Short answer
“We need an AI roadmap” deserves a strategy answer, not a chatbot by the end of the quarter. Three layers decide it: where AI removes real friction in your customers’ workflows, where it changes your own cost structure, and whether it threatens your category’s pricing. Ship one, write the other two.
The competitor’s launch demo looks slick. Two prospects asked about your AI story this month. A board member forwarded an analyst chart with a steep line on it.
Engineering has a prototype that impresses in demos and does nothing in production, and the honest internal answer to what AI should do for your customers is a shrug wearing a roadmap slide. That gap between the pressure and the answer is the actual subject here.
The three layers, in decision order
Order matters here as much as content, because the layer that generates board comfort is not the layer that generates customer value, and doing them in the wrong sequence produces an expensive demo and a flat usage graph. Work through them in the order below and the ninety-day answer assembles itself out of things you were going to need anyway.
1. The customer workflow layer, where you build
Find the steps in your customers’ actual jobs where they burn hours that your data could collapse: triage, drafting, classification, lookup. This comes from watching the work in customer interviews rather than from a feature brainstorm, and one shipped collapse of a genuinely hated hour beats five copilots nobody opens twice.
2. The cost structure layer, quietly and in parallel
Where AI changes your own economics: support deflection, onboarding automation, content operations. None of it appears on the product page of a board deck and all of it appears in the margin line, which is the number most boards read first. This layer is often the fastest to produce a defensible result.
3. The category threat layer, where you write
Could an AI-native entrant deliver your core outcome without your product’s shape? If the honest answer is plausibly yes, that is a strategy question about where your moat actually lives: workflow lock-in, proprietary data, integration depth, or trust. It deserves a written page the board reads, whatever the conclusion.
“Every company I have seen ship AI feature theatre had the same thing missing beforehand, and it was not machine learning expertise. It was a working way of finding out what customers actually spend their day doing.”
What to bring the board in 90 days
Three artifacts, and together they convert the question from “do you have an AI roadmap” into “this team knows what it is doing,” which was the real question underneath. Notice that only one of the three is a shipped feature. The other two are evidence of thinking, and boards are considerably better at valuing that than founders expect.
- One shipped, measurable workflow win with usage data attached, not a demo.
- The internal cost-layer experiments with their numbers, however modest.
- A one-page category position, written, that says where the moat is and what would change your mind.
- An honest note on what you deliberately chose not to build, which is the part that signals judgement.
The trap worth naming out loud
AI pressure is the current costume of an old problem: building for the board’s fears rather than for customer evidence. Companies with a working evidence engine, meaning interviews, usage truth and someone with the authority to decide, metabolise AI the way they metabolised mobile and then cloud before it. Companies without one ship feature theatre and call it a roadmap. The gap the pressure exposes is almost never an AI gap.
If the deeper issue is that priorities flip whenever an external signal arrives, the AI conversation is a symptom and our piece on strategy that changes every week is the more useful read. If the issue is that nobody owns the call at all, capacity versus direction covers that, and the roadmap nobody owns describes how it looks from inside the team.
AI pressure, and no answer you believe?
Thirty minutes, no pitch deck. You will leave with the three-layer answer sketched for your product.
Book a Product Strategy SessionFrequently asked questions
Our competitor’s AI launch looks far ahead. Are we late?
Watch their usage rather than their launch video, which is genuinely hard from the outside but not impossible: look at whether the feature appears in their docs, their pricing, their customer stories and their job listings six months later. A great deal of first-wave AI functionality in B2B SaaS shows demo-grade adoption. Shipping the workflow customers actually run beats shipping first, and it beats it by a wide margin over two years.
Do we need AI-specialised product leadership for this?
You need product leadership that runs on evidence. The AI-specific part is learnable and moving too fast for specialisation to hold its value; the judgement part is neither. Be suspicious of anyone selling AI strategy detached from your customers’ workflows, because that is a deck rather than a direction.
What if the honest answer is that AI does not help our customers much?
Then write that down with the reasoning and bring it to the board, which takes more nerve than shipping a copilot and is worth considerably more. A well-argued position that the category is not yet AI-shaped, with the conditions that would change your view, is a defensible strategic statement. A copilot nobody uses is a defensible-looking one that fails quietly in twelve months.
How do we stop the prototype becoming the strategy?
Put a usage bar on it before it ships and agree the bar in advance. Prototypes become strategies because nobody ever defines what success would look like, so the demo’s reception in the room becomes the only available evidence. Name the metric, the population and the window first, and the prototype either earns its place or is retired cleanly.
The board is asking whether you have thought about this seriously. One shipped workflow win and two written positions answer that far better than a feature announcement, and they leave you with something customers actually use.