AI Roadmap Prioritisation Is Solving the Wrong Problem

Every major roadmap tool now has an AI button — and if you've pressed it, you already know it doesn't actually help you prioritise.

5 min read·1 June 2026·Fredrik Göth

You open Productboard, Airfocus, or Aha!. You click the AI button. A roadmap appears. It looks reasonable. Features are scored, grouped, and sequenced. Someone who didn't know better might think you'd done the work.

But you know something is off. Because the tool didn't ask you about the constraint that's actually driving your next quarter. It didn't ask what your company is betting on this year. It didn't ask whether you're optimising for retention or acquisition right now, or whether engineering is at 60% capacity because of a platform migration nobody put on the roadmap. It just sorted your backlog and dressed it up as a plan.

That's the uncomfortable truth about AI roadmap prioritization in its current form: it's automating the part of prioritisation that was never the hard part.

"The problem is almost never that teams don't know how to prioritize. It's that they're optimizing for the wrong thing."
— Shreyas Doshi

What AI roadmap tools actually do

Sort. Score. Visualise. Summarise. These are the four things current AI prioritisation features are genuinely good at. Give a tool your backlog, some impact and effort estimates, maybe a few user votes, and it will produce a ranked list faster than you could in a spreadsheet.

That is useful. It is not prioritisation.

Prioritisation is reasoning under constraint with strategic context. It's the conversation where you say: "I know Feature B scores higher on user impact, but if we ship it before Feature A, we lose the enterprise deal that's funding Q3, and the team doesn't have capacity for both anyway because two engineers are on the infrastructure work we committed to six months ago." That reasoning requires holding your company's strategy, your current resource reality, your competitive position, and your actual bets in context simultaneously.

No current AI roadmap tool does that. Because none of them have access to it.

Speed is the wrong metric

My experience is that the most dangerous thing about the current generation of AI prioritization for product managers is how good it looks. A beautifully formatted roadmap, produced in ten minutes, with scores and colour coding and a logical-seeming sequence.

The problem is you can now make a faster mistake.

If the strategic context is missing from the tool's reasoning, producing output faster doesn't help. It just means you get to the wrong place sooner and with more confidence. I've seen teams bring AI-generated roadmaps into leadership reviews that looked air-tight and were fundamentally wrong because the tool had no way of knowing the company had quietly shifted its ICP three months earlier.

Shreyas Doshi put it well when writing about prioritisation failures: "The problem is almost never that teams don't know how to prioritize. It's that they're optimizing for the wrong thing." AI roadmap tools, as they exist today, can optimize your backlog against your backlog. They can't tell you whether your backlog is aimed at the right thing in the first place.

Why the gap exists

The structural reason is straightforward. These tools are trained and configured around backlog data: feature requests, user feedback, effort estimates, ticket histories. That's the data they have access to. That's what they can reason about.

What they don't have is your strategy document, your board commitments, your sales pipeline, your engineering capacity by squad, or the competitive move your roadmap needs to anticipate. That context lives in Notion, in your head, in a slide deck from last quarter's offsite, and in a conversation you had with your CRO last Tuesday.

Until a tool can hold all of that as persistent context, its AI prioritization layer is a visualisation feature wearing a strategy hat.

A test to apply to any tool claiming AI prioritisation

Ask it to justify why Feature A should ship before Feature B given a specific constraint you name out loud. Not a generic constraint. Your constraint. "We have one iOS engineer available for six weeks and we've committed to three enterprise customers that the bulk export feature ships in Q3."

If the tool can reason through that trade-off and show its work, you have something worth using. If it gives you a re-ranked list based on impact scores without engaging with the constraint, you have a visualisation tool. That's fine. Just don't call it prioritisation.

A genuine AI prioritisation layer would need to hold company strategy, current bets, resource constraints by team, customer segment priorities, and competitive pressure in active context. Use that list as your evaluation standard. Most current tools satisfy none of it.

What to do with this

Stop evaluating AI roadmap tools on speed and output quality. Start evaluating them on what context they can actually hold and reason with. If the answer is "just my backlog," use it for what it's good at, and keep the strategic reasoning where it's always lived: in the room with the people who know what the company is actually trying to do.

That part isn't going to be automated any time soon. And pretending it has been is the fastest way to produce a roadmap that looks right but isn't.

Fredrik Göth is a CPO and product leadership consultant working with product teams across Europe.

References

Shreyas Doshi — Prioritisation failures and optimising for the wrong thing (2023)