative, including doing nothing. It is the bridge between a real customer job and what the team builds, and it has to be testable, not a slogan.

The core mental model

The great PM question is not "what should we build." It is "what progress is the customer trying to make, and how can we uniquely help them succeed."

Jobs To Be Done as the lens

You already use JTBD. The proposition use of it is to separate three job types and design against all three, because a feature list ignores two of them.

• Functional. What they want to do. Trade easily, learn a skill.
• Emotional. How they want to feel. Confident, in control, less anxious.
• Social. How they want to be perceived. Smart, capable, respected.

Value Proposition Canvas

Two halves that must fit. The customer profile describes reality. The value map describes your offer. Fit means your pain relievers match the top pains and your gain creators match the important gains.

Moats, weak and strong

Differentiation has to be defensible, not just different.

Early-stage products compete on minimum value. Scaling products build defensible value through loyalty and switching costs.

The proposition formula

Feasibility Analysis

Most product failures are not design failures. They are feasibility failures, ideas that looked brilliant in a room and collapsed against reality. The discipline tests three lenses, desirable, feasible, viable, before serious resources are committed. All three must be strong.

The three questions (DVF)

Pillar 1: Desirability

Not "users like the idea." It means users actively choose your solution over current alternatives. Strong problems are frequent, painful, and underserved.

• Observe behaviour, not opinions. Ask about past actions, not future intent.
• Measure willingness to pay and current spend on the problem.
• Run MVP pilots and track usage, retention, adoption.

Pillar 2: Feasibility

Capability, not ambition. Three sub-checks.

Pillar 3: Viability

Can it survive as a business? Who pays, when, and why they keep paying. Cost structure and unit economics decide it.

Business model and sizing tools

Lean Canvas (9 blocks) is for fast assumption testing on early ideas, focused on risk and learning speed. Business Model Canvas (9 blocks) is for full operational clarity on mature products. Market sizing nests three numbers.

SOM is the one that matters, because SOM multiplied by price tells you whether the business is worth building. Forecasting uses both top-down (start from macro data, narrow to segment) and bottom-up (start from existing customer data, scale up) to cross-check.

Assumption mapping and the killing blow

Every idea rests on assumptions. Test the high-impact, high-uncertainty ones first.

Rule of thumb. One month of feasibility analysis saves six months of wrong engineering.

Platform Strategies

A platform creates value by connecting interdependent user groups rather than building value internally. The PM job shifts from builder to orchestrator. Success comes from incentives, balance, and governance, not from shipping more features. Not every product should be a platform.

Product mindset vs platform mindset

Types of platforms

Network effects, the real moat

Two kinds, and one warning.

• Same-side effects. More users like me makes the platform more valuable to me. Drives viral growth. WhatsApp, Wikipedia.
• Cross-side effects. More of the other group makes it more valuable to mine. Creates lock-in. Uber drivers and riders.

The chicken-and-egg problem

A platform has to win one side before the other will join. Three sequencing strategies.

Most successful platforms use hybrid sequencing. The deeper truth is platform tension. You cannot optimise all sides equally. Helping customers can hurt margins, helping suppliers can raise costs. PMs choose priorities and accept the sacrifice.

Governance prevents collapse

The orchestration levers are quality control (who can participate), pricing rules (prevent exploitation), standards and policies (trust and safety), and algorithms (how matching happens). Health is tracked on both sides at once.

Product vs platform vs ecosystem

Design Thinking, Personas & Journeys

Module 2 turns a validated opportunity into something usable. Design thinking is the spine. The course frames it as fast, not slow, because you build the right thing once instead of the wrong thing ten times. This topic is revision of ground you already work in, kept to what the quiz tests.

The five stages

You run this loop already. The exam-relevant framing is the sequence and what each stage's strongest method is.

Journey maps and their cousins

A journey map captures actions, touchpoints, thoughts, emotions, and opportunities at each step. The exam tends to test the difference between the map types, which is about perspective.

User Stories & Feature Prioritisation

This is the PM-rigor core of Module 2. Stories carry the user's intent into the backlog, and four scoring frameworks decide what gets built first. The frameworks differ by what they optimise for, so the exam tests when to use which, plus the RICE and ICE arithmetic.

User stories

A short statement of who, what, and why, written from the user's view and focused on benefit, not implementation.

The "so that" clause is the part that matters. It carries the value, which is what prioritisation scores against. A story without a benefit is a task in disguise.

Four prioritisation frameworks

MoSCoW

Sorts features into Must have, Should have, Could have, Won't have. The discipline is that "Won't have (this time)" is an explicit decision, not a backlog graveyard.

Kano

Not all features are equal. Some delight, some are merely expected, some are indifferent. Kano splits them into basic expectations (absence causes anger, presence is unnoticed), performance features (more is better), and delighters (unexpected joy). Delighters decay into expectations over time, which is why yesterday's wow becomes today's baseline.

RICE and ICE, with the math

Value vs Complexity

A 2x2 that the lecture leaned on for fast triage. Plot each feature by value delivered against complexity to build.

Prototyping

Prototyping is test-driving the car before buying it. You commit only once it feels right. The PM-relevant decisions are choosing the right fidelity for the question, using a design system to prototype at scale, and measuring whether the prototype actually worked.

Why prototype at all

It spots flaws early instead of paying for them late, turns abstract ideas into something tangible, gets reactions on real versions rather than descriptions, and lets you fail fast and small. None of this is new to you. The PM framing is that a prototype is a risk-reduction instrument, and its fidelity should match the risk being tested.

Fidelity, matched to the question

Tools by stage

• Early. Paper, Balsamiq.
• Mid. InVision, Proto.io, Canva.
• High. Figma, Adobe XD, Sketch.
• AI-assisted. Uizard, Google Stitch, UX Pilot, Figma Make.

Prototyping at scale

At company scale, consistency comes from a Design Language System (Google Material, Microsoft Fluent, Swiggy's DLS). A DLS provides ready components, which speeds prototyping and smooths the developer handoff, and keeps web, mobile, and offline branding consistent. The system evolves with new patterns as the product grows.

Measuring effectiveness

At scale, prototypes are reviewed for security, accessibility (WCAG audits built into the design stage), and regulatory compliance before rollout, then soft-launched or beta-released before full release.

UX/UI Design & Evaluation

UI is what users see and touch. UX is how the product feels to use. You know both. The exam value here is the named evaluation frameworks the course standardises on, and the working relationship the lecture defined between product owners and designers.

UI and UX, the course split

A brand is more than a logo. It ensures recognition and consistency across touchpoints through logo usage, colour modes, and typography, and it builds trust online and offline.

Evaluation frameworks to name

These are tools you apply already. The course wants them named precisely.

• Nielsen's 10 Usability Heuristics. Rules of thumb for evaluating an interface. Visibility of system status, match to the real world, user control and freedom, consistency and standards, error prevention, recognition over recall, flexibility and efficiency, aesthetic and minimalist design, error recovery, help and documentation.
• Morville's Honeycomb. Qualities of good UX. Useful, usable, desirable, findable, accessible, credible, valuable.
• Double Diamond. The design process shape. Discover and Define (problem space), Develop and Deliver (solution space). Diverge then converge, twice.
• POUR. Accessibility principles. Perceivable, Operable, Understandable, Robust. The backbone of WCAG.

What designers need from product owners

The lecture made this explicit, and it is worth holding because it is the collaboration model the course expects a PM to run.

• Align on the problem before design starts.
• Give intent-based feedback during design, not pixel policing.
• Respect the design stages and involve design early.
• Champion the design when designers are not in the room.
• Treat designers as user advocates, not aesthetic police.

Technology Selection & Architecture

A PM orchestrates technology without being the engineer. This topic is the learner-mode part. The vocabulary of a tech stack, the common architecture patterns, and the build-versus-buy spectrum, so you can hold a real conversation with engineering and judge constraints.

What technology actually is

Not just code. It is tools plus platforms plus frameworks plus infrastructure.

• Tools as utilities. VS Code, testing tools, deployment tools, Figma.
• Platforms as blueprints. Cloud platforms like Azure, AWS, Google Cloud.
• Frameworks as ecosystems. React and Angular (frontend), Flutter and React Native (mobile).
• Infrastructure as backbone. Compute, networking, storage, databases.

The stack has recurring parts. Frontend, backend, database, APIs and integration, security and compliance, testing and QA, DevOps, and increasingly AI-based no-code and low-code platforms.

Architecture patterns

Architecture is the blueprint of how frontend, backend, data, and infrastructure interact. Four patterns cover most products.

Build vs buy, the new spectrum

The old binary has become a gradient. AI changed where the cheap, fast options sit.

The selection method the lecture gave. Inventory the current stack and assess fit, assess the team's competencies and skill gaps, review development practices and tooling, then review infrastructure and cloud capability including security and compliance.

