Where AI actually belongs in your product
There is an electric toothbrush on sale with “with A.I.” printed on the box. That is roughly where we are: the label has become the feature, and adding it appears to sell more units.
The pressure on founders is real. If you are not shipping something with an AI angle, it feels like you are missing an opportunity — or worse, being left behind. So teams bolt an assistant into the corner of the interface, tick the box, and move on.
I think that is the wrong instinct, and I think it is a product problem rather than a technology one. The question is not whether to put AI in your product. It is where.
The bar is genuinely high
It is worth being clear that the sceptical position is not the safe one either. When Klarna launched its AI assistant, within a month it was handling two-thirds of all customer service chats — 2.3 million conversations, the equivalent work of 700 full-time agents, resolution times down from 11 minutes to under two, a 25% drop in repeat enquiries, and an estimated $40 million profit improvement in a year.
That is not a gimmick. That is a business function rebuilt.
The broader pattern is that the large “wide” platforms have now established themselves, and that is enabling a wave of “narrow” tools built on top of them. The infrastructure question is largely settled. What is unsettled — and where the value now sits — is knowing which specific problem to point it at.
Assistants are the obvious answer, which is the problem
Design patterns are already establishing, and the dominant one is the AI assistant: a chat panel bolted onto an existing product. They tick the box. They can be genuinely useful. But most of them feel opportunistic rather than considered — a response to the moment rather than to anything a user asked for.
My view is that as this matures we will see much less “Powered by AI” and far more seamless experience, where the user does not know or care that AI is involved. It will move from being an exciting, delightful feature to being table stakes — and the products that win will be the ones where it is doing something specific and invisible rather than something general and announced.
The balance to strike is between gimmick and value. Which means starting from the experience, not from the technology.
Two ways to find the right spot
Both of these are ordinary product techniques. Neither is about AI. That is exactly why they work — they find the problem first, and only then ask whether this particular tool is the right fix.
Customer journey mapping
Map every interaction a customer has with you, from first contact through to post-purchase, across every channel. For each phase — registration, onboarding, first session, and so on — capture four things: what the customer actually does, which part of your service they touch, what they are thinking, and how they feel.
The feelings row is the one people skip and the one that matters. It is where you find that registration is delightful, onboarding is confusing and frustrating, and the first real session is good again. That dip in the middle is worth more than any amount of speculation about what to build.
Then, and only then, go through each identified opportunity and ask whether AI improves it. Some will be obvious candidates — shortening a training flow, surfacing the right template at the right moment, triaging a support queue. Many will not, and the discipline is in leaving those alone.
The speedboat technique
A collaborative workshop with users, or potential users. Draw a speedboat. Ask what is pushing them forward and what is dragging them back.
The anchors are the things slowing customers down. The engines are what is working and could work harder. Prioritise the biggest anchors and the most impactful engines, then build an action plan against them.
It is a fast, cheap way to get honest input, and because it is framed around the boat rather than around your roadmap, people tell you things they would not otherwise say.
What this changes
Run either exercise and the conversation shifts. You stop asking “where can we put AI?” and start asking “which of these specific, evidenced problems does it solve better than anything else?”
Usually the answer is one or two places, not ten. And those one or two tend to be unglamorous — the pain point in onboarding, the repetitive triage nobody enjoys, the personalisation that was always too expensive to do by hand. They rarely make good launch copy. They are usually the ones that show up in retention.
The founders I see getting this right share a habit: they treat AI as one option in a product decision rather than as the decision itself. They are also, almost without exception, the ones with someone senior owning product — because the failure mode here is not technical, it is a roadmap driven by what is fashionable rather than what is evidenced.
If this is where you are
This is a lot of what I do. I work with founders as a fractional CTO and CPO — technical and product leadership in one seat, which in practice often means running exactly these sessions, then making sure what comes out of them is buildable and worth building.
If you are staring at an AI roadmap you are not confident in, or you have shipped an assistant and cannot tell whether it is earning its place, tell me what you are working on. A short conversation is usually enough to tell whether there is something real there.