The CPO Playbook
Process
17 août 2026 · Dernière mise à jour
Contenu à réviser
FR / EN
Login

AI Tools for Product Managers.

A map of the category, a stack that is actually implementable, what is publicly known about how large tech companies work, and a benchmark by company size. Followed by the part that matters most: what AI does not take off the PM's plate.

Two caveats before reading. First, this category moves fast — tool capabilities and pricing change quarterly, so treat every product name as a starting point for your own evaluation, not a recommendation. Second, section 03 describes tooling at Airbnb, Stripe and Meta: these companies build heavily on internal tools that are not publicly documented. What follows is the publicly reported pattern and the product culture it reflects — read it as an illustration of three approaches, not as a verified inventory.

01The category map

— six families of tools, by the job they do

General-purpose assistants — daily use

Claude · ChatGPT · Gemini Drafting PRDs and specs, synthesising user interviews, generating user stories, roadmap brainstorming, stakeholder updates. Limitation: not natively connected to your Jira, analytics or feedback data — you paste context in, which caps how much they can reason about your actual product.

AI inside the product stack

Notion AI Document generation, automatic note summarisation, spec structuring — sits where the knowledge base already lives.
Jira Product Discovery Assisted prioritisation, feedback synthesis, opportunity suggestions. The discovery layer above the delivery backlog.
Productboard Automatic clustering of customer feedback, pattern detection across sources, roadmap support.

Discovery & user research

Dovetail Interview analysis, automatic insight extraction, intelligent tagging — turns a pile of transcripts into a searchable insight repository.
Sprig In-product feedback analysis, automatic summaries, UX friction detection.

Analytics & product intelligence

Amplitude Automated behavioural analysis, anomaly detection, predictive insights. Also the reference tool for the KPIs on the PLG page.
Mixpanel Automatic insights, funnel and retention analysis, suggested actions.

Delivery & product ops

Linear Ticket summarisation, assisted prioritisation, automatic bug triage.
Height Near-autonomous backlog management, workflow automation, agent-style task execution.

Emerging — AI-native

LogRocket Galileo AI Session analysis with automatic detection of UX problems — the closest thing to an automated friction-finder.
Athenian Engineering-team performance analysis and delivery insights. Note this measures the team, not the product — useful for an org, sensitive to deploy.

02A stack that is actually implementable

— six layers, one tool each, no overlap

LayerToolRole
BrainClaude / ChatGPTDrafting, structuring, thinking through problems
KnowledgeNotion AIPRDs, specs, the team knowledge base
DiscoveryDovetailInterviews, user insights, synthesis
FeedbackProductboardCentralisation and prioritisation
AnalyticsAmplitudeUsage, funnels, retention
DeliveryLinearTickets, backlog, execution

The end-to-end workflow

CollectInterviews land in Dovetail with automatic tagging and synthesis; customer feedback flows into Productboard from every channel.
AnalyseAmplitude surfaces churn and drop-off points; the assistant cross-reads qualitative insights against the behavioural data to find where they agree — and, more usefully, where they contradict each other.
DecideProductboard scores and prioritises; the resulting PRD is drafted and structured in Notion.
ExecuteLinear tickets are generated from the PRD; bug and ticket summaries are produced automatically.
CommunicateStakeholder updates and roadmap narratives drafted from the same source material, so the story stays consistent across audiences.

On the time savings

Reported gains cluster around 40–60% less time spent on drafting, synthesis and documentation. Treat that as directional: it comes from vendor and practitioner reports rather than controlled study, and it measures the artefact-production part of the job — which was never the part that determined whether the product succeeded.

03Three approaches at scale

— Airbnb, Stripe, Meta: what the tooling says about the product culture

Airbnb

Pattern
Internal data tooling, an in-house experimentation platform, and a strong research culture that is deliberately not tool-centric.
AI is used for
Search, pricing and personalisation — inside the product more than inside the PM workflow.
PM style
Design and product intuition. Few plug-and-play tools; judgement and taste carry more weight than tooling.

Stripe

Pattern
SQL and internal data tools, documentation-heavy culture, delivery tooling close to what Linear offers.
AI is used for
Developer support, documentation and API intelligence.
PM style
Execution and developer experience. PMs sit unusually close to engineering; the written document is the primary unit of thinking.

Meta

Pattern
Massive internal data infrastructure and A/B testing at a scale few companies can replicate.
AI is used for
Recommendation, feed ranking and ads — AI is the product, not a tool beside it.
PM style
Metrics and optimisation. Decisions are driven by experiment results at a volume that makes intuition secondary.

The transferable lesson

None of these stacks is copyable — they rest on internal platforms built over a decade. What is transferable is the alignment: each company's tooling matches how it actually decides. Airbnb invests in research because it decides on judgement; Meta invests in experimentation because it decides on data. Pick tools that match your decision-making culture, not the culture of a company you admire.

04Benchmark by company size

— the constraint changes, so the stack changes

Startup — 0 to 50 people

Stack
Claude or ChatGPT · Notion AI · Linear · Amplitude or Mixpanel
Goal
Move fast, keep tooling minimal, maximise individual output.
The constraint
Individual capacity. One PM operates as a small augmented product team; every tool added must pay for itself immediately.

Scale-up — 50 to 500 people

Stack
Claude / ChatGPT · Notion AI · Productboard · Dovetail · Amplitude · Jira Product Discovery
Goal
Structure discovery, align several teams, prioritise defensibly.
The constraint
Coordination and clarity. The bottleneck stops being individual output and becomes agreement between teams on what matters.

Enterprise — 500+ people

Stack
Internal tools with embedded AI · Amplitude / Snowflake / Looker · in-house experimentation platforms
Goal
Scalability, governance, a single unified source of data.
The constraint
Organisational complexity. Tool choice becomes a governance and procurement question as much as a product one.

05What actually changes — and what does not

— the strategic reading

Three trends are visible across the whole category:

  • Automation of low-value PM work — notes, tickets, summaries, status updates. This is where the time savings come from.
  • Augmented discovery — extracting insight from raw qualitative and behavioural data at a volume no PM could read manually.
  • The arrival of PM "copilots", and early attempts at semi-autonomous product tooling that executes tasks rather than only suggesting them.

No tool currently replaces three things, and they are the three that determine whether a product succeeds:

  • Strategic prioritisation — choosing which problem is worth solving now, against opportunity cost.
  • Business understanding — knowing what a number means for the company, not just whether it moved.
  • Organisational alignment — getting people who disagree to commit to one direction.

Where the bottleneck moved

The PM becomes less of a writer and more of a decider. The constraint used to be producing the artefacts — specs, decks, summaries. Now that those are cheap, the constraint is choosing the right problems. A tool that helps you produce twice as many specs for the wrong roadmap has made things worse, not better.

The practical distinction worth holding onto: the best PMs orchestrate AI tools rather than simply using them. Using a tool means faster artefacts. Orchestrating means designing the chain — where insight enters, where it is challenged, where a human decides — so the speed compounds into better decisions instead of just more output.

← Back to the Playbook