Why AI Isn’t Working for Your Business
(It’s Not the Technology)

Businesses know AI is supposed to help them. They’ve heard it enough times. What they’re less clear on is why, despite trying, it still doesn’t feel like it’s working. The answer isn’t that AI isn’t ready. The answer is that the way AI gets implemented for small service firms is broken at every level.

The Staffing Reality

The most direct route to operational AI is hiring someone who knows how to build it. So what does that cost?

In the UK, the median salary for an AI systems engineer is between £75,000 and £85,000 a year. Experienced engineers command over £100,000. For a 15-person recruitment agency or a local accountancy firm, that’s not just expensive — it may be more than the owner pays themselves.

And even if a firm could afford the salary, they’d need to manage a highly technical role in a field that changes every few months. Most small business owners aren’t equipped to evaluate a candidate’s AI engineering credentials, let alone manage them day-to-day.

The Consulting Trap

So you look at consulting firms. This is where small businesses often go next, and where they get the second shock.

• Enterprise AI projects typically start at £80,000–100,000 just to begin.
• Full implementations often run between £200,000 and £500,000.
• Most firms expect you to have an internal IT team to work alongside them.
• Timelines of 12 months or more are standard.

These firms aren’t built for a 20-person service business. They’re built for enterprise clients with dedicated IT departments, legal teams, and project managers. A Penrith recruitment firm or a Carlisle HR consultancy isn’t their client, and the product they sell isn’t designed for one.

The DIY Failure Mode

So you try to do it yourself. This is where most small businesses end up — and where the failure is most insidious, because it happens slowly and silently.

You start using ChatGPT for emails. You add an AI tool that promises to integrate with your CRM. You build a few Zapier automations. It works at first. The problem is the architecture.

This isn’t a failure of AI. It’s a failure of how AI was built into the business. Point solutions and chatbot overlays don’t add up to infrastructure. They add up to technical debt that makes the underlying problem worse.

What Actually Works

Operational AI infrastructure is different from AI tools. It’s a connected system: your CRM, email, documents, and calendar linked together with a decision layer that can reason over all of it and take actions.

A practical example:

• Your CRM flags a client who hasn’t replied in two weeks.
• Your email drafts a personalised check-in based on the history of that relationship.
• You review and approve it in 10 seconds. The system sends it.
• The interaction is logged, and the next follow-up is already scheduled.

That’s not a chatbot. It’s infrastructure. And it only works when the systems are genuinely connected — not through a series of fragile webhooks, but through a designed architecture with proper error handling, monitoring, and the ability to scale as the business grows.

What Real Infrastructure Requires

Building AI infrastructure that actually works in a production environment requires three things that DIY approaches almost never include:

• Monitoring — so you know when something breaks, before a client does.
• Error handling — so the system can recover without wiping state or corrupting data.
• Scalable design — so when you add a new process or new team member, the system grows with you rather than requiring a rebuild.

None of this is exotic. But it’s the gap between “we played around with AI” and “AI is a compounding advantage in our business.”

The Real Question

The question isn’t whether AI is good enough. It is. The question is who builds it for businesses that can’t hire an £80k engineer or pay a £200k consulting fee. That’s the gap Expanza was built to fill. Architect-level systems, built for UK service firms with 5–50 staff, at price points that make sense for a regional business.