This plan is designed for organisations that have decided to do something with AI but are not sure where to start. It assumes you have a team, a budget, and a willingness to be honest about what needs to change.
Before you start
Establish these foundations before choosing a tool or sharing data.
Apply it to your work
Three phases with a review before expanding
- Weeks 1–4 · assess
Map the process, evaluate the opportunity and choose one goal.
- Weeks 5–8 · pilot
Prepare, configure, test with a small group and iterate.
- Weeks 9–12 · review
Measure against the baseline, check governance and plan the next step.
Name an owner
One person who is responsible for this initiative. Not a committee. Not everyone. One person who will make decisions, remove blockers, and be accountable for outcomes.
Set one goal
One process. One outcome. One way to measure success. Not three goals. Not five. One. The more focused you are, the more likely you are to succeed.
Set the boundaries
Decide what data may be used, which actions require human approval, and how errors will be detected and handled. Check relevant security, privacy and contractual requirements before sharing information with a provider. Use synthetic, anonymised or otherwise approved test data as appropriate. Make sure anonymisation is effective for the intended use.
Agree a stop condition, a rollback plan and who is responsible for ongoing review. Consequential decisions need safeguards proportionate to their impact. A calendar plan does not replace that assessment.
Weeks 1-4: assess and choose
Week 1: map the process
Take the process you want to improve and map it in detail. Every step. Every handoff. Every decision point. Every place where data is entered, transformed, or used.
Do not map how the process is supposed to work. Map how it actually works, including the workarounds.
Week 2: identify the AI opportunity
Look at the process map. Where is there repetitive manual work? Where are decisions made based on pattern recognition? Where is data being transformed from one format to another?
These may be opportunities for improvement. Compare AI with simpler process or automation changes, then choose a narrow opportunity with a useful potential benefit and manageable risk.
Week 3: evaluate tools
Research tools that address your specific opportunity. Not best AI tools lists. Tools that solve your specific problem. Evaluate them against your workflow, budget and approved test data, within the boundaries set before the project began.
Week 4: decide and plan
Choose a tool. Define the implementation plan. Set a budget. Identify the people who need to be involved. Set specific, measurable success criteria.
Weeks 5-8: build and integrate
Week 5: prepare the data
Clean the data the AI will use. Establish the data quality standards it needs. Build or configure the connections between your systems.
Data preparation matters, alongside task fit, tool capability, configuration and the safeguards around its use.
Week 6: configure and test
Set up the AI tool. Configure it for your specific use case. Run it against approved historical or representative test data and compare its outputs to known-good results.
If the outputs are not reliable, review the data, task design, configuration and model or tool fit. Do not proceed to a live pilot until the agreed safeguards and acceptance criteria are met.
Week 7: pilot with a small group
Roll out to a small team. Five to ten people who understand the process and can provide honest feedback. Give them training and support. Collect feedback daily.
Week 8: iterate based on feedback
Fix what is broken. Adjust what is awkward. Remove what is unnecessary. The pilot group's feedback is more valuable than any planning document.
Weeks 9-12: measure and scale
Week 9: measure against baseline
Compare the AI-assisted process to the baseline you established in week one. Is it faster? More accurate? Cheaper? By how much?
If the improvement is not measurable, understand why before proceeding.
Week 10: train the wider team
Expand beyond the pilot group. Train each team with hands-on practice, not just presentation slides. Provide ongoing support for the first two weeks after training.
Week 11: check the governance in practice
Review the ownership, safeguards and monitoring established before the pilot. Check whether they are working in day-to-day use, update the documentation and make sure issues have a clear route to resolution.
Week 12: review and plan next steps
Assess the overall outcome. Document what worked and what did not. Decide whether to expand this AI capability or start a new one.
Common pitfalls
- Skipping the data preparation because it feels slow
- Choosing a tool before defining the problem
- Training once and assuming adoption will follow
- Measuring activity instead of outcomes
- Moving to the next AI project before the first one is stable
The 90-day principle
Ninety days is a useful constraint because it forces focus. You cannot do everything in 90 days, so you have to choose the thing that matters most.
A focused starting point makes it easier to measure the result honestly and decide whether to build further.
From the Yopla source archive dated 2026-02-15. Refreshed for this edition.
All insights