AI Product Management
What a product manager does
Begin with people, problems, outcomes, and choices; not features or technology.
- Read slowly, say unfamiliar ideas aloud, and try the ten-minute practice before looking at the model reasoning. Use it to check your work, not as a script to copy.
- Each chapter assessment samples 6 questions from a bank of 8. The pass mark is at least 80%, so you need 5 of 6 answers correct.
- Single-select questions have one correct answer. Multi-select questions require the complete correct selection. Explanations point back to the reading.
- The assessment is one knowledge check. It does not assess the quality of your practical portfolio. Complete the on-the-job lab, ask a reviewer to challenge it or follow the stated solo route, and save the revision before moving on.
- Try the understanding check first, without looking back. You do not need perfect recall. Note what you can explain, what your work demonstrates, and what needs another pass.
- Start the evidence log in Chapter 1. After each chapter, add your work and one weakness you found. Note the revision you made and the skill you will practise next.
- Build the required project in the course project workspace. Save one checkpoint after each chapter. Course completion requires every chapter assessment pass and instructor approval of your final project. Revise and send a new version if changes are requested.
By the end of this chapter
- Tell the difference between a product, a project, and a feature.
- Describe the PM’s job as a set of decisions rather than a list of meetings.
- Write a clear problem statement without smuggling in a solution.
- Explain how user value, business value, feasibility, and responsibility constrain one another.
Read the lesson
Choose your project and keep a Project Passport
NeighbourCare is our fictional community service. Start with its transport-information journey unless you already have a suitable low-risk project. ResolveAI is the proposed staff-facing AI draft assistant introduced in later chapters. It is one possible response for NeighbourCare. Repairing the service, using simpler software, and building nothing remain options.
Your Project Passport is a short note that keeps the same project clear across the course. Create it in a notebook or document, then update it when evidence changes your direction. Do not collect confidential workplace or personal information for practice. The class pack’s Product Picker sheet and AI-Product-Project-Catalogue.md offer alternative contexts; NeighbourCare remains the default.
| Passport field | What to write |
|---|---|
| Project and user | NeighbourCare by default; name the person and one task they need to complete. |
| Problem and current journey | What happens today, where progress fails, and the consequence. |
| Evidence and unknowns | Separate observations, supplied or synthetic examples, assumptions, and open questions. |
| Outcome and boundaries | One result to improve, guardrails, and actions or data outside the project. |
| Decision and owner | The next decision, who would own it in a real service, and what could change it. |
| By the end of this chapter, you can… | Proof you can show |
|---|---|
| Tell the difference between a product, a project, and a feature. | Explain it aloud without jargon |
| Describe the PM’s job as a set of decisions rather than a list of meetings. | Complete the guided worked example |
| Write a clear problem statement without smuggling in a solution. | Produce one reviewable workplace artefact |
| Explain how user value, business value, feasibility, and responsibility constrain one another. | Respond to the chapter’s challenge question |
Before Chapter 1: examine one PM decision at a time
'The PM owns the product' can sound as though the PM is everyone’s boss. A product manager helps the group decide which problem to address, what result matters, what to try next, and what evidence would change the plan. The PM usually does this without managing the designer, engineer, researcher, or analyst.
Imagine a family planning a large meal. One person keeps these details visible:
- Guest list.
- Allergies.
- Budget.
- Timing.
- Menu choices.
- Responsibilities.
That person does not have to cook every dish. They help make sure the meal meets the need and the separate parts work together. Product management is similar, but decisions continue after launch because the product keeps serving people.
Key terms
| Word | Meaning in this course |
|---|---|
| Problem | A difficulty or unmet need experienced in a particular situation. It is not the proposed feature. |
| User | The person who operates or directly experiences the product. |
| Buyer | The person or organisation that chooses or pays. The buyer and user can be different. |
| Outcome | A meaningful change, such as completing a task faster or making fewer dangerous mistakes. |
| Word | Meaning in this course |
|---|---|
| Output | Something the team produces, such as a screen, report, API, model, or feature. |
| Evidence | Something observable that supports or weakens a claim: behaviour, interview detail, data, or a test result. |
| Assumption | A belief the team is currently treating as true but has not yet proved. |
| Trade-off | A choice where gaining one benefit may cost time, money, simplicity, speed, safety, or another benefit. |
Walk through Maya's first ten minutes
- The director says, 'We need an AI chatbot.' Maya writes the sentence down without agreeing or disagreeing. She treats it as a request that must be understood.
- She asks, 'Who is struggling, during which task?' This changes the conversation from technology to a person and a moment.
- She asks, 'What happens today?' The answer reveals three conflicting pages, repeated phone calls, and abandoned journeys. That is the current baseline.
- She asks, 'What harm or cost follows?' Residents miss transport; staff repeat explanations; the organisation receives avoidable calls. These are consequences, not vague frustration.
