I’ve watched dozens of SMEs bin their AI tools within weeks of buying them.

The pattern is always the same. Someone gets excited about ChatGPT or Claude. They subscribe to the premium version. They try it for a few tasks. It doesn’t work perfectly. They give up.

The problem isn’t the technology. The problem is expecting AI to work like software when it actually works more like a junior hire.

We’re building an AI and automation apprenticeship at The Coders Guild, launching in 2026. What I’m seeing in SMEs right now is people experimenting without strategy. Someone in the team is “doing a bit with AI” but there’s no plan, no vision, no structure.

That’s a recipe for disappointment.

The Technical Bit Isn’t the Difficult Bit

Most SMEs make the same mistake. They hand AI to their most technical person and say “sort this out.”

That person can set up agents. They can configure tools. They can write prompts.

But the technical stuff isn’t what makes AI useful. Understanding your business is what makes AI useful. Understanding your processes, your service design, your product, your data management.

According to research from Latent Bridge, organisations getting good results invest 70% of AI resources in people and processes, not technology.

The failure happens because you’re trying to automate workflows you haven’t properly mapped or measured.

You Need to Tell AI About Your Business

If you want AI to help you be more efficient, you need to know what your state of play is right now.

Who needs help. Where the bottlenecks are. What actually happens when a customer places an order or when a project kicks off.

You need to create documents. Onboard the AI into your company, into your workflow. Have people comfortable with it the same way they’d be comfortable with a new person joining the team.

This is where most SMEs discover something uncomfortable.

They haven’t actually documented their processes in a way that would make sense to explain to anyone.

AI can help you document things. You can use it to create standard operating procedures, reference documents, process maps. But you need to do that foundational work first.

The Chicken and Egg Problem

You need documented processes to use AI effectively. But AI could help you document those processes.

When SMEs realise they’re stuck in this loop, the answer is simple. Start with the people.

Draft things out. What would you do to onboard someone correctly? What’s missing? Build those things first. Get a good understanding of your business, your processes, your value.

Then start looking at how to implement AI more widely.

What Good Onboarding Actually Looks Like

I’ve onboarded hundreds of apprentices into SMEs over the past decade. The pattern is clear.

When you’re onboarding a school leaver with no workplace experience, they need to know loads more than usual. The teams that make that effort, that put in the extra work, get better results from having the apprentice.

The companies that do it well prepare materials about the business. The people. How the company makes money. How the organisation exists. Who does what.

All of these things are useful for a functioning AI helper.

If you want to use ChatGPT Teams or Claude for Teams properly, you want projects and documents that describe how your business ought to work.

Notice I said “ought to work” not “does work.”

By documenting what you’re doing now and having people review those documents, you start exposing pain points and opportunities for improvement. The act of preparing to use AI properly makes your business better before the AI does anything useful.

We experienced this recently when we switched from ChatGPT to Claude as our main shared resource. The transition forced us to think about what had changed in the past year. Redoing that work in a different way gave us more insight into what we’re doing as a business currently.

It wasn’t some dramatic revelation. It just got better organised because we redid it.

Where the Analogy Breaks Down

AI isn’t a person. The differences are obvious.

You’re going to look after a real person more than you’ll look after AI. AIs don’t get tired. They don’t respond badly to feedback. They don’t need sick days or holidays.

But the preparation is similar.

Maybe the better comparison isn’t even a junior hire. What would you do to prepare for a consultant coming in to help with something specific? You’d prepare materials. You’d ask them what they needed to know.

Those are the sorts of things you can do with AI as well.

The Expectation Problem

People think AI is something it isn’t.

According to Virtual Sapiens, the main misconception with creators and small business owners is expecting AI to complete the entire workflow. It doesn’t do everything for you.

The data backs this up. Research from MIT and RAND Corporation shows 70-85% of AI initiatives fail to meet expected outcomes. More striking: 42% of companies abandoned most AI initiatives in 2025, up sharply from 17% in 2024.

The tools are improving. The models are getting much better at asking for extra information when they need it.

But the preparation gap stays the same.

People aren’t doing the work needed to get good results. So they’re either disappointed with what they get, or they’re blaming the tools.

What to Do This Week

If you’re reading this and thinking “we need to do the prep work,” start here.

Talk to your team. Have a think about what you need to understand your business properly. What does your AI need to know?

What custom GPTs or Claude projects do you need to create to make this working environment work for everyone, not just as a content producing machine or data crunching tool?

Those use cases are fine. They’re useful. But they’re only scratching the surface.

Real business efficiency comes from mapping out whole processes and having AI assist throughout. From lead generation to onboarding. From project kickoff to delivery.

The more context you provide, the better you get at training AI to give you the outputs you want.

According to Mind Studio, 76% of enterprises now include human-in-the-loop processes. Organisations implementing these workflows report accuracy rates up to 99.9%, compared to 92% for AI-only systems.

The combination of AI speed and human judgement catches errors that either would miss alone.

The Window Is Closing

We’re at a critical point for SMEs and AI adoption.

The tools are getting better. The costs are coming down. The capabilities are expanding.

But if SMEs don’t shift how they approach AI implementation, we’re heading for widespread disillusionment.

The businesses that figure this out now, that treat AI like a team member requiring onboarding and management rather than magic software that should just work, will have a significant advantage.

The ones that don’t will keep binning their subscriptions after a few disappointing weeks.

Your choice.