AI Won't Save a Weak Product.
It Exposes It Faster.
Leonid Goriev
Founder Alty
May 25, 2026
A few weeks ago I spoke with a CPO at a UK retail bank. They had just shipped an AI assistant inside their app. Engagement was up. Retention hadn't moved. The same gap that existed before the launch was still there. AI had only made it visible faster.
This is not an unusual story. We are seeing the same pattern across UK financial services.
A team ships an AI feature. The launch post goes out. Three months later, the metrics that mattered before still haven't moved. Sometimes they have worsened.
The AI wasn't the problem. The product underneath it was.
The shift no one in product is naming clearly
According to GP Bullhound's Consumer Subscription Software Report 2025, 40% of searches now end without a click. Users get an AI summary and move on. By 2028, AI search traffic is projected to surpass traditional search.
This is not a marketing story. It is a product distribution story.
Until now, weak products could survive on SEO, paid acquisition, and brand recall. The discovery layer was forgiving. You could outspend the gap between what your product actually was and what users actually needed.
AI search closes that escape route.
When someone asks an AI assistant "what's the best way to send money internationally from the UK," the assistant doesn't pick the loudest brand. It picks what the underlying data suggests is genuinely worth recommending. Trust signals. Retention curves. Editorial citations. App ratings.
A bank that has spent the last three years patching its UX does not make that list. A Wise, a Starling, a Monzo, or a Revolut, depending on the use case, does.
That is the part most product teams in UK banking are not internalising fast enough.
The trap: bolting AI onto a weak product
The most common reaction we are seeing right now is to ship an AI feature fast. Show the board momentum. Get a headline.
The teams doing this end up in one of two places.
In the first, the AI feature works in isolation but doesn't improve the product. Retention stays flat. CAC stays the same. The AI line item now sits in the budget as a fixed cost, not a return.
In the second, the AI feature accelerates whatever weakness was already there. If onboarding was broken, the AI assistant gives users one more reason to drop out before they understand the product. If retention was thin, the AI feature gives them one more thing to lose interest in.
GP Bullhound's report puts it directly: the most successful AI integrations won't feel like AI. They reduce friction. They make the product more responsive. That is not a content take. It is a structural prediction about which products survive the next 24 months.
What the teams handling this well actually do
Three patterns. Each one is a product decision before it is an AI decision.
They fixed the core experience before AI became the conversation.
We worked with Recharge — a power bank rental service where the product's entire value depended on instant activation. Users needed a charged phone in the next sixty seconds, not after a five-minute onboarding flow. The fix wasn't a better app. It was removing most of the app. Registration gone. Login gone. Onboarding compressed from minutes to seconds.
That kind of decision is the precondition for AI to add value later. Without it, AI is decoration.
They treat AI as a product decision, not a feature decision.
The teams making AI work are not asking "where can we add AI?" They are asking "what part of this product is still slower or less responsive than it should be, and would AI close that gap?"
The starting point is the friction, not the technology.
They think about retention as structure, not as a campaign.
Bundling reduces churn by up to 50%, according to the report. Not because users get more options. Because the product becomes harder to leave. That is a product and design decision. Most teams treat it as a pricing one and wonder why churn doesn't move.
The product question for 2026
Not "are we using AI" — most teams are.
But: if an AI assistant were asked to recommend a product in your category, would yours come up? And if someone followed that recommendation, what would they find?
The question leads somewhere specific: how does a user open this for the first time, what do they do in the first sixty seconds, and why do they come back.
If you want to see more of how we approach this, our work is a reasonable place to start.
Spotted by Yelyzaveta Yatsuk, our Partnership Manager.
Source: GP Bullhound Consumer Subscription Software Report, October 2025. gpbullhound.com
