60% of small businesses now use AI. Only 23% got anything back and they're renting their results.

Eighteen months ago, "are you using AI yet?" sorted businesses into two groups, and one of them had an advantage.

Ask it today and it sorts nobody. Sixty per cent of small businesses use AI, up from eighteen per cent the year before. Ninety-seven per cent of executives have deployed an agent of some kind. Your competitor has it. The bloke who does your books has it.

So the question stopped being useful, and almost nobody replaced it with the better one.

What did it actually pay you?

Only 23% of executives report a significant return from AI agents. Forty-eight per cent describe their adoption as a "massive disappointment". Seventy-five per cent admit their AI strategy is "more for show than actual guidance".

And the part that should stop you cold: investment levels don't correlate with returns. Fifty-nine per cent of companies spend over a million a year on this. Spending more is not what separates the winners.

Which kills the usual response to disappointing AI results, which is to buy more AI.

The disappointed majority bought a capability and went looking for a use case. They got a tool that writes, or summarises, or answers, then asked the business to find something valuable for it to do. Predictably it ended up drafting emails nobody needed.

The 23% started at the other end: with a specific, expensive problem that was already costing them money every day. The technology was the last decision they made, not the first.

But there's a reason so many businesses ended up holding a capability and hunting for a use. They didn't go looking for one. It was sold to them.

Be blunt about where most "AI strategy" actually came from. It came from an email. From HubSpot, or Salesforce, or your phone system, or whichever software you already pay for monthly. Good news, we've added AI. Usually one tier up.

That isn't a product built for a problem you have. It's a retention feature. These companies live or die on churn, and the cheapest way to make a subscription feel indispensable is to sprinkle AI on the thing you were already buying.

So you get an AI that writes subject lines, summarises the call you were already on, and scores leads with a number nobody acts on. Real technology, doing very little.

It isn't supposed to move that number. It's supposed to make leaving feel expensive.

Here's the part that should genuinely bother you. While you use it, you're teaching it.

Every call transcript. Every objection your best salesperson handled brilliantly. Every reason a deal died. Every pattern in who buys from you and what tipped them over. That's the accumulated commercial intelligence of everyone who has ever worked in your business, and it's going daily into a system you rent.

Then stop paying. It isn't deleted or handed back. It's switched off, on the other side of a wall, still running. You can export your contacts, a CSV of names and numbers, the least valuable thing in there. You cannot export what it learned.

So it was never yours. You were renting it, and the rent is now non-negotiable, because leaving means walking away from every pattern your business spent three years teaching it.

That's not a software subscription. That's sharecropping. You did the work, on their land, and the harvest stayed.

Ask why the version you're renting is so thin. Look at what these models can actually do, then look at what your CRM's AI is permitted to do for you. Write a subject line. Summarise a call. Score a lead out of a hundred.

That gap is not a technical limitation. It's a pricing decision.

An AI that took everything your business knows and went and acted on it, at volume, would be worth ten times what you pay. It would also be the last thing that vendor ever sold you, because a customer with that much leverage doesn't need the rest of the platform.

So you get a version calibrated carefully. Useful enough to renew. Never useful enough to make you dangerous.

While everyone was busy adopting, a second shift happened, and it reprices every enquiry you receive.

Buyers now arrive having already researched. They've had an extended back-and-forth with an assistant that shortlisted options for them. Two consequences, both expensive.

The enquiries you get are worth more. They're pre-qualified, further down the funnel, and know roughly what it costs. Every one you fumble is a more expensive fumble than it used to be.

And they won't wait. Someone who just had a fluid, instant conversation with an AI then fills in your contact form. Yours replies in four hours, if it's a weekday. They didn't wait. They were comparing three suppliers in one window.

Our system can reply in under five seconds. We deliberately don't.

A four-second reply tells the person one thing: a machine did this. Nobody types a considered response to a web form in four seconds. So you win the speed metric and lose the conversation, because the moment someone knows they're talking to software they stop explaining themselves and start giving one-word answers.

We set ours to forty-five seconds. Fast enough to be remarkable, since your competitors measure this in hours. Slow enough to be believed. And those forty extra seconds go into making the reply worth reading.

Which generalises: the businesses winning with this stuff optimised the customer's experience, not the vendor's metric.

Your CRM is full of people who once put their hand up. They enquired, downloaded something, got a quote, then went quiet. Not because they said no, but because the timing was wrong or nobody had the hours to chase them.

You have already paid for every one of them. The ad spend is gone whether you follow them up or not.

Following them up properly isn't a hard problem to understand. It's a hard problem to staff. Nobody has the headcount to personally text four thousand old leads, handle the "not right now"s, requalify the ones whose circumstances changed, and book the live ones.

It's unglamorous and entirely solvable, and it's the best-aimed thing you can point AI at, for one reason: the return is measurable in a week, in booked meetings rather than hours saved.

One partner's dormant database produced $1.4 million in fee value, $725,000 on the first pass and $675,000 on a second, before they'd paid us a commission. Eight thousand dormant contacts for another client produced 596 enquiries, 248 booked calls and $400,000 in a single month. A mortgage broker with nine hundred subscribers got nine booked calls in the first twenty-four hours, a hundred and forty in total, and settled sixteen loans from leads he'd written off.

None of those are AI stories. Nobody in them bought a model.

Stop asking whether you're using AI. You are. Everyone is. It tells you nothing about whether you're winning.

Sort the retention features from the revenue features. Most of the AI in your stack arrived as an upgrade notice from a vendor who needs you to renew. Very little of it was designed to make you money.

Ask who keeps the learning. If the intelligence your business generates only ever accumulates inside a platform you rent, you're improving an asset you'll never own.

Start with the leak, not the tool. Where does money already fall out of your business every week? Unanswered calls. Slow replies. Leads that went quiet. Fix the expensive leak, then choose the technology. That's the easy part.

The AI advantage was never going to come from having it. It was always going to come from aiming it.

The businesses in that disappointed 48% aren't there because they bought the wrong product. They're there because they never chose one. It was added to their subscription, at a higher tier, by a company whose actual objective was that they not cancel.

Find the most expensive thing your business does badly, and point something at it.

For most businesses we meet, that thing is a database full of people who already said yes once, and never heard from anyone again.