Deployment & Release Management

Deployment is how code gets from a developer's machine to real users safely. The PM does not run the pipeline, but plans and coordinates around it. The four ideas to hold are packaging, environments, configuration, and CI/CD.

The path code travels

1. Packaging

Bundle the application code and its dependencies into one installable, versioned, signed unit. Docker is the standard tool. Packaging matters for three reasons.

• Consistency. The same package tested in staging is the one deployed to production.
• Reproducibility. Developers, testers, and operations all work with identical builds.
• Portability. A container image moves across machines and clouds without the "works on my machine" problem.

2. Environments

Environments de-risk deployment by testing code in safe, controlled spaces before it reaches users. The bare minimum is development, testing, and production, with UAT, pre-prod, and load-test added as needed. They let teams work in parallel and make it possible to track, roll back, and debug issues.

3. Configuration

Configuration is the set of instructions telling an app where to connect. Environment variables (which database), application settings (API endpoints, flags), infrastructure settings (which cluster), and credentials (API keys, passwords, stored in a secret manager). The code stays the same across environments. Only the configuration changes.

4. CI/CD

The tool map

DevOps wraps the whole loop. Planning, version control, CI/CD, then monitoring and observability through logs, performance monitoring, and reporting.

AI & Machine Learning Foundations

Before agents and LLMs, the base layer. Machines do not see words or images, only numbers. Everything in AI flows from that one constraint. Machine learning teaches a system from examples and data instead of explicit instructions.

What AI is, and the paradox

Plain language. AI is a machine doing tasks we associate with human intelligence. The surprising part is which tasks are easy for it.

Moravec's Paradox. AI is good at tasks humans find hard (chess, large-scale calculation) and bad at tasks humans find easy (walking across a cluttered room, picking up a cup). The things evolution made effortless for us are the hardest to automate.

How machine learning differs from coding

In traditional software you write the rules. In ML you supply examples and data, and the system learns the rules itself.

Data outranks algorithms. Feature engineering, choosing which input data matters, is part of modeling, not the algorithm. Tom Mitchell's formal definition is worth holding. A program learns from experience E on task T measured by P, if its performance on T as measured by P improves with E.

Three types of machine learning

Real applications often combine all three. Reinforcement learning with human feedback (RLHF) is now standard in LLMs.

Neural Nets, GenAI & LLMs

Neural networks are how modern AI represents the world in numbers. The 2017 transformer broke the speed limit on sequence models, which made today's LLMs possible. Generative AI gives machines the power to create new data, not just classify it.

Neural networks

Plain language. A neural network is layers of simple units that pass numbers forward. An input layer takes data, hidden layers find the right representation, an output layer gives the answer. The hidden layers are where the learning lives.

Training is a one-time process of adjusting weights. The network starts random, outputs an answer, the difference from the correct answer is the loss, and backpropagation uses that loss to nudge the weights. Repeat until the output is correct. It is data and compute intensive, which is why GPUs are needed for these massively parallel workloads.

From RNNs to the transformer

The hard problem in language is sequence. Older approaches processed words one at a time.

Generative and predictive AI

In a complete AI product these are layers, not rivals. Predictive AI is the base. Generative AI is the icing.

The generative timeline worth recognising. ELIZA (1966), GANs for realistic images (2014), the transformer (2017), GPT-3 (2020), ChatGPT for the public (2022), multimodal text-image-audio-video (2023).

Training LLMs & LLM Security

An LLM is trained in stages, each adding more assistant-like behaviour. Understanding the stages explains why models are helpful but also why they hallucinate and can be attacked. Treat an LLM as an inscrutable artifact built by lossy compression.

What an LLM is

LLM training is a form of lossy compression. The model compresses vast text into parameters, keeping language patterns rather than exact information. That is why it is fluent but can be wrong on specifics. The course framing. Think of LLMs as inscrutable artifacts.

The four training stages

Scaling laws and their limit

Performance improves as parameters and training data grow, which looks like "intelligence for free" without changing the architecture. But this approach may be reaching its limit, which is why new architectures are being explored.

Why LLMs struggle with reasoning

Kahneman's dual-process theory maps cleanly onto LLM behaviour.

Most LLMs behave like System 1 thinkers. Newer reasoning models and agents try to approximate System 2 with reasoning trees and reinforcement learning, but human-like reasoning is still unsolved.

LLM security

A PM cannot fix these but must name them for the engineering and security teams.

• Jailbreaking. Crafted prompts bypass safety filters, sometimes using encoded inputs or universal suffixes.
• Data poisoning. Malicious content in training data plants backdoors. Data must be validated.
• Prompt injection. Hidden prompts in user input or data sources mislead the model into unintended output.

Foundations of AI Agents

An agent is more than a chatbot. It perceives, decides, and acts on an environment, and crucially it makes non-deterministic decisions that adapt. An LLM becomes an agent when you give it the ability to act, tools, and knowledge.

What makes something an agent

Three components define it.

• Sensors. Perceive the environment. Text, images, sensor data.
• Effectors. Act on the environment. Classifications, recommendations, physical actions.
• Decision-making. Makes stochastic, non-deterministic decisions based on inputs and experience.

Autonomy and control

Agents sit on two axes, and the sweet spot for production is usually semi-autonomous with a human in the loop.

An LLM plus three things

An AI agent is an LLM augmented with the ability to perform actions (not just answer), access to tools (web, APIs), and access to knowledge bases. That combination enables goal-directed behaviour beyond a plain LLM.

Agent types and software eras

• Simple reflex. Condition-action rules, no memory. Thermostat.
• Model-based reflex. Keeps an internal model of the world. Roomba.
• Goal-based. Chooses actions to reach a goal. Google Maps.
• Utility-based. Weighs options to maximise a utility function under trade-offs. Ride assignment.
• Learning. Improves from feedback over time. Self-driving cars, spam filters.

Agent Architectures & Environments

PEAS is the practical framework for defining an agent before you build it. Architecture choices trade reasoning depth against speed, and the environment, observable or not, deterministic or not, decides how hard deployment will be.

PEAS, the PM's scoping tool

Define any agent with four questions.

Architecture patterns

Environments decide difficulty

How multiple agents talk

• Direct messaging. Point to point. Scales poorly with many agents.
• Broadcast. Send to all. Easy but can overwhelm channels.
• Blackboard (message queue). Agents post to a shared repository, recipients read what is relevant. Scales best. Used by Netflix and Google.

LLMs as the Brains of Agents

The LLM is the reasoning core of an agent. Getting good behaviour out of it is prompt engineering, a real skill. Tooling and RAG extend the brain beyond its training data so it can act and stay accurate in specialised domains.

What the brain can do

The LLM contributes five core capabilities. Natural language understanding (intent, sentiment), natural language generation (coherent text), reasoning (following frameworks like RICE, still limited), summarisation, and translation.

Prompt engineering

Crafting prompts that give the model clear, unambiguous context. It is a skill that improves with practice. The named types are the exam target.

RAG, grounding the model

Retrieval Augmented Generation connects external documents or databases to the LLM. Instead of relying only on training data, the model retrieves specific knowledge to generate its answer. This is what keeps it accurate in niche domains like medical journals or a company's own history.

Tooling turns an LLM into an agent

Tooling gives an agent access to external systems so it can do more than generate text. An LLM alone cannot act in the world. Add a weather API and it gives real-time answers instead of generic historical ones. Add payments or booking APIs and it can transact.

Tools, Memory, Planning & Frameworks

This is the hands-on layer. Agents need memory to hold context, containers to run consistently, and a choice between visual tools like n8n and code frameworks like LangChain. As a PM your job is to prototype and communicate, not ship production systems.

Agent memory

Containers, the consistency layer

Inspired by shipping containers, software containers isolate an app with a consistent environment. Docker is the standard. The VM comparison is the exam-friendly framing.

Automation vs agents

n8n is a visual workflow automation tool, like Zapier but more powerful and self-hostable. You connect nodes. Trigger, regular, AI agent, and memory nodes. It can call GPT or Gemini, which makes it great for building agents without heavy code. The first agent is a chat trigger, an AI agent node, a chat-model node with your API key, and an optional memory node.

Frameworks, when to use which

Responsible Agents & Alignment

An AI system can hit its goal and still behave wrongly. Alignment is making the machine's purpose match human values. For a PM the job is to name the biases, define what fairness means for your case, and remember that a proof of concept is not deployment.

What alignment is

AI must achieve its goal and avoid unintended harmful side effects. A robot vacuum maximising dust collection might cause damage. Technically succeeding, behaving wrongly. Norbert Wiener warned in 1960 that the purpose put into the machine must be the one we actually desire, not a colourful imitation of it.

