Why I think businesses are using AI in the wrong place

What it got me thinking about

In his Daily Spark piece, published on 24 October 2026, Slok looks at revenue per worker, margins and productivity indicators. His argument is that the wider financial payoff from AI is not yet showing up clearly beyond the large technology companies.

It reminded me of an experience I had with a pest control company. It was a small job, but it made me think about how easily automation can wear away a customer's trust.

The appointment that nobody seemed to know about

We needed a hornet's nest removed. We booked an appointment and received an automated email with a time slot. The team then arrived an hour before that slot. I wasn't there to meet them, so they moved on.

We called over the weekend and got through to an AI call agent. It told us we could rebook, but that the company's phone lines were closed on Saturdays and Sundays. That was confusing, because we'd made the original booking on a Saturday and someone had answered. Perhaps that had been an AI agent too; we hadn't necessarily realised at the time.

In practice, we still had to wait until Monday morning to speak to a person and get the appointment rebooked. Then the texts started arriving, asking us to rebook. They sounded personal, but they didn't seem to know that we'd already done it, or that we'd spoken to an actual member of the team.

That was the frustrating part. The company was communicating with us, but the messages didn't reflect where things actually stood. We'd already paid a deposit. Honestly, if we hadn't, we wouldn't have continued with them.

Sounding personal is not the same as understanding

Arriving early was a service problem in its own right. I can't say AI caused that. But the disconnected calls and messages made the experience worse. I had to keep track of what had happened while the systems contacting me appeared to have no shared understanding of it.

A message using your name and a friendly tone can look like good customer service. When it ignores the conversation you've just had, it feels quite different. It gives the impression that someone is paying attention, then shows you that they aren't.

Now imagine that on a £10,000 installation

Getting a hornet's nest removed is a relatively small purchase. It's irritating when it goes wrong, but it isn't the same decision as spending £10,000 on solar and a battery. With that sort of order value, I want a different experience, and I think most customers would too.

They've invited someone into their home, explained what matters to them and discussed practical compromises. If the follow-up then forgets that conversation, why should they feel confident the installation will reflect it? At the moment, I wouldn't ask AI to take responsibility for that relationship.

I would put AI behind the scenes first

This is my view on where some businesses may be going wrong; it isn't an explanation proved by Apollo's analysis. There are other reasons an AI investment might take time to pay off, including integration, poor data and the work needed to check the results.

I would start with the jobs the customer doesn't need to see: organising survey notes, grouping photos, extracting a kit list, checking it against an approved catalogue and preparing a draft handover. Keep measurements separate from estimates. Show what's missing. Make the information easier for a person to use.

Reading through AI slop is still work

If a salesperson has to wade through pages of AI slop just to work out where a conversation got to, what have we saved? They may as well have written the emails or had the conversations themselves. A quick generation time means very little if the next person spends half an hour reading and untangling it.

That is the design principle behind Solful: give the reviewer a clear choice between realistic options supported by the site survey. Those options should meet what the customer has asked for and reflect the needs the surveyor has identified. The surveyor can then decide and confirm, with any uncertainty visible.

The final email should stay close to what the surveyor actually said about the job. Review the design, check the pricing and confirm the recommendation, without reading a whole new essay to see whether it still means what you meant. That is how this should save time.

Help people carry the conversation forward

The surveyor still visits, listens and makes a recommendation. Back in the van, once parked, they talk through the job. AI carries that knowledge into the project and prepares the paperwork for review. The customer's requests and the surveyor's recommendations stay distinct, so the office can see what was discussed and what still needs deciding.

I would judge it on whether the whole job gets easier, including the time spent checking and correcting. Fewer facts to chase, a clearer proposal and less admin left at the end of the day. Keep people responsible for the relationship, and make the unseen work easier for them.