Where to start with AI: a practical guide for UK businesses
Most businesses don't struggle to find AI tools. They struggle to know where to start. This guide sets out a practical path that keeps risk low and makes results measurable.
1. Start with a problem, not a tool
The most common mistake is choosing an AI product first and then looking for a use for it. Start instead with a task that costs your team real time or money every week: answering the same customer questions, drafting documents, re-keying data between systems, or searching for information.
2. Pick one workflow
Choose a single, well-understood workflow rather than an organisation-wide programme. A good first candidate is frequent, repetitive, low-risk if a draft is wrong, and easy to check by a person.
3. Measure a baseline
Before changing anything, record how the workflow performs today: time per task, volume per week, error rate or response time. Without a baseline, you can’t prove the AI made a difference.
4. Prove it in 30 days
Run a small pilot with a handful of users. Keep a person reviewing outputs, collect feedback weekly, and compare results with your baseline. If it doesn’t beat the baseline, change the approach or stop. Both are useful outcomes.
5. Set clear rules for data
Decide what information staff may and may not share with AI tools, and provide approved tools so people aren’t tempted to paste sensitive data into public chatbots. In 2023, Samsung restricted generative AI tools after an engineer uploaded internal source code to ChatGPT. Read more in our case studies. Check every tool against UK GDPR, including where data is stored and whether it is used for training.
6. Train people, not just systems
Adoption depends on confidence. Short, hands-on sessions built around your own workflows work far better than generic AI training.
7. Scale what works
Once a pilot beats its baseline, roll it out in phases and pick the next workflow. Over time, these focused wins add up to a genuine AI capability.
- One workflow chosen, with an owner
- Baseline measured
- Data rules written and approved tools provided
- 30-day pilot with human review
- Results compared and a decision made