Three research approaches. Inverse reinforcement learning (train on appropriate behaviours, not just outcomes), Constitutional AI (give the AI a ruleset, like Asimov's laws), and mechanistic interpretability (understand decisions at the neuron level). All are open problems.

Four ethical frameworks

The Trolley Problem shows why encoding ethics is hard. Different frameworks give different answers, and there is no universal right one. Related distinction. Equality (same support for all), Equity (support by need), Justice (remove systemic barriers). You must decide which your system optimises for and test edge cases explicitly.

Bias, named for engineering

A PM does not fix these but must name and explain them.

• Historical bias. Training data is outdated.
• Representation bias. Data reflects only a subset who were sampled.
• Measurement bias. You measure what is easy, omitting key variables.
• Algorithmic bias. The wrong algorithm skews results.
• Aggregation bias. Aggregate conclusions that fail at the individual level (Simpson's Paradox).
• Population bias. Key demographic groups excluded entirely.

Model evaluation and deployment

Project Management & Methodologies

Execution is where strategy either ships or dies. This topic sets the vocabulary. Project versus product versus program, the lifecycle every project runs, and the three methodology families. The methodology choice is driven by how much requirements are likely to change, not by preference.

Project, product, program, portfolio

You held the PM-versus-PjM split in Module 1. Here is the object each one manages.

The project lifecycle

Monitoring and controlling runs in parallel with execution, not after it. That overlap is the part exams probe.

Three methodology families

Kanban's core discipline is the work-in-progress limit. Capping how many items are in any column at once exposes bottlenecks and forces the team to finish before starting, which is the opposite of the busy-but-nothing-ships failure mode.

Agile & Scrum Framework

Agile is a mindset of four values. Scrum is the most common framework that operationalises it, with fixed roles, artifacts, and ceremonies. The exam tests the difference between the two and who owns what, especially the product owner versus scrum master split.

Agile is a mindset, not a process

The Agile Manifesto sets four values. Each prefers the left without discarding the right.

• Individuals and interactions over processes and tools.
• Working software over comprehensive documentation.
• Customer collaboration over contract negotiation.
• Responding to change over following a plan.

Scrum's three pillars and values

Scrum stands on transparency, inspection, and adaptation, supported by five values. Commitment, courage, focus, openness, respect. Empiricism is the underlying idea. Knowledge comes from experience, and decisions are based on what is observed.

Roles, the ownership map

Artifacts and ceremonies

The sprint itself is the container event, usually one to four weeks. Inside it, planning opens, the daily standup syncs the team in 15 minutes, the review shows the increment to stakeholders, and the retrospective improves the process. Review inspects the product. Retro inspects the team.

Sprint Planning, Estimation & Tracking

This is the quantitative heart of agile execution. Stories get sized in relative points, not hours. Velocity turns points into a forecast. Burndown shows whether the sprint is on track. Jira is where it all lives. The exam tests why points beat hours and how velocity is used.

Good stories pass INVEST

A backlog story should be Independent, Negotiable, Valuable, Estimable, Small, and Testable. If it fails one, it usually needs splitting or rewriting before it enters a sprint.

Why story points, not hours

Story points measure relative effort, complexity, and uncertainty, not clock time. Two reasons this beats hour estimates.

• Humans compare better than they measure. People are poor at "how many hours" and good at "is this bigger than that." Relative sizing plays to the strength.
• Points are person-neutral. A senior and a junior disagree on hours but can agree a task is twice another. Points stay stable across who does the work.

Velocity and capacity

Capacity is how much the team can actually take on next sprint given leave, holidays, and other commitments. Velocity says what the team usually delivers. Capacity adjusts that for this sprint's reality. Plan to capacity, forecast with velocity.

Tracking with burndown

A burndown chart plots remaining work against time. The ideal line falls steadily to zero by sprint end. Reading the gap between actual and ideal is the daily health check.

Roadmap, Backlog & PM Skills

The roadmap connects strategy to the backlog. The exam-relevant shift is from date-driven timeline roadmaps to outcome-driven Now-Next-Later, and the discipline of a backlog that is prioritised and groomed rather than a wishlist. Two communication structures, SCQA and STAR, round out the PM toolkit.

From vision to task

Work decomposes the same way each time. An Epic is a large body of work, a Feature is a shippable slice of it, a Story is a user-facing unit of value, and a Task is the technical step under a story.

Roadmap types

The backlog is a decision, not a wishlist

A healthy backlog is prioritised, sized, and continuously groomed. Backlog grooming (refinement) is the recurring work of clarifying, splitting, re-estimating, and re-ordering items so the top is always ready to pull into a sprint. Prioritisation reuses the Module 2 frameworks. MoSCoW for deadline cuts, Value vs Effort for fast triage, RICE for defensible ranking.

Two communication structures

SCQA structures a proposal or a strategy memo so the recommendation lands with context. STAR structures how you describe past work, in reviews or interviews, so the outcome is explicit rather than implied.

Stakeholder Alignment & Leading without Authority

A PM owns outcomes but rarely owns the people who deliver them. The job runs on influence. This topic covers mapping stakeholders by interest and influence, managing each quadrant differently, and the specific moves that build influence when you have no formal authority.

Map before you manage

Not every stakeholder gets the same time. Plot them on interest against influence and treat each quadrant differently.

Leading without authority

The PM has responsibility without command. Influence is built, not granted. The moves the course stressed.

• Credibility through competence. Be the person who understands the customer, the data, and the trade-offs better than anyone in the room.
• Shared goals over orders. Frame the work as the team's win, not your request. People commit to outcomes they helped shape.
• Reciprocity and trust. Help others hit their goals and the support returns when you need it.
• Data over opinion. When you cannot pull rank, evidence is the lever. A clear metric ends a debate authority cannot.
• Transparency. Surface trade-offs and constraints openly. Hidden reasoning breeds resistance.

Managing up

Stakeholder work runs in every direction, including upward. Give leadership the decision, the options, and your recommendation, not raw status. Bring problems with a proposed path, not just the problem. The SCQA structure from Topic 04 is the tool here. Situation, complication, question, answer, so a busy executive can decide in one read.

Ethics in Product Management

Legal and ethical are not the same line. A PM ships choices that shape behaviour, attention, and trust, so ethics is a product decision, not a compliance footnote. This topic covers the legal-versus-ethical gap, the frameworks for reasoning through it, and the specific traps in AI products.

Legal is the floor, not the standard

Frameworks for reasoning

The same four lenses from Module 3 apply to product decisions, not just AI alignment.

These conflict on hard cases, which is the point. A growth tactic can be utilitarian-positive (more users served) and deontologically wrong (it deceives). Naming the conflict is how a PM makes the trade-off visible instead of defaulting to whatever ships fastest.

Common product traps

• Dark patterns. Interfaces designed to trick. Hidden unsubscribe, pre-ticked consent, confirm-shaming. Legal in many places, corrosive to trust.
• Engagement at any cost. Optimising for time-on-app can amplify harm, addiction, and outrage. The metric succeeds while the user loses.
• Privacy erosion. Collecting more data than the feature needs because it is technically possible.
• Manipulative defaults. Defaults that serve the business over the user. Defaults are powerful precisely because most people never change them.

AI-specific ethics

AI products add a layer of risk that maps to Module 3. Bias in outcomes across protected groups, lack of transparency in automated decisions, accountability gaps when an agent acts wrongly, and consent over how data trains models. The PM job is to define what fairness means for the specific use case and to insist a proof of concept is not deployment until it is monitored for these harms.

Marketing Strategy Framework

A working marketing strategy needs three things in sequence. A clear read of your environment (5Cs), a deliberate choice of who you serve (STP), and a portfolio view of where to invest (GE-McKinsey).

The 5Cs Framework

Plain language. Before you do anything, look at five things that shape whether your product will work. Your customer, your company, your competitors, your collaborators, and the broader context around all of it.

Formal. A situational analysis covering:

• Customer. Needs, segments, buying behaviour, decision criteria.
• Company. Internal capabilities, resources, brand equity, cost position.
• Competitors. Direct rivals, indirect substitutes, new entrants.
• Collaborators. Distributors, suppliers, channel partners, alliances.
• Context. PESTEL forces (political, economic, social, technological, environmental, legal).

STP: Segmentation, Targeting, Positioning

Plain language. Divide the market into groups, pick which groups you will serve, then decide what you want to mean to them.

The four segmentation bases

Psychographic segmentation is the strongest of the four because it gets at why people buy, not just who they are. Two 35-year-old men in Bangalore with the same income can have completely different product preferences depending on values and lifestyle.

GE-McKinsey Matrix

Plain language. A 3x3 grid for deciding where to invest, where to hold steady, and where to exit across multiple product lines or business units.

Formal. A portfolio analysis tool plotting business units on two axes. Industry attractiveness (market size, growth rate, profitability, competitive intensity) and competitive strength (market share, brand strength, customer loyalty, cost position).

