How to adopt AI across a business
By the Day Seven team · Updated 14 October 2026
Adopt AI in seven steps: check readiness, prioritise a few high-value tasks, run a short pilot, put light governance in place, train the people who will use it, measure the result against a baseline, then scale what worked. Most stalled programmes skip the middle steps — they buy licences and run an awareness session, but never pick specific tasks, set rules or measure.
Summary
- Start from tasks people repeat every week, not from a tool.
- Pilot with a small group and a fixed end date before buying at scale.
- Agree data rules and human checks before the pilot starts, not after.
- Train on the real tasks in the pilot, then measure time, quality and sustained use.
1. Check readiness
Readiness is less about technology than about clarity. Before choosing tools, answer four questions: which approved AI tools people can use today, what data may and may not go into them, who owns the decision to expand use, and where staff already use AI informally.
Informal use is common and is usually the best evidence of where value lies. Ask about it without blame — a short anonymous survey works.
- Approved tools and the licence or account type (consumer, business or enterprise).
- Data classification: what is public, internal, confidential or personal.
- An owner for AI decisions, even if part-time.
- Existing informal use and the tasks it covers.
2. Prioritise a few use cases
List candidate tasks and score each on value, frequency, risk and ease. A good first use case is repeated weekly by many people, has a clear quality bar, and uses data you are allowed to process.
| Criterion | Question to ask | Good first candidate |
|---|---|---|
| Frequency | How often is the task done, and by how many people? | Weekly or more, by most of the team |
| Value | What does a faster or better result unlock? | Time freed for client or revenue work |
| Risk | What happens if the output is wrong? | Errors are caught by an existing review step |
| Data | Does the task need personal or confidential data? | No, or only within an approved enterprise tool |
| Ease | Can people do it in tools they already have? | Yes, with no integration work |
3. Run a short, bounded pilot
Choose one team, two or three tasks and a fixed period — four to eight weeks is typical for a first pilot. Record a baseline before anyone is trained, and decide in advance what result would justify expanding.
A hackathon or short build sprint can compress the discovery part: a cross-functional group spends a few days building working prototypes against real tasks, which gives you something concrete to pilot.
4. Put light governance in place
Governance at this stage should match the risk of what you are doing and how complex your organisation is: a short policy may be enough for low-risk use, while regulated or higher-risk uses need more depth. Either way it needs to say which tools are approved, what data is off limits, when a person must check or sign off output, and how to report a problem.
If personal data is involved, UK GDPR still applies; the ICO's guidance on AI and data protection and on data protection impact assessments is the starting point. Organisations wanting a formal management system can look at ISO/IEC 42001.
5. Train on the real tasks
Generic awareness sessions rarely change behaviour. Train the pilot group on the exact tasks you selected, in the tools they are allowed to use, with the data rules built into the exercises. Leaders need a different session: how to set expectations, judge output and manage risk.
6. Measure against the baseline
Measure three things: time per task including checking, the share of the group still using the approach weeks later, and quality or rework. Report capacity released separately from cash saved — they are not the same thing.
7. Scale what worked
Expand only the use cases that met the success criteria. Scaling usually means more licences, a refreshed policy, training for the next groups and named champions in each team. Re-run the measurement for each new group rather than assuming the pilot result transfers.
An illustrative example
This is a hypothetical scenario, not a Day Seven client result.
A 40-person professional services team finds that people already draft client emails and meeting summaries with personal ChatGPT accounts. The firm approves one business-tier tool, bans client-identifiable data in prompts until a DPIA is complete, and pilots meeting summaries and first-draft proposals with eight people for six weeks. Baseline times are recorded in week one; training covers the two tasks only. At week six the firm compares time including checking, rework and continued use, and decides whether to expand.
Frequently asked questions
Sources
Related
Want help applying this to your organisation?
Training is delivered on-site at your premises or live online, scoped after a short conversation. Individuals can join the in-person Leeds AI Fundamentals Workshop.