Strategy & Planning

How to build a product roadmap with AI

Strategy & Planning3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

A product roadmap is the connective tissue between company strategy and what the team actually ships. It answers three questions: what are we building, in what order, and why. Most operators draft one at the start of a planning cycle, then watch it go stale the moment priorities shift — which is usually within two weeks. Keeping it current, stakeholder-ready, and grounded in real signals is a persistent maintenance job, not a one-time document.

AI feels like a natural fit here because the hardest parts of roadmapping are synthesis and structuring — exactly what LLMs are good at. You have scattered inputs: customer feedback from support tickets, sales call notes, engineering estimates, strategic bets from the last board meeting. Turning that pile into a coherent, prioritized plan with clear rationale is time-consuming cognitive work that a well-prompted model can accelerate significantly.

ChatGPT, Claude, and Gemini can genuinely help you draft a roadmap structure, pressure-test your prioritization logic, write the rationale behind each initiative, and translate a rough outline into a polished stakeholder document. They work well as thinking partners for frameworks like RICE scoring or opportunity-solution trees. The limitation isn't the quality of the output — it's that none of these tools have access to your actual data, and nothing they produce persists or updates automatically.

Skip the manual work → See the Starch version

Strategy & Planning3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ClaudeChatGPTGemini

Step-by-step

1 Gather your raw inputs before opening any AI tool — paste in recent customer support themes, sales objections, engineering backlog items, and any strategic priorities from your last planning session. The quality of your roadmap output is directly tied to what you feed in.

2 Open Claude or ChatGPT and paste your inputs with a structured prompt asking the model to identify the top themes, group them into potential initiatives, and flag gaps. Ask it to return a structured list, not prose, so the output is easier to work with.

3 Use a second prompt to score or rank the initiatives using a framework you specify — RICE (Reach, Impact, Confidence, Effort) works well. Provide your own rough estimates and ask the model to surface inconsistencies or flag where your confidence scores look optimistic.

4 Ask the model to draft a one-paragraph rationale for each top initiative, written for a non-technical stakeholder audience. Paste this into your roadmap document — it saves the 'why are we doing this' writing that often gets skipped.

5 Run a dedicated prompt to generate the 'what we're NOT building' section. This is often the most useful part of a roadmap for alignment, and LLMs are good at inferring reasonable exclusions from a prioritized list.

6 Use Gemini or ChatGPT to convert your structured outline into a presentation-ready narrative — either a written brief or slide-by-slide talking points — by describing the audience (board, team, investors) in your prompt.

7 Manually copy the final output into whatever tool your team actually uses — Notion, Google Slides, Linear, or a shared doc — and circulate for feedback. The AI produced the draft; the distribution and revision cycle is still entirely manual.

Prompts you can copy

Here are 12 customer support themes and 8 sales objections from last quarter. Group them into 4-6 product initiatives, name each one, and explain the underlying user problem each solves. Return as a structured list.

Score each of the following 6 initiatives using the RICE framework. I'll provide rough estimates for each dimension — flag any where my Confidence or Effort scores seem inconsistent with the description.

Write a 3-sentence stakeholder rationale for each of these 5 roadmap items. Audience is a non-technical board of directors. Explain what we're building, why now, and what success looks like.

Given this prioritized roadmap, what are the top 5 things we are explicitly NOT building in the next two quarters? Write these as clear, defensible exclusions with one-line reasoning for each.

Turn this roadmap outline into talking points for a 15-minute all-hands presentation. Structure it as: context, the 3 big bets, how we prioritized, and what we need from the team.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No connection to your actual backlog — you copy-paste items from Jira, Linear, or Notion by hand, and the roadmap immediately diverges from ground truth the moment a ticket changes.

Customer feedback lives in Intercom, support tickets, and sales call notes; none of it flows into the LLM session automatically, so you're always working from a manually curated, probably incomplete sample.

Nothing persists between sessions — next quarter's planning cycle starts from a blank prompt window, and the structured rationale you spent two hours building last time isn't there to build on.

Outputs drift run to run; the RICE scoring table you carefully formatted in April looks structurally different when you re-run the same prompt in July, requiring manual reconciliation every cycle.

Stakeholder-ready formatting is still a manual step — the LLM produces solid content but you're copying into slides or docs yourself, applying your brand, and fixing layout every single time.

Meeting decisions don't feed in automatically — action items and priority shifts from your last roadmap review exist in someone's notes, not in the document, until a human manually reconciles them.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — an agent builds the persistent apps and automations that handle this workflow on your live business data, so your roadmap isn't a document you update manually but a surface that stays connected to what's actually happening.

The Knowledge Management starter app gives your roadmap a home that's connected to your existing docs and searchable by the whole team — no more 'where's the latest version' conversations or stale Google Docs floating around.

Connect Notion from Starch's integration catalog and the agent queries your actual backlog, project pages, and decision logs live — your roadmap rationale can reference real initiatives, not a copy-pasted snapshot from two weeks ago.

Meeting Notes captures your roadmap review sessions in real time, extracts decisions and priority changes, and archives them in searchable history — so 'didn't we deprioritize that last month?' has an answer you can find in seconds.

Describe the roadmap structure you want in plain English — 'build me a quarterly roadmap view organized by initiative, with RICE scores, owner, and status pulled from our Linear tickets' — and the agent builds it as a persistent app, not a one-off document.

The Presentation Agent (currently in development — request beta access) will convert your roadmap directly into a polished deck for board or all-hands use, without the Sunday-night slide formatting session.

Automations run on a schedule — wire up a weekly summary that checks for new customer feedback themes across your connected tools and surfaces anything that should influence the roadmap, delivered to Slack before your planning meeting.

Get closed-beta access →

Toolkit