Branding

A brand is not a logo. It is the emotional and psychological relationship a customer holds with a company, built through consistency across touchpoints and time. Strong brands earn premium prices and survive product mistakes.

What a brand actually is

Plain language. A brand is the gut feeling a customer has when they hear your name. The logo is the trigger. The feeling is the brand.

Formal. Per Prof Shelby Hunt's commitment-trust theory of marketing, a brand is the trust accumulated through consistency across every customer touchpoint over time. Trust is earned slowly through repeated delivery on a promise. It is destroyed quickly through one breach.

Brand equity drivers

• Differentiation. How distinct your brand is from alternatives.
• Relevance. How well it fits the customer's life and needs.
• Awareness. How readily customers think of you in the category.
• Esteem. How highly the brand is regarded.
• Emotional connection. The non-rational pull a customer feels.

Functional vs Symbolic Differentiation

Brands can earn their premium two ways. By being functionally better, or by meaning something the customer wants to wear or signal.

Kapferer's Brand Identity Prism

Plain language. A six-sided map of what a brand is. Three sides are how the brand presents itself (sender), three are how the customer experiences it (receiver).

Smell of the Place

Sumantra Ghoshal's concept. The contrast he used was Calcutta in summer (heavy, oppressive, draining) vs Fontainebleau in spring (light, energising, alive). The same person performs differently in different environments. Brands work the same way internally.

Customers feel the smell of the place through every employee interaction. A company built on compliance, control, and contract cannot deliver a brand experience that feels warm and trustworthy, no matter what the ads say.

Services Marketing Triangle

For service businesses (which is most digital products), the brand is delivered through three connected marketing motions.

• External marketing. Promises made to customers (ads, positioning, brand campaigns).
• Internal marketing. Equipping and motivating employees to actually deliver on those promises. Training, culture, tools.
• Interactive marketing. The moment the employee meets the customer. This is where the brand is either kept or broken.

Promotion Strategy & the 7Ms

Promotion is not "let's run ads". It is a seven-question checklist that forces you to be specific about what you are trying to achieve, who you are talking to, where, what you will say, how much you will spend, when, and how you will measure whether it worked.

The session was titled "7P" in the program calendar, but the framework actually taught was the 7Ms. Use 7Ms.

The 7Ms Framework

1. Mission

What the campaign is actually trying to do. Awareness, trial, repeat purchase, or sales conversion. These are not the same thing and require very different campaigns. For a high-involvement product like a baby health product, you cannot skip awareness and trial and go straight to sales. The customer needs to know you, trust you, and try you before they will buy you repeatedly.

2. Market

Who is the audience. This is where communication personas come in.

3. Media

Where to reach them. Digital, traditional, or both. The question is not "what is trendy" but "where does my market spend attention". An older rural segment may be reached more efficiently through regional television and local print than through Instagram, even in 2026.

4. Messaging

What to say. The biggest distinction here is brand ad vs sales ad.

5. Money

How much to spend. The lecturer taught ground-up budgeting, which is more useful than top-down "we have 10 crores".

6. Moment

When to say it. Timing matters more than most people assume. The lecturer's example. LinkedIn posts on Friday evening underperform because the audience has logged off for the weekend. Tuesday and Wednesday mornings outperform. Same content, very different reach, just by changing when it was posted.

7. Metrics

Did it work. This is where most marketing fails. Vanity metrics (likes, impressions, video views) are not the same as outcome metrics (conversions, trials, repeat purchases). Define the metric upfront, before the campaign launches. Otherwise you will retroactively define success around whatever number looks best.

Digital vs Traditional Marketing

Digital and traditional are not opposites, they are complements. Most successful campaigns blend both.

Digital Marketing

Build the digital asset first, drive traffic to it second. Search marketing is SEO plus SEM plus GEO, and you need all three running in parallel. Social media is its own ecosystem with platform-specific algorithms and an emerging trust crisis.

The Sequence: Assets First, Traffic Second

Before spending on advertising, build the digital assets traffic will land on. Website, e-commerce store, mobile app, social media presence. Without these, even free traffic has nowhere to go.

Search Marketing: SEO, SEM, GEO

The three search disciplines do different jobs. Run all three.

Keywords: short-tail vs long-tail

Long-tail keywords are usually the better starting point for a small brand. Less competition, higher conversion. Tools to find them. Google Ads Keyword Planner for volume estimates, SimilarWeb.com for competitor traffic analysis.

SEO and the EEAT Framework

Google ranks pages on four criteria, abbreviated EEAT.

• Experience. Does the content reflect first-hand experience?
• Expertise. Is the author qualified to write on the topic?
• Authoritativeness. Do other credible sources cite this page?
• Trustworthiness. Is the site secure, transparent, and accurate?

Tactics that used to work (keyword stuffing, link farms) now get penalised. The modern playbook. Write genuinely useful content, get cited by authoritative sources, and aim for the answer box at the top of search results.

SEM and the Bid + Quality Score Equation

SEM placement is not determined by who bids highest. Google ranks ads using a combination of bid amount and quality score (relevance, expected click-through rate, landing page quality). A lower bid with a higher quality score can outrank a higher bid with a poor landing page.

GEO: Generative Engine Optimization

The emerging discipline. AI search engines (Claude, Perplexity, Gemini, ChatGPT search) generate answers rather than serve links. Optimising for them requires different tactics.

• Q&A format. AI models pull from content structured as questions and answers.
• Citations. Content that cites credible sources gets reused as the source for AI answers.
• Conversational language. Match how users phrase questions naturally, not how SEO keyword tools phrase them.

SEO is not dead. Roughly 96% of searches in 2026 still happen on Google. GEO is additive, not a replacement.

The Ansoff Matrix

While discussing Atomberg's growth, the lecturer introduced the Ansoff Matrix as the framework behind product expansion decisions.

Social Media Marketing

The Facebook algorithm in plain language

Facebook (and by extension Instagram) ranks what you see based on four factors:

• Affinity. How close is the relationship between you and the poster.
• Interaction. Have you engaged with this person or page before.
• Content type. Some formats (video, carousel) are weighted higher at different times.
• Recency. Newer content beats older content.

You cannot trick the algorithm long-term. You can only design content that scores well on these four dimensions naturally.

Influencer tiers and the maths of cost-per-engagement

The lesson. Nano-influencers often deliver more value per dollar than mega-influencers because their engagement rates are dramatically higher and their audience trusts them as a peer, not a celebrity.

Platform roles

• WhatsApp / WhatsApp for Business. Community building. QR codes on packaging that drop customers into caregiver groups.
• YouTube. Educational long-form. Second-largest search engine in the world. Backlinks from YouTube descriptions help SEO authority.
• X / Twitter. Social listening, hashtags, complaints. High public risk if you have unhappy customers (Q Ratio later closed their X account for this reason).
• Reddit / Quora. Q&A, niche communities. Strong SEO value for long-tail keyword discoverability.
• Instagram. Visual storytelling. The "lifestyle" arm of most consumer brands.
• LinkedIn. B2B, professional credibility, thought leadership.

Episodic content example: KitKat

KitKat ran a heist storyline starting April 1, releasing short episodic videos over weeks. The point was not the chocolate. It was attention. Episodic content trains an audience to come back, which compounds organic reach without paid spend.

Integrated platforms

Tools like Hootsuite and Zoho Social manage scheduling, analytics, and engagement across platforms from one dashboard. For any brand running on three or more platforms, manual posting becomes the bottleneck.

The Posting Zero Trend

An emerging behavioural shift. Roughly 10% of social media users globally posted nothing in 2025, up sharply from prior years. Three reasons:

• Enshittification. Platforms became commercial and lost the "share with friends" feel.
• Perception anxiety. Users fear judgment, comparison, professional consequences.
• Dead Internet Theory. A genuine concern that bot traffic now exceeds 50% of all internet activity. Posting into a sea of bots feels pointless.

For marketers, this matters because if audiences are posting less, user-generated content campaigns become harder. The mechanics of word-of-mouth marketing are shifting underneath the strategy.

Consumer Behaviour

Three psychological drivers shape every purchase. Motivation, perception, and memory. Most brands sell features when they should sell benefits. The customer journey from attention to satisfaction varies dramatically by product involvement.

The AIDAS Model

The customer journey for any product moves through five stages.

The time spent in each stage varies wildly by product involvement.

High-involvement products (long journey) demand sustained attention across all five stages. A car ad that drives Action without earning Desire wastes spend.

The Three Psychological Drivers

1. Motivation: Maslow's Hierarchy

Customers buy to satisfy needs, and those needs sit on a hierarchy.

This matters for messaging. A baby products brand sells to safety (parents protecting their child). A luxury watch sells to esteem. The same product can be repositioned to a different level. A car can be sold on safety, belonging, or self-actualization depending on the segment.

