AI skills for
product managers

For PMs, AI skills come in two layers. The first is using AI in your own work: synthesising discovery, analysing data, drafting specs and communication, and prioritising faster. The second is AI product judgement: deciding where AI belongs in your product, how to evaluate it, which model fits and what it does to your economics.

Both are now expected. McKinsey's 2025 survey found 88% of organisations using AI in at least one function, but only about a third scaling their programmes.² Gartner expects more than 40% of agentic AI projects to be cancelled by 2027 due to costs, unclear value or weak risk controls.¹ PMs who can separate valuable AI from hype are worth more than PMs who add AI everywhere. And research shows AI's help is uneven: a Harvard Business School study found strong gains on tasks inside AI's capabilities and worse accuracy outside them.³ PMs need that instinct for their own work too.

Organisations using AI in at least one function (McKinsey, 2025)

88%

Agentic AI projects Gartner expects cancelled by 2027

40%+

Fall in GPT-3.5-level model cost, Nov 2022–Oct 2024 (AI Index 2025)

280×

Typical product manager pay in India (AmbitionBox, Oct 2026)

₹22.1–24.5 lakh

Table: Where AI skills come up in PM interviews (editorial summary)

## Should we add an AI ordering assistant?

User problem: repeat customers spend 40 s rebuilding usual orders.
Alternatives: "reorder" button (cheap, deterministic), AI assistant (flexible).
Prototype eval (200 real sessions):
- Repeat orders: AI correct 94%; reorder button covers 70% of cases
- New discovery: AI correct 61% — below our 85% bar
Cost: large model ₹1.40/order (too high); small model ₹0.20/order, quality -2 pts
Decision: ship a reorder button now; AI assistant for repeat orders only,
small model, with confirm-before-checkout. Revisit discovery in Q3.

It compares AI with a simpler alternative, uses real data, sets a quality bar, factors in cost and scopes the feature to where it works.

Table: How the PM job has changed with AI (editorial summary)

• Two examples of AI-accelerated PM work, with how you verified it
• A synthesis traced back to real customer evidence
• One AI feature decision with an eval, cost and scope
• A view on model choice for a feature
• An example of choosing not to use AI
• A verified AI for Product Managers test result

Show leverage and judgement: "Used AI to synthesise 3,000 tickets and 40 interviews in 1 day (themes validated against raw data); scoped AI ordering assistant to repeat orders after 200-session eval; feature lifted repeat-order speed 35% at ₹0.20/order."

"How do you use AI as a PM?"

Synthesis, analysis, drafting and prototyping — with verification against real data and your own judgement on decisions.

"Should we add AI to this?"

Start with the user problem, compare with simpler options, prototype, evaluate quality and cost, and scope to where it works.

"How would you measure an AI feature?"

An offline eval set, online metrics such as task success, corrections and escalations, plus cost and latency.

"Which model would you use?"

The cheapest that meets the quality bar on your eval set, with room to switch as prices and models change.

"AI summarised our interviews. How do you know it's right?"

Spot-check themes against transcripts, look for missing minority views, and keep quotes linked to sources.

Table: Product roles and typical pay in India (AmbitionBox, October 2026; indicative)⁶

• Associate PM — AI-accelerated discovery and specs.
• Product manager — AI judgement in their product area.
• Senior / group PM — AI strategy across products.
• AI product manager — specialist ownership of AI products.
• Head of product — AI-era team and portfolio design.

1. Week 1: Your workflow — Use AI for synthesis, analysis and drafting; verify everything.
2. Week 2: AI judgement — List AI ideas for your product; compare with non-AI options.
3. Week 3: Evaluate one — Prototype, build a small eval set, estimate costs.
4. Week 4: Prove it — Write the decision and take the AI for Product Managers Certification.

Next: the AI Product Manager Certification. Guides: AI product management, product manager interview questions and will AI replace product managers?.

• Feedback triage — AI clusters, PM validates.
• Specs — first drafts in minutes, rewritten for clarity.
• Data questions — AI-generated queries checked before use.
• Prototypes — quick AI-built demos for early testing.
• Stakeholder updates — drafted fast, numbers checked.
• AI feature reviews — eval results and cost.

"What's a mistake you've seen with AI features?"

Shipping without an eval set or fallback — and discovering confident errors from customers.

"How do falling model costs change your roadmap?"

Features that were too expensive may become viable; revisit cost assumptions each quarter.

• "AI wrote my PRD" with no editing story.
• Unverified synthesis.
• AI everywhere without alternatives.
• Ignoring cost.
• No measurement plan.

Table: A first-90-days plan (editorial guidance)

New PMs are usually expected to understand the product, users and data quickly. AI can compress that ramp-up dramatically if you use it to summarise and cluster existing material — and then check the clusters against raw evidence. Use the time saved to talk to more customers. By the second month, pick one AI idea that leadership is excited about and run it through a proper evaluation: real cases, a quality bar, cost per use and a simpler alternative. Whatever the result, write it up clearly. A well-reasoned 'not yet' earns as much credibility as a launch, and PMs who make evidence-based AI decisions early get handed the most important AI work later.

Adoption and scaling figures come from McKinsey's 2025 State of AI; cancellation forecasts from Gartner's June 2025 press release; uneven AI performance from the Harvard Business School/BCG study; cost trends from Stanford HAI's 2025 AI Index; work-design trends from Microsoft's 2025 Work Trend Index; and India pay from AmbitionBox (read October 2026). The decision example, formats and history are editorial. Science of Apps publishes the free certification linked here. Last reviewed October 2026.

• Discovery synthesis — turning research and feedback into themes and insights.
• PRD — product requirements document.
• Evaluation set — real cases used to measure an AI feature's quality.
• Confident error — a wrong AI answer presented as certain.
• Unit economics — revenue and cost per user or transaction.