- She asks, 'What evidence do we have?' Eight call recordings and support logs exist, but nobody has watched residents use the pages. She labels the recordings evidence and the unobserved journey an evidence gap.
- She asks, 'What result would be worth achieving?' The team suggests more completed journeys with no increase in incorrect eligibility guidance. That is closer to an outcome plus a guardrail.
- Only now does she list possible responses. She could repair the pages, simplify the journey, or improve staff tools. She could add search, test a bounded assistant, or do nothing. AI is one option, not the starting answer.
Example team conversation
| Person | What they might say |
|---|---|
| Director | Can you commit to launching the chatbot this quarter? |
| Maya, PM | I can commit to returning with a tested problem definition and options. I cannot responsibly commit to the solution until we know which journey is failing and whether an assistant improves it. |
| Designer | I can observe the journey and test where people lose confidence. |
| Engineer | I can identify which information sources are current and whether the rules are accessible. |
| Maya, PM | Good. Our next decision is whether the information problem is clear enough to choose a response. |
First ten-minute practice
Employees with non-standard contracts cannot confidently determine their remaining leave or the approval step when policies and payroll records disagree, which causes repeated HR contacts and delayed requests. We have 46 tagged support tickets from the past month, but still need to learn which contract types and journey steps produce the disagreement. Notice that this answer does not promise a chatbot.
Common mistakes
- Calling a feature an outcome: 'launch chatbot' describes what the team ships, not what improves for a person.
- Using 'users are frustrated' as evidence without describing what people did, could not do, or said in context.
- Assuming the loudest stakeholder represents all users or affected people.
- Trying to write a perfect solution before the team agrees on the decision and the evidence gap.
Questions to answer as you read
- I can explain the PM's job without saying 'CEO of the product.'
- I can separate a problem, output, and outcome using an everyday example.
- I can mark a sentence as evidence, assumption, or opinion.
- I can name at least three possible responses before choosing one.
Products, projects, and features
A project has a planned beginning and end: moving an office, organising an event, or migrating a database. A product keeps helping a group of people make progress. It needs ongoing:
- Discovery.
- Delivery.
- Support.
- Measurement.
- Improvement.
A feature is one capability within the product. A chatbot describes a possible feature, not a product decision.
The PM keeps the team focused on a result users can achieve repeatedly and an outcome the organisation can sustain. Choose technology after agreeing on the behaviour and constraints.
How the PM works with specialists
The PM brings together the work of specialists:
- Designers.
- Engineers.
- Researchers.
- Salespeople.
- Lawyers.
- Data analysts.
The PM does not need to be the best at each discipline. The job is to make their evidence and the decision clear. In many organisations the PM recommends rather than commands. Influence comes from a clear problem, credible evidence, visible trade-offs, and follow-through.
| Question | The specialist contributes | The PM keeps visible |
|---|---|---|
| Do people need this? | Research and design evidence | Target user, unmet need, alternatives, severity |
| Can we build and operate it? | Engineering and data judgement | Scope, dependencies, failure modes, cost, reversibility |
| Should the organisation fund it? | Leadership, finance, sales and strategy | Outcome, opportunity cost, business model, non-goals |
| Can we do it responsibly? | Security, privacy, legal, accessibility and domain experts | Affected people, controls, proof, residual risk, owner |
Write a problem statement
A useful problem statement names:
- A specific person.
- Their situation.
- Their struggle.
- The consequence.
- The evidence.
“We need a chatbot” requests a solution. “Residents abandon the transport-assistance journey because eligibility and booking instructions conflict across channels” is a problem hypothesis. The team still has to verify it.
Worked example
The director wants an AI assistant for NeighbourCare.
| A rushed answer | A stronger beginner answer |
|---|---|
| Build a chatbot so residents can get answers faster. | Residents arranging non-emergency medical transport cannot reliably determine eligibility and the next booking step. In eight support calls, five people repeated information and three received conflicting guidance. We need to reduce unresolved journeys without giving unauthorised eligibility decisions. |
The example identifies:
- The person and moment.
- Evidence.
- Consequence.
- Desired improvement.
- An important boundary.
Several responses remain possible:
- Repair the content.
- Train staff.
- Simplify a form.
- Add search.
- Add a bounded assistant.
Use four checks: value, viability, feasibility, and responsibility
- Value: Does the person make meaningful progress? A click, session, or generated answer is not automatically value.
- Viability: Can the organisation sustain the service? Consider revenue and funding, cost and risk reduction, or mission impact.
- Feasibility: Can the team build, integrate, support, and change it? Consider the available skills and systems, time, and money.
- Responsibility: Can people use it safely and fairly? Check privacy, accessibility, and explanation. Include correction and human recourse.
On-the-job lab: create a one-page opportunity brief
Use the product selected in your Project Passport. NeighbourCare remains available as the fully worked fallback. Treat the catalogue choice as a problem context, not permission to assume AI is the answer.
- Name one target user and the exact moment they are trying to make progress.
- Write what they do today, including workarounds and other people involved.