2. Perception: Not Reality, but What Customers Perceive

Customers do not respond to reality. They respond to their perception of it. Two phones with identical specs can be perceived very differently depending on brand, packaging, and social proof.

Tools to build perception:

• Logos and taglines. The Nike swoosh signals performance before any context is added.
• Social proof. Reviews, ratings, "10 million users" badges.
• Influencers. Borrowed credibility. Authenticity test. Does this influencer actually use this product?

3. Memory and Learning

Three types of memory shape what customers remember about a brand.

• Sensory memory. Smell, sound, texture. Starbucks' coffee aroma when you walk in is a deliberately engineered sensory memory.
• Short-term memory. The 30 seconds after seeing an ad. Most ads die here.
• Long-term memory. Jingles, taglines, repeated exposure. "I'm lovin' it" sits in long-term memory for billions of people.

Three learning mechanisms move information into memory:

The FAB Framework

Features, Advantages, Benefits. Most marketing fails because it sells features when it should sell benefits.

FAB in B2B

Different stakeholders in a B2B sale care about different levels of FAB.

• Procurement cares about features (spec comparison, compliance).
• Operations cares about advantages (will it actually save time).
• CEO or CFO cares about benefits (will it move the business).

A pitch deck that targets only one level of FAB will fail in front of the other stakeholders.

ELM: Elaboration Likelihood Model

When customers are highly involved in a decision, they think rationally (central route). When they are not, they respond to emotional or peripheral cues (peripheral route).

This drove the Q Ratio messaging decision. Baby products are high-involvement. Use the central route. Trust, safety, evidence-led messaging.

Product-as-hero vs User-as-hero

A long-running messaging question. Q Ratio A/B tested both for baby products:

• Product-as-hero. Hero shot of the product, specifications-focused.
• User-as-hero. Mother holding baby, product in supporting role.

User-as-hero (mother and baby) won. For high-involvement, emotional categories, the user belongs at the centre of the frame.

AI's Impact on Consumer Behaviour

How AI is reshaping the customer journey:

• Hyper-personalization. Every customer sees a different version of the product.
• Predictive intent. AI guesses what you want before you search for it.
• Conversational commerce. Buy through chat, not through a browse-and-add-to-cart flow.
• Frictionless experience. One-tap purchases, AI agents handling the comparison work.

Accelerating Growth

Every new product starts with a credibility problem. The growth journey moves from signaling (earn trust) through launch, growth hacking, growth loops, hyper-scaling, and finally a durable moat. Skip any stage and the product stalls.

Calendar note. This topic maps to the "Building a Scalable Business" sessions in the program calendar.

The Full Growth Sequence

The running case was VoiceGen, a real Bengaluru-based AI speech therapy startup (name changed by the lecturer). Patients with stroke, autism, stammering, or Parkinson's use the app for AI-coached speech therapy in multiple Indian languages.

The Four Types of Novelty

Every new product faces one or more credibility problems. The lecturer added a fourth (provenance) to the standard three.

Signaling: Overcoming Novelty Without a Track Record

Plain language. You cannot prove you are good yet. Signaling is the act of giving credible markers that lower the buyer's perceived risk.

Practical signaling tactics:

• Board of advisors. Recruit credible names. This is not a board of directors (legal term) but a public-facing list of respected people who vouch for you.
• Academic and hospital partnerships. Tie-ups with IIT, AIIMS, or major hospital chains.
• Early data. Even imperfect data is better than no data. VoiceGen shared 83% accuracy benchmarks from 50 patients early on.
• Visible thought leadership. LinkedIn presence, blogs, research publications.
• Influencer endorsements. Senior clinicians publicly using the product.

Pre-launch and Launch

Pre-launch checklist (work backwards from T-zero)

• Beta testing with a small group (VoiceGen ran 12 weeks with real patients).
• MVP deployment with measurable benchmarks (VoiceGen's 50-patient trial showed 83% accuracy, below the 92% target. Back to the lab).
• Legal and regulatory clearances (patient data handling, compliance).
• Pricing validation.
• Content infrastructure (FAQs, onboarding, website).
• Internal alignment across operations, support, and product teams.

Soft Launch vs Big Bang

VoiceGen soft-launched with six hospitals, demo booths in rehab centres, and clinician workshops. The right call for a medtech product where negative first impressions are very hard to recover from.

Growth Hacking

Plain language. Short-term, low-cost tactics to get the first wave of customers to try the product.

VoiceGen's growth hacks. Seeding neurologists, before-and-after voice clips, WhatsApp caregiver groups, gamified recovery challenges, hospital desk flyers, referral fees.

Growth Loops

Plain language. Mechanisms where existing customers bring in new customers automatically. The product itself does the marketing.

VoiceGen's clinical referral loop is the textbook example for medtech. Doctors got a dashboard showing patient progression. When a new patient walked in, the doctor showed prior patients' improvement curves (anonymised) to build trust. The doctor was marketing themselves, but in the process, was also marketing VoiceGen.

Hyper-Scaling

Plain language. Grow 2x or 3x year-over-year while keeping costs flat. The term originated in cloud infrastructure (hyperscalers like AWS) but now broadly means non-linear growth.

VoiceGen's hyper-scaling moves. Hospital chain partnerships, Ayushman Bharat integration, free academic licenses to speech therapy training institutes. The academic license play is the same playbook used by SAP, Salesforce, and Spotify. Train students on the tool. They graduate. They bring it into their workplaces. Free distribution for the next 20 years.

Durable Moats

Plain language. The barrier that stops a well-funded competitor from copying you. Borrowed from medieval castles, where the moat (and the crocodiles) kept invaders out.

The term was popularised in business by Warren Buffett. Five types of moat:

• Trust. The hardest to build, hardest to copy. Brands like Tata, Patagonia.
• Network effects. Each new user makes the product more valuable for existing users. Covered in Module 1 (Platform Strategies).
• Data. Proprietary data that compounds over time. Google search data is the canonical example.
• Regulatory. Licenses, certifications, approvals that take years to obtain.
• Specialisation. Deep expertise in a narrow domain.

VoiceGen's moats:

• Proprietary clinical data. Every patient session improves the AI model. The lecturer advised them to drop prices to acquire more users and build an unassailable data moat before competitors catch up.
• Multilingual capability. Fine-tuning AI for Telugu, Marathi, Tamil, Bengali phonetics is genuinely difficult. The more languages they support, the harder it gets for a newcomer to compete across the same territory.

Sustainable Product Management

Two parts. First, intellectual property as the legal moat that protects everything else. Second, sustainability as a design-stage discipline where 75% of the impact is decided before the product is ever made.

Part 1: Intellectual Property

Plain language. IP is the legal barrier that stops competitors from copying what you built. Without it, anything digital can be cloned in weeks.

The four main types

Patents: what is and is not patentable

Other practical patent points:

• First-to-file. Patents go to whoever files first, not whoever invented first. File early.
• Pharma timing. Drug patents are filed at the start of human trials. By the time the drug is approved (14 years later), only 6-7 years of patent life remain. This is why approved drugs are expensive.
• Cost. Basic patent filing in India. Rs 15-25K for a starter filing. Serious work runs into lakhs.

Trademarks and the knock-off problem

The lecturer's anecdotes:

• Holiday Inn in Suzhou, China. The lecturer stayed there, found it sub-standard, called Holiday Inn's regional manager, who confirmed they had no property in Suzhou. The hotel was a complete brand knockoff.
• Jaipur sweets (JMB / Jambi). One legitimate brand spawned 30+ "Jambi" knockoffs in Udaipur alone.
• Bawarchi biryani (Hyderabad). Wi-Fi Bawarchi, Green Bawarchi, Pink Bawarchi all coexist. None are the original.
• Bikanerwala. A family dispute that became a years-long trademark fight.

India's trademark enforcement is improving but slow. File trademarks early. Defend them visibly.

Trade secrets

The Coca-Cola and Pepsi formulas sit in physical vaults in Atlanta. Only a small number of people know the full formula. No patent, no expiry. Just secrecy.

Open source is not "free for anything"

The NDA trap with investors

Most major VCs refuse to sign NDAs. They argue the volume of pitches they see makes NDAs unworkable. If they like your idea, they may share it with their portfolio companies that are competitors. The lecturer's company Avika saw this firsthand. Three of five investors asking for meetings were investors in their direct competitors. Be careful what you share, and assume what you share will travel.

Part 2: Sustainability

Definitions and how they connect

Sustainability. Meeting today's needs without compromising the ability of future generations to meet theirs.

The Triple Bottom Line

Three pillars to evaluate sustainability against.

• People. Social inclusivity, fair labour, accessibility, community impact.
• Planet. Environmental impact, carbon, resource use, biodiversity.
• Profit. Economic viability over the long term.

