AI enablement: a practical first 90 days for your company

Companies adopt AI in two ways. Some buy licences and hope; some pick a few workflows, set rules, run a pilot and measure it. The second route is slower for the first month and faster after that. Here is a 90-day plan for the second route.

Days 1 to 30: understand and choose

• Interview the people who do the work, not only the people who manage it. Ask where time goes, which tasks repeat, and where errors are costly.
• List 10 to 20 candidate use cases. Score each on value, how likely it is to work, whether the data is available, and the risk if the AI is wrong.
• Pick one first use case. The best first choice is valuable, low-risk, and small enough to finish in 6 to 8 weeks.
• Write down a baseline: how long the task takes today, how often it goes wrong, how much it costs. Without a baseline you cannot show an improvement.

Days 15 to 45: set the rules

Write a short acceptable-use policy while the pilot is being planned. It should cover which tools are approved, what data may be entered into them, when a human must review the output, and who is accountable. Keep it to two pages that people will read.

Check which of your planned uses fall under the EU AI Act and what that requires. Most business uses are low-risk, but a few categories, such as screening candidates, carry specific obligations.

Days 30 to 75: run the pilot

• Build or configure the use case with a small group of real users, ideally five to fifteen.
• Create a test set of real examples before the pilot starts, and score the system on it. This is the evidence for the go-live decision.
• Hold a weekly review with users: what worked, what failed, what they stopped using. Fix the top issue each week.
• Train the pilot group on the habits that matter: check the output against the source, know what the tool is bad at, and know when to escalate.

Days 75 to 90: decide and plan the next step

• Compare the pilot's results to the baseline, using time saved, quality and cost, plus the users' own verdict.
• Decide: roll out, adjust, or stop. Stopping is a legitimate result and is cheaper than a failed rollout.
• If you roll out, name an owner for the process, set up monitoring, and schedule the next use case from your list.

Common ways this goes wrong

• Starting with the most ambitious use case instead of the most finishable one.
• No baseline, so success is a matter of opinion.
• Rules written by legal alone and ignored by everyone else.
• Training run once as a presentation instead of practised on real tasks.
• No owner after the pilot, so the tool decays.

If you want help with any part of this, from the assessment to the pilot, that is the work we do under AI Enablement.