Most UK businesses have tried AI. Far fewer have turned those experiments into something that runs every day and saves measurable time and money. Here is why AI projects stall, and the six steps that get them working.
Most UK businesses have now tried AI in some form. Someone on the team uses ChatGPT for emails, a director has experimented with Copilot, and there was probably a chatbot trial that never quite went anywhere.
What most have not done is turn those experiments into something that runs every day, saves measurable time and money, and does not rely on one enthusiastic person to keep it alive. That gap, between a promising trial and a working system, is where most AI projects stall.
This guide explains why that happens and sets out the six steps we use to get AI working properly inside a business. Each step links to a deeper article in this series.
Why do so many AI projects stall?
AI projects rarely fail because the technology does not work. They fail because the conditions around the technology were never set up.
In 2024, the analyst firm Gartner predicted that at least 30% of generative AI projects would be abandoned after the proof of concept stage by the end of 2025. The reasons it gave were poor data quality, weak risk controls, rising costs and unclear business value. None of those is a technology problem. Each one is a planning problem.
In smaller businesses, the pattern usually looks like this:
- The tool came before the problem. Someone bought licences, then went looking for a use for them.
- The AI knew nothing about the business. It gave generic answers, so staff stopped trusting it.
- Nobody owned the risk. There was no policy on what data could go in, so either everyone shared too much or a manager banned it altogether.
- Costs crept. Seats, add-ons and usage charges added up without anyone tying them to results.
- Nothing was measured. With no before-and-after figure, the project quietly lost its budget.
The good news is that every one of these is fixable, and the fixes are rarely expensive. They just need doing in the right order.
The six steps to getting AI working in your business
The framework below is adapted from the approach large cloud providers recommend to enterprise technical teams. We have translated it for UK SMEs, where there is usually no in-house AI team and the person making the decision is also running the business.
Step 1: Choose the right tool for the job
Different AI models and products are good at different things. The best choice is usually the one that fits the systems you already use, not the one with the most headlines. A business running on Microsoft 365 will often get more from Copilot than from a separate tool nobody logs into.
Start with the task, then choose the tool. Read more: ChatGPT, Claude, Gemini or Copilot? Choosing the right AI tool for your business.
Step 2: Balance cost, speed and quality
Not every task needs the most powerful model. Drafting a customer reply and analysing a 60-page contract are different jobs with different price tags. Matching the model to the task, and setting sensible usage limits, keeps spend predictable as adoption grows.
Read more: How much does AI cost a business?
Step 3: Teach the AI about your business
Out of the box, AI knows a great deal about the world and nothing about you. Connecting it to your own documents, price lists, policies and past work is what turns generic answers into useful ones. For most SMEs, this means a well-organised knowledge base, not an expensive custom model.
Read more: Teaching AI about your business.
Step 4: Put guardrails in place
Guardrails are the rules that stop AI saying something wrong, off-brand or risky, particularly anywhere a customer can see it. They range from a one-page staff AI policy to technical controls on what a website assistant is allowed to discuss.
Read more: AI guardrails and governance for SMEs.
Step 5: Keep data safe and compliant
Under UK GDPR, your business remains responsible for personal data your team puts into an AI tool. The right account type, settings and supplier agreements make the difference between a compliant setup and an accidental data breach.
Read more: AI and GDPR: keeping client data safe.
Step 6: Move from assistants to agents
Once the foundations are in place, AI can do far more than draft text. AI agents can carry out whole workflows: reading an enquiry, checking your CRM, drafting a quote and booking a follow-up, with a person approving the steps that matter.
Read more: AI agents for business: automating whole workflows.
Why the order matters
It is tempting to jump straight to Step 6. Agents are where the headline savings are. But an agent built on a tool nobody uses, with no business knowledge, no guardrails and unclear data handling, is exactly the kind of project that gets abandoned.
Working through the steps in order means each stage pays for itself before the next begins. A well-chosen tool with a good knowledge base can save several hours per person per week on its own. That saving funds the automation work that follows.
What the first 90 days look like
For a typical UK business with 10 to 100 staff, a sensible first phase looks like this:
- Weeks 1 to 2: Audit where time goes, which systems you use and where your data sits. Pick two or three processes worth improving.
- Weeks 3 to 6: Choose and configure the tool, build the first knowledge base, write the AI policy and train the team.
- Weeks 7 to 12: Automate the first workflow end to end, measure the result and decide what comes next.
Everything is built inside your own accounts, so you own it, and nothing stops working if you change supplier.
How ThursdayThinking.ai helps
We work with UK businesses at every one of these steps. Most start with our AI Ops Audit, a two-week review that maps where AI will make the biggest difference and what it will cost. From there, our Fractional AI Operator service builds and runs the automations inside your own systems.
If you are not sure where to start, see how we approach AI transformation or email hello@thursdaythinking.digital.