Circularity

The waste of one process becomes the input of another. Opposite of linear "take-make-dispose".

Real examples:

• Paper mill fly ash to bricks. Combustion waste becomes building material for the factory.
• Effluent treatment. Chemical plants cleaning wastewater and reusing it in their own processes.
• Airport bottle refill stations. Reduces single-use plastic by enabling reuse.
• Kabadiwalas (Indian recycler collectors). India had a circular economy in practice for decades before the term existed.

The 75% Rule

The 6R Framework

The Bamboo Toothbrush Case

Does sustainability help business?

Empirical evidence says yes. Examples cited:

• S&P data. ESG-strong companies outperform broader market indices over a decade.
• Gallup data. Purpose-led companies have significantly lower employee turnover.
• Patagonia, Fab India. Built their entire brand moat around sustainability. Brand trust as a competitive advantage that competitors cannot easily replicate.

The AI Infrastructure Problem

The lecturer's framing was striking. $1.5 trillion was spent on AI infrastructure in 2025 (roughly one-third of India's GDP). $8 trillion projected over the next five years. Yet the system is hitting bottlenecks.

Power and water

• Several US states (Maine and others) have banned new data centres entirely.
• The Tamil Nadu chief minister reportedly declined a major data centre proposal because it would drain power and water from manufacturing (which employs tens of thousands) for a few hundred data centre jobs.
• Andhra Pradesh has approved data centre proposals but lacks the power infrastructure to run them.
• In Nevada, aircraft graveyards are being dismantled. The jet engines are being repurposed as power generation for data centres. That is how desperate the energy situation has become.

The data problem

All human-generated internet data has essentially been used to train current AI models. New training data is increasingly synthetic data (AI-generated). The problem. Training AI on AI-generated data narrows the model's outputs mathematically. The model becomes worse at the extremes and clusters around the mean.

The energy stagnation problem

No major energy breakthrough in 50 years. Nuclear is the only credible large-scale option, and only China is building meaningfully. This is the underlying bottleneck for AI scaling that the trillion-dollar-investment narrative does not address.

Sales & Distribution Strategy

Distribution is the path from manufacturer to customer. Direct vs indirect, B2B vs B2C, intensive vs selective vs exclusive. Pick the wrong path and the product never reaches buyers. Pick the right one and watch conflicts emerge between partners you now depend on.

The Channel Choice: 2x2 of Options

Direct vs Indirect: The Tradeoff

Omnichannel: Multiple Paths, One Brand

Most modern brands run several channels in parallel. Lenskart runs 2,000+ physical stores, its own e-commerce, marketplace listings (Amazon, Flipkart), and a social media presence simultaneously.

Why bother with so many?

• Different customers prefer different channels. Older buyers may want to visit a store. Younger buyers may want an app.
• Brand recall through touchpoints. Each visible channel reinforces top-of-mind awareness. When you think "online food delivery", you think Swiggy first because Swiggy is everywhere.
• Risk reduction. If one channel underperforms, others keep revenue flowing.
• Data control. Your own channels give you data marketplaces will not share.

B2B vs B2C: Why Channels Look Different

The B2B Sales Orientation: Farmer vs Hunter

Neither orientation alone is enough. A team of all hunters churns existing customers. A team of all farmers stops growing. Great salespeople have both, and the skills can be developed.

Increasing Customer Switching Costs

If a customer can leave easily, you cannot defend your account. Switching costs come from:

• Learning costs. The team has invested time learning your system. Switching means re-learning.
• Economic costs. Discounts, bundled services, custom pricing that disappear on exit.
• Relationship costs. Deep trust with your team is hard to replicate.
• Brand costs. "We use IBM" carries credibility. Switching loses that signal.

The HP/IBM Migration Model

IBM is the canonical end-state. Some Fortune 500 clients invite IBM into their strategic planning sessions because IBM understands their technology roadmap better than they do. At that point switching is unthinkable.

Distribution Intensity: How Many Outlets?

With exclusive distribution, the partner choice is critical. One bad partner can damage the brand across an entire market. Due diligence is non-negotiable.

Agency Problems: When Partners Misbehave

You (principal) hire a distributor or retailer (agent) to act in your interest. Two things can go wrong.

The lecturer's example of moral hazard. You offer a retailer 50% extra margin to recommend your product to every customer. The retailer accepts the margin, pockets it as a discount to undercut competitors, and recommends nothing. You are paying for promotion that is not happening.

Fixes

• Due diligence (for adverse selection). Verify partner capability before signing. More important for exclusive partners.
• Mystery shopping (for moral hazard). Send your own team to retail outlets as customers. Observe the actual behaviour.
• Audits. Periodic verification of promotional activity.
• Performance-based incentives. Pay on results, not on activity claims.

Channel Conflict

The Free-Rider Problem

Free-rider fixes

• Exclusive territories. Geographic exclusivity so retailers do not compete head-on.
• Cooperative advertising funds. Compensate retailers specifically for educational activity.
• Product versioning. Different SKUs for different channels. Samsung sells Galaxy F series only on Flipkart. Customers cannot compare prices directly.
• Performance rewards. Better margins to retailers who demonstrably do the work.

Customer Service

Customer service is not a support team's problem. It is a cross-functional discipline. When delivery fails, technology fails, supply chain fails, and support fails simultaneously, the customer does not separate them. The brand takes the hit, regardless of which function caused the breach.

The Blinkit Complaint Case

What failed:

• Promise broken. 10 minutes became more than 60.
• Technology failed. Tracking estimates fluctuated wildly.
• Operations failed. Delivery partner assignment was poor.
• Support failed. No conversation history between agents.
• Empathy missing. No acknowledgment of the urgency.

Fixing only the support response misses the point. The problem is systemic.

What Are You Actually Selling?

Before designing service, get clear on what the customer is buying. Almost never the literal product.

This is Clayton Christensen's Jobs-to-be-Done framing. Customers hire products to do a job. Get the job right and service expectations follow naturally.

The Three Pillars of Customer Centricity

The tacit knowledge problem

Knowledge in heads is not knowledge in systems. The support team knows what customers complain about most. The engineering team knows what breaks most often. The sales team knows what objections customers raise. If these teams do not talk, that knowledge stays trapped. CRM tools (Zoho, Salesforce) are partly attempts to surface tacit knowledge across functions.

The Customer Journey

Zero Moment of Truth (ZMOT)

Coined by Google in 2011. Before any customer enters a store or app, they research. Reviews, YouTube videos, articles, comparison content. This research phase is the Zero Moment of Truth. If your brand does not show up in that research, you are not in the consideration set.

Procter & Gamble had previously identified:

• First Moment of Truth. Entering the store.
• Second Moment of Truth. The purchase decision.

Google added the zeroth. The research that happens before either.

Service Encounter

The actual experience. This is where cross-functional alignment is tested. Operations, technology, support, logistics all have to perform. If any one fails (the Blinkit case), the whole experience fails.

Post-Purchase: Cognitive Dissonance

After buying, customers often experience uncomfortable doubt about their decision. "Did I pick the right car? Maybe I should have got the Compass instead of the Creta."

Brands that reduce cognitive dissonance turn buyers into advocates. Tactics:

• Follow-up communication. "How are you finding the car? Here is what other Creta owners love."
• User communities. Owner groups where positive experiences get reinforced.
• Educational content. Help the customer get more value from their purchase.

Why this matters. Customers continue to research even after they buy, looking for validation. If your brand is visible during that post-purchase research, you reduce dissonance and build loyalty.

Zone of Tolerance

Every customer has two service levels in their head.

• Desired service level. What they wish for.
• Adequate service level. The minimum they will accept.

The gap between these two is the zone of tolerance. If you deliver inside the zone, the customer is satisfied. Below the zone, complaints. Above the zone, occasional delight.

When the zone is narrower (less forgiving)

• Service is expensive.
• Customer involvement is high (emergency, urgent need).
• Many alternatives are available.
• New product was launched with high hype.
• Customer has had prior bad experiences.

When the zone is wider (more forgiving)

• Service is free or low-cost.
• Customer involvement is low.
• Few alternatives exist.
• First-time user, expectations not yet set.
• Customer trusts the brand deeply.

Strategy: under-promise and over-deliver

Narayana Murthy's motto when building Infosys. Set realistic expectations and beat them consistently. This widens the perceived zone by anchoring the desired level lower.

Measuring Service Quality: The RATER Framework

Five dimensions customers use (often unconsciously) to judge service quality. Developed by Prof Parasuraman.

Reliability is the most important. A brand that promises 10 minutes and delivers 40 destroys reliability, which is the foundation everything else sits on.

E-service quality extensions

For digital products, RATER extends with:

• Efficiency. How quickly can the customer complete their task.
• Fulfillment. Was the order delivered as promised.
• System availability. Does the platform work when needed.
• Compensation / Recovery. When things go wrong, is the resolution satisfactory.