- Write the pain or risk in observable terms. Avoid ‘frustrating’ unless you can say what happened.
- Add the consequence for the user and the organisation.
- List three pieces of evidence you have and three things you are only assuming.
- List at least three possible responses, including a non-AI option and the option to do nothing.
- Write one outcome measure and two guardrails that must not worsen.
- Ask a reviewer to circle every claim that lacks a source. Revise the brief.
Work to share for review
- A one-page brief with user, situation, current journey, evidence, assumptions, outcome, guardrails, alternatives, non-goals, owner, and next decision.
- A dated reviewer note showing one challenged assumption.
- A second version that visibly changes after the challenge.
Check your understanding
Five-minute vocabulary review
| Word | Meaning you can use at work |
|---|---|
| Product | A continuing system for delivering a repeated outcome to a group of people. |
| User | The person operating or experiencing the product; not always the buyer. |
| Stakeholder | A person or group that affects the decision or is affected by it. |
| Outcome | A meaningful change in behaviour, ability, risk, cost, revenue, or mission progress. |
| Constraint | A real limit such as law, time, cost, system capability, accessibility, or authority. |
- Explain why “build a chatbot” is not a problem statement.
- Give an everyday example of a product, project, and feature.
- Name one decision a PM makes and three specialists who improve it.
- Explain why a useful product could still be unsustainable or irresponsible.
Evidence and revision log
| Prompt | Your note |
|---|---|
| What did I make this chapter? | Name the artefact and where it is saved. |
| What did a reviewer challenge? | Record the question, objection, or missing evidence. |
| What did I change after feedback? | Describe the revision and why it is stronger. |
| What will I practise again? | Choose one skill, not an entire topic. |
Sources and further reading
- AI Product Management Ultimate Edition: Teslim Salawu, August 2026
Original 303-page student-class-pack edition. This lesson preserves guided week 1, pages 15 to 21. The full book includes the mastery manual, studios, technical build labs, apprenticeship, templates, and references.
- GOV.UK Service Manual: Learning about users and their needs
Official guidance on starting with user needs and evidence rather than a preferred solution.
- Government Digital and Data: Product manager role
One public-sector role framework for comparing product skills and responsibility levels; requirements vary by employer.
Adapted reading from Teslim Salawu’s AI Product Management Ultimate Edition, original guided Week 1, pages 15 to 21. Original class-book illustrations are retained.
Required project checkpoint
Write the opportunity brief
Define the transport-information problem and keep several responses open.
Start with
Read the Amina story and the worked opportunity brief in Chapter 1. Open the student class pack from /courses/ai-product-management#course-resources. Keep the Project Passport in your project notes; use NeighbourCare as the common case.
Name the moment
Write who needs help and what they are trying to do. Start with Amina arranging non-emergency medical transport. Describe finding the current rule and next booking step before naming a chatbot.
Expected result. A short problem statement about a person and a failed journey, with no solution hidden inside it.
Separate facts from guesses
Copy the lesson’s fictional eight-call summary into a supplied-evidence section: five people repeated information and three received conflicting guidance. List what those calls do not establish, such as prevalence across all residents. Do not turn the counts into original research.
Expected result. An evidence list with a source label and an assumptions list that makes missing evidence visible.
Compare responses
Write at least three responses. Include repair of conflicting instructions and a non-AI route. Keep doing nothing visible long enough to explain its consequences. For each, state what it might improve and what the team would need to learn.
Expected result. An options comparison that can support a decision other than building AI.
Define the first result
Choose an outcome about completing the journey. Add two guardrails using the chapter’s concerns, such as incorrect eligibility guidance and accessibility. Leave unknown targets marked to be established rather than inventing improvements.
Expected result. One outcome and two guardrails, each linked to a user consequence.
Challenge and save
Read the brief as a resident, support worker, and finance owner in separate passes. Circle unsupported claims. Revise one and keep both the first and revised wording in the same record.
Expected result. The first version of your opportunity brief, with a visible revision and a question for Chapter 3 research.
Save these outputs
- Opportunity brief with the problem, supplied evidence, assumptions, options, outcome, guardrails, and revision.
If you get stuck
- My brief starts with “build a chatbot.”
- Replace the solution with the resident’s task and observable breakdown. Move the chatbot to the options list.
- I have no real interviews.
- Use the supplied fictional case and label it. Record the evidence you would still need; do not invent participants or quotes.
Your checkpoint template
Project: NeighbourCare Record: Opportunity brief Person and situation: Problem and consequence: Supplied evidence and source: Assumptions and unanswered questions: Option 1 and trade-off: Option 2 and trade-off: Option 3 and trade-off: Outcome to improve: Guardrail 1: Guardrail 2: First recommendation and next evidence: Claim revised after self-review:
In the student classroom, save this checkpoint in the project workspace. After the final chapter, send the complete project for instructor review. This owner review shows the instructions; it does not save or send learner work.