AI that adds suppliers instead of capacity

By Jason Sibley

The problem

You signed up for an AI tool to get more done without hiring anyone new. That was the plan. Six months on, look at what's actually on the books: the subscription itself, an agency that came in to "set it up properly", and a freelancer who fixes the output before it goes anywhere near a customer.

Say you run a small lettings agency. You bought a tool to draft landlord emails and summarise viewing notes. Someone had to configure it, connect it to your systems, write the prompts: a contractor, a fortnight's work. The tool runs now, mostly, but the drafts still need a proper read-through before they go out, so you've quietly added a part-time pair of eyes to your costs. You didn't hire a person. You hired three suppliers instead.

Or say you run an online gift shop and bought an AI tool for product descriptions. The output reads fine until it invents a feature the product doesn't have. Now a freelance copywriter checks every batch before it's published: the subscription was meant to replace that job, and instead created a smaller, stranger version of it.

The tools mostly work. What goes wrong is everything around them.

What the numbers say

Coface, the trade credit insurer, ran its 2026 UK Payment Survey in July 2026 (sample size: 608) and published the results on 15 September 2026. Worth flagging: this is a survey of companies generally, not small firms specifically. Even so, the headline is hard to ignore. 80% of companies using AI said the technology had changed how they use external providers, and most of those reported increased reliance on outside expertise, not less. Coface's own framing is blunt: AI is introducing another cost layer, rather than cutting supplier spend. Among large companies, customer support was the only function where more businesses brought work in-house than pushed it out to suppliers.

One extra point from the same survey: average agreed payment terms have stretched to 57 days, with 89% of UK companies reporting payment delays over the past year. Slower cash coming in makes an extra layer of supplier cost sting more, not less.

The ONS tells a similar story from a different angle. Its report "Artificial intelligence in UK businesses: 2023 to 2026", released 20 July 2026 and based on the Business Insights and Conditions Survey from June 2026, found that self-reported AI use among businesses with 10 or more employees rose from around 12% to around 35% since late 2023. That's a real jump. But adoption is described as relatively shallow: the average number of AI technologies per adopting business only rose from about 1.4 to about 1.6. Most adopters are using one or two tools at most. And the gap by size is wide: 28% of businesses with 0 to 9 employees reported using at least one AI technology, against 49% of those with 250 or more staff.

There's a reassuring bit in the same report. Most businesses report AI hasn't changed their overall headcount so far. Around half, across size bands, reported no impact at all. Just under 7% of medium-sized businesses reported a decrease. Among businesses specifically using AI to improve operations, 63% reported no change in headcount, 1% an increase, and 6% a decrease. So far, AI isn't thinning out payrolls in any broad way. The Coface numbers suggest the cost may be showing up on the outside bill instead.

Why it happens

A few patterns keep showing up.

The tool gets bought before anyone's agreed which workflow it's actually for. Someone sees a demo, likes it, signs up. Then the team has to work out afterwards what it's meant to replace.

Nobody inside the business owns it. It sits with whoever happened to sign up, and when they're busy with something else, it drifts.

Setup gets outsourced, and the knowledge of how the thing actually works leaves with the contractor when the invoice is paid. Nobody on the inside can explain why a prompt is written the way it is, or adjust it when something changes.

And because the output isn't quite trusted, it goes back out again, to a freelancer or an agency, for a check and a fix. That loop can run for months without anyone noticing it's become a fixed cost.

There's a skills angle here too. The ONS found training or retraining existing staff is the most common way businesses build AI skills; recruiting new AI-skilled staff or outsourcing roles remains relatively uncommon. Among firms with 10 or more employees who cite a lack of AI expertise as a barrier, around 62% respond by training people they already employ. Yet only 11% of businesses with 10 or more employees say more than half their workforce has had any AI-related training. Most businesses know training is the right lever. Most haven't pulled it properly yet.

Worth knowing, too, how most businesses actually get their AI tools in the first place. Government research from DSIT found that among AI adopters, most bought ready-to-use external software rather than building anything in-house or outsourcing development. For text generation tools specifically, 71% bought external software, 14% built in-house, and 9% outsourced the development. Buying off the shelf is fine. Trouble starts after the purchase.

How to keep the work in-house

Pick one workflow. Not five. One thing the AI tool does, end to end, that someone can point to and say "that's what it's for." If you haven't settled this yet, Pick one AI workflow and stick with it for 30 days is worth a read before you sign anything else up.

Name an internal owner. A real person, not "the team." Someone who can explain, in a sentence, what the tool does and why.

If you do bring in a supplier to set something up, ask for a written handover and a how-to guide your own team can follow without them. Not a call where they talk you through it once. A document.

Train the person who'll actually use the thing day to day. Given how much more common this is than hiring or outsourcing for AI skills, it's the plainer option, and often the one businesses skip.

Keep a simple list: every AI tool and every supplier attached to it, marked "in" (done internally) or "out" (paid to someone else), with an owner against each line. It takes an afternoon, and it'll probably surprise you.

Then review it after 90 days. Ask one question: did headcount-hours or supplier spend actually go down because of this tool? If the honest answer is no, that's useful information, not a failure.

When outsourcing is the right call

None of this means never use a supplier. A one-off build, say a custom integration or a data migration, is often cheaper to buy in than to learn from scratch. Specialist work, the kind that needs deep technical knowledge you'll only use once, is what outsourcing is for. Where it goes wrong is paying someone outside for the same ongoing task, month after month, when it could have been trained into someone on your own team.

The one decision to make this week

Look at your current AI tools. Pick the one workflow that matters most. Write down who inside your business owns it. If the answer is nobody, that's this week's job.

For a wider look at getting more from the people you already have, see The "do more with the team you have" week.

A soft word before you go

If you'd like a second pair of eyes on which AI tools are earning their keep and which have quietly become supplier spend, My AI Apprentice's SMB AI sessions are built for exactly that conversation, and the point is that your own team leaves able to run the thing. No pressure either way.