Pricing Strategy

Price is the only marketing element that generates revenue. It is also the most flexible. Three strategies (cost-based, competition-based, customer value-based) plus price elasticity plus new-product positioning (skimming vs penetration) are the toolkit.

What Price Actually Is

Price is not just money. It is the sum of all values a customer gives up.

• Money. The actual rupees paid.
• Time. Waiting, travelling, learning curves.
• Effort. Switching cost, onboarding pain.
• Risk. Uncertainty about whether the product will work.
• Experience. The 20-rupee chai vs 300-rupee Starbucks coffee. You are not paying for the coffee.

The Three Pricing Strategies

Cost-Based Pricing

Two methods:

• Markup pricing (cost-plus). Costs + fixed margin. A car costs Rs 20 lakh to make, add 20%, sell at Rs 24 lakh.
• Break-even pricing (target return). Work backwards from desired profit. Need Rs 20 lakh profit on a Rs 1 crore investment, work out volume and price.

Competition-Based Pricing

Price relative to competitors.

Customer Value-Based Pricing

The most powerful and the hardest to do well. Price based on what the customer believes the product is worth, not on what it costs to make.

How to measure willingness to pay

• Customer surveys and feedback. Direct conversations reveal what customers value.
• A/B testing. Show two prices to two cohorts. Measure adoption. Dropbox might test basic plan vs family plan at different prices.
• Van Westendorp Price Sensitivity Meter. Four-question survey covered in Module 6 (Market Research).

Price Anchoring

Present a high reference price next to your actual price to make the actual price feel reasonable.

Anchoring shows up everywhere:

• "Was Rs 999, now Rs 349" e-commerce labels.
• Netflix subscription tiers where Premium makes Standard look reasonable.
• Salary negotiations where the first number quoted anchors the conversation.
• Real estate listings where an overpriced unit makes the adjacent unit look like a bargain.

EDLP vs High-Low Pricing

Price Elasticity of Demand

Plain language. How much does demand change when price changes.

Formula:

Categories by elasticity

Why elasticity varies

• Over time. Petrol demand is inelastic short-term (you still need to commute) but elastic over years (people switch to public transport or EVs).
• Across consumers. Cigarette demand is inelastic for heavy smokers, elastic for teenagers.
• By context. Cab rides are more inelastic during peak urgency. This is why surge pricing works.

Decisions elasticity helps you make

• If demand is inelastic. Raise price. You will lose few customers and earn more per unit.
• If demand is elastic. Lower price. The volume gain will more than offset the per-unit loss.
• Strategic discounting. Discount elastic products to drive volume. Do not discount inelastic ones, you are giving away margin.
• Build defensibility. Brand, prestige, unique features all make demand more inelastic.

New Product Pricing: Skim vs Penetrate

Skimming risks

• Slower market share build.
• High margins attract competitors.
• Early adopters who paid full price feel cheated when price drops.

Penetration risks

• Requires deep funding to survive early losses.
• Trains customers to expect low prices, making future price rises painful.

Pricing Must Align with Positioning

The single most important meta-principle. A premium brand cannot use penetration pricing without confusing customers and damaging the brand.

Data-Driven Decisions & Biases

A PM's job is to move from gut feel to evidence. The first discipline is recognising the biases that quietly corrupt judgement before any data is even collected.

Why decide with data

Plain language. Every PM has instincts, but instincts alone do not convince a team to spend money or change direction. Data turns "I think onboarding is confusing" into "activation dropped from 62% to 48%", a fact that forces action. The goal is never to remove uncertainty (no tool does that) but to get closer to the truth and reduce risk.

Formal. Data-driven decision making is the practice of grounding product choices in measurable evidence rather than opinion, while accepting that evidence narrows uncertainty rather than eliminating it. The faculty framed the whole module as building confidence in decisions, not certainty.

The biases that distort judgement

The faculty drilled four biases in the live session. Recognising them is the prerequisite for clean analysis, because a biased question produces biased data no matter how good the maths.

Decisions sit on top of context

Plain language. The quality of any analysis depends on how well you understand the business situation around it. Pick the wrong problem and the cleanest maths is still useless.

The faculty kept returning to the 5Cs (Company, Customers, Competitors, Collaborators, Context) as the lens that decides which questions are worth asking and which variables are worth measuring. Tools are easy to learn; the ability to frame the right problem is what separates a strong PM.

Types of Data

Before you can analyse anything, you need to know what kind of data you are holding. The type dictates which method is even valid.

Qualitative vs quantitative

Plain language. Quantitative data is numbers you can do maths on (orders, session time, revenue). Qualitative data is descriptive or categorical (genre, sentiment, an interview quote). You can count and average the first; you have to code or interpret the second.

This distinction matters immediately for regression (Topic 5). A qualitative variable like movie genre cannot be fed into a model as text, or as 1-2-3-4, because the model would invent a ranking that does not exist. It has to be converted with dummy coding.

Variable types

Cross-sectional vs longitudinal

Formal. Cross-sectional data is a snapshot of many subjects at one point in time (all customers surveyed this week). Longitudinal data follows the same subjects over time (one cohort's retention across D1, D7, D30). Cohort analysis in the metrics topics is longitudinal thinking.

Primary vs secondary, audit first

The faculty's practical rule: before collecting primary data (a fresh survey or interview), audit the secondary data you already hold, which is app analytics, reviews, support tickets, sales calls. Most companies already capture far more than they use. Collecting data you already have wastes the respondent's attention and lowers future response rates.

Analytics & Insight

Having data is not the point. Turning it into an insight that drives an action is. Most teams fail not for lack of dashboards but for stopping at the number.

Descriptive vs inferential

Plain language. Descriptive analytics tells you what happened (last month's average order value was ₹2,000). Inferential analytics uses a sample to draw conclusions about a larger population, with a stated confidence level.

Formal. Regression and A/B testing both live in the inferential category, and both use the p < 0.05 significance rule covered in Topics 5 and 8.

The chain that matters: data to insight to action

The faculty repeated this in nearly every session. The differentiator is whether you extract an insight from the data and convert it into an action.

Synergistic thinking is the faculty's term for combining several methods so they inform each other. No single technique is trusted alone; together they converge on the real problem.

Using AI responsibly for analysis

AI can crawl and monitor competitors, analyse review sentiment and tone, draft and launch surveys, and generate research reports (for example Displayr). The faculty's guardrails:

• Define the strategic question first. Do not use AI because it is trendy.
• Audit your data sources before feeding them in.
• Pilot the AI against a human analyst before scaling.
• Comply with data-protection law (GDPR, CCPA, India's DPDP Act 2023). Personal data needs consent; public data such as earnings calls and job postings generally does not.

Market Research

How to gather primary data well: pick the research mode that fits the question, choose the right survey method, and design questions that do not poison the answers.

Three research modes

The faculty's ideal sequence: explore qualitatively to understand the "why", then validate and quantify with a survey, then prove causation with an experiment. This is the same hierarchy that reappears in regression (a starting point, not proof) and A/B testing (the causal step).

Survey modes, the cost-quality-reach trade-off

Face-to-face still wins when the population lacks technology or language access (rural consumers), the topic is sensitive (health), or the respondent is already on the premises (in-store).

Questionnaire design rules

• Pilot test with 10 to 15 people before launch to catch errors.
• No double-barrelled questions. "Is it useful and inexpensive?" mixes two things; split it.
• No leading questions. "Is the chapter too lengthy?" primes the answer. Ask "Rate the length 1 to 5" instead.
• Order general, then specific, then personal. Never open with age or income; it tanks response rates. Keep personal questions optional and last.
• Keep around 80% closed-ended for response rate; use the remaining open-ended ~20% to capture the "why".
• Repeat a question in different wording as an attention and bot check.
• Time-filter out responses finished far below the median completion time; use CAPTCHA against bots.
• Explain the purpose upfront to motivate honest answers.

Tools vs panels

Survey tools build the questionnaire: Qualtrics (enterprise, considered strongest), SurveyMonkey, Google Forms, Microsoft Forms, SurveySparrow (chat-style, AI-assisted). Panel providers supply the respondents: Toluna, Nielsen Consumer Panel, Amazon Mechanical Turk, Prolific. Even paid panels carry quality risk, since people who take surveys for income rush and straight-line, so the wording and time checks above still apply.

Regression Analysis

The core quantitative tool: measure how one or more predictors relate to an outcome, predict the outcome, and read whether the relationship is real or noise.

What it is

Plain language. Regression draws the best-fit line through your data so you can say how a predictor moves the outcome, and predict the outcome for new inputs. Think of it as a smarter average: instead of treating every past period equally, it tilts a line to follow how the variables move together.

