Workflow of the day ·
Run a fair two-week trial of one AI task
Editor: Owen Hagger
Plan a two-week trial of one AI task with a baseline, a simple daily log and a fair way to judge it, so you decide on your own evidence, not gut feel.
When to use it
When someone in the business says "we should be using AI" and nobody can say whether it actually helps. Most firms try five things at once for a week, get mixed results and give up. Pick one task that happens often, like daily site reports, meeting actions or checking subcontractor applications. Run it properly for two weeks with one or two people, then decide. This prompt plans the trial. It doesn't tell you the answer.
The prompt
You are helping a UK construction director run a fair two-week trial of one AI task in their business. The business is a [SIZE AND TYPE OF BUSINESS, e.g. regional main contractor with around 40 staff]. The task we are trialling: [TASK, e.g. writing the daily site report from the site manager's notes] Who will run the trial: [ROLES, e.g. two site managers on different jobs] How the task is done now: [CURRENT METHOD, IN A FEW LINES] What we already know from our own records about how long it takes or how often it goes wrong: [PASTE YOUR OWN FIGURES, OR WRITE "NONE"] The AI tool we will use, and that it is our company's approved account: [TOOL] Give me: 1. A one-paragraph trial plan: what we are testing, who, for how long, and what a good result would look like. 2. A baseline section: what to measure in the first two or three days without AI, so we have something fair to compare against. 3. A simple daily log as a table with columns: Date, Person, Job, Time taken, Used AI (yes or no), Corrections needed, Notes. 4. Three to five questions to ask the people in the trial at the end of each week. 5. A short decision sheet for day 14 with three options: keep it, change it and trial again, or stop. Say what evidence would point to each. 6. A short list of rules for the trial, including what information must not go into the tool. Rules: - Only