Formal. Regression estimates the relationship between a dependent variable Y (the outcome, "what keeps you up at night": NPS, churn, CLV, revenue, DAU) and one or more independent variables X (the predictors). One predictor is simple linear regression; two or more is multiple linear regression.

A residual is the gap between an observed value and the model's prediction for one data point. The mean alone is the baseline predictor when you have no other information; a good predictor shrinks the residuals and tilts the flat average line into a fitted line.

Reading the output, three numbers

Worked case A, trial duration vs conversion

Real regressions run on the faculty's 24-row SaaS trial dataset. A linear fit gives a negative slope (longer trials, lower conversion) but explains only part of the story:

Students suspected the real shape was an inverted-U (a sweet spot, not a straight decline). Adding a squared term confirms it:

Worked case B, DVD sales and dummy coding

Predicting first-three-weeks DVD sales across 250 titles. Pre-orders alone are a remarkably strong predictor:

Genre is qualitative, so it cannot be coded 1-2-3-4 (that invents a ranking). It is dummy coded: one 0/1 column per genre, and with four genres you include only N−1 = 3 dummies. The omitted genre (here Family) becomes the base every other genre is compared against. Including all N triggers the dummy variable trap (perfect redundancy that breaks the model).

Adjusted R² stays at about 0.94: genre and the release gap add nothing meaningful beyond pre-orders and box office. The lesson the faculty wanted is that more variables is not better; logic and significance decide what stays.

Correlation is not causation

Regression measures correlation, not cause. The birds-and-stock-market story: a regression can return a "significant" link between birds on a balcony and market returns, but acting on it bankrupts you because there is no logical mechanism. A result is only trustworthy when a prior logic or theory explains it.

Product Metrics

Which numbers actually matter, mapped to the user journey, anchored by one North Star, and how to avoid chasing a metric until it stops meaning anything.

Goodhart's Law

Formal. When a measure becomes a target, it ceases to be a good measure. Turn a metric into a goal and people optimise the number, not the value behind it.

The three cautionary tales: the nail factory (paid per nail gives tiny useless nails; paid per weight gives huge useless nails; the factory closes), the cobra bounty (people bred cobras for the reward), and the 70% test-coverage mandate (engineers wrote tests that tested nothing). Guardrail: pair any target metric with a complementary one to catch gaming.

Vanity vs actionable metrics

Plain language. A vanity metric looks impressive but does not tell you whether the product creates value (downloads, raw views, registered users, likes). An actionable metric ties to behaviour or revenue. The test: if this number changes, do I know what to do?

Metrics across the user journey

NPS

North Star and leading vs lagging

The North Star metric is the one metric that reflects core value and predicts long-term growth; every other metric should ladder up to it. Examples: Netflix hours watched ("the competitor is sleep"), Zomato orders per active user, Airbnb bookings per traveller, WhatsApp messages per DAU, Spotify listening hours.

Leading indicators predict the future (activation, WAU, engagement); lagging indicators report the past (revenue, churn, CLV). The balanced scorecard (1992) says watch three perspectives, financial, customer and operations, or you go myopic. Governance metrics (GDPR, DPDP) round it out.

Digital & Media Metrics

The e-commerce and advertising formula stack, plus the diagnostic tools: funnels, segmentation, and the analytics platforms that surface them.

Early warning and funnel analysis

Most churn is predictable before revenue drops. Weekly warning signals: DAU down about 10% week-over-week, activation falling for new cohorts, session duration declining, feature adoption slowing after release. Act on these leading signals rather than waiting for the lagging revenue line.

The funnel tells you where users drop; you still diagnose why (price, trust, quality, UX) with reviews, surveys, regression and A/B tests. Segmentation breaks an aggregate apart: overall activation of 55% can hide Tier-2 cities at 42% behind Tier-1 at 68%. Cut by geography, behaviour (seasonal vs regular buyers) or demographics, then form a hypothesis before acting.

E-commerce formula stack

Acquisition cost and lifetime value

Retention and advertising metrics

Tools and the combination principle

GA4 tracks the full lifecycle (acquisition, engagement, monetisation, retention); an engaged session is over 10 seconds, or a conversion, or 2 or more pageviews. Mixpanel is the common Indian choice for funnels and cohorts; Amplitude suits advanced cohort analysis at scale. The faculty's big rule: never read a metric in isolation. Pair high CAC with LTV-CAC, sessions with RPS, CPM with engagement, registered users with MAU.

A/B Testing

The causal tool. Randomly split users between two versions, change one thing, and use statistics to decide whether the difference is real, not the boss's opinion.

What it is and why randomization matters

Formal. An A/B test is a randomized controlled experiment comparing two versions of one element on a single outcome. Randomization means every user has an equal chance of either version, which spreads confounders (time of day, device, mood, campaigns) evenly across both groups, so any difference is caused by the change. This is what makes it the gold standard for causation, unlike regression's correlation.

It overrules the HiPPO effect: the data decides, not the highest-paid opinion. The treatment-and-control logic is the "you cannot both send and not send the same person" counterfactual, so you compare a randomly-assigned control instead.

The 8-step process

• 1. Gather insights from analytics, support tickets and interviews (for example Flipkart saw cart abandonment).
• 2. Set goal and metric, one primary metric (conversion, CTR, sign-up, retention).
• 3. Form a hypothesis with logic, in the form "If [change], then [outcome], because [reason]."
• 4. Build variations, one variable at a time, changing only the button colour and nothing else.
• 5. Run the test for 1 to 4 weeks; aim for about 1,000+ sessions per variant. Do not peek and stop early (p-hacking).
• 6. Analyse and calculate lift, the difference in conversion rates.
• 7. Check statistical significance with the Z-test and p-value.
• 8. Decide and monitor long-term, rolling out the winner and watching for the novelty effect fading at 1 week, 1 month, 3 months.

Worked example, Nike "Add to Bag" button

Red vs black button. Do not stop at the conversion rate: compute the lift, then test significance.

A/B/n testing

Compare 3 or more versions at once instead of sequential A/B rounds. Example: a bank loan-repayment email with 7 behavioural-science variations (scarcity, social proof, loss aversion, authority, reciprocity, default, timing) plus a no-email control, across 2M customers. Social proof plus scarcity won, and some variants lifted deposits by 30% or more. Key statistics point: with 3 or more groups you use ANOVA or regression, not the two-proportion Z-test.

Pitfalls

• Stopping too early or running too short (or endless); small samples are noise.
• Testing multiple variables at once, so you cannot tell which caused the result.
• No predefined success metric, or ignoring significance.
• Only looking at averages; segment, because a winner overall can hurt mobile or new users.
• Cross-contamination, where test groups interact or share.
• Over-optimising for heavy users and short-term, novelty-driven thinking.

Tools: Optimizely, Adobe Target, AB Tasty, Kameleoon, Convertize. Netflix built its own (Shakespeare) to let non-engineers run tests; democratising experimentation turned "Start Watching" into "Get Started" for a global sign-up lift.

Course Wrap-Up (context)

The course closeout covered two cross-cutting topics that sit outside the Module 6 metrics syllabus. They are kept here for completeness and separated from the quiz material in Topics 1 to 8.

The four PM documents

PM writing inverts academic writing: brevity signals confidence, caveats read as "cover your ass", and metrics must be specific ("load under 200ms", not "fast"). Roughly 60% of a PM's time is documentation, which is how a PM leads without authority.

Five startup funding stages

Frameworks & Methodology

Every framework, method, model and formula from all six modules, grouped by area and tagged with its module and topic. Use this as a single revision checklist.

About this Notebook

A short note on what this is, how it is built, and how to get the most out of it before the exam.

What this is

A personal study notebook for the BITSoM Product Management with Generative and Agentic AI program. Each faculty session is condensed from its transcript, pre-read and lecture notes into clean, structured notes, then cross-checked for accuracy. All six modules are complete here, spanning Product Thinking, Design and MVP, Agentic AI, Agile Execution, Go-To-Market, and Data and Metrics.

How it is organised

• Modules. Each module is a collapsible group in the sidebar, holding its numbered topics.
• Topics. Every topic gives a plain-language read first, then the formal definition, real company examples, links back to earlier modules and to PM or design work, and a dark "Remember for the Quiz" recap at the end.
• Frameworks and Methodology. A single cross-module index of every framework, model and formula, tagged by module and topic and filterable with the chips at the top.

How to use it

• Learn. Read a topic top to bottom. The plain-language line is the one to internalise first.
• Self-test. Cover the recap and try to reproduce each point before reading it.
• Revise. The Frameworks and Methodology tab is the last-mile checklist. Filter to one module the night before its quiz.

Closing

Metrics, models and tests are only useful when they sharpen a decision. Keep asking what action a number would change, and the frameworks will do their job. Good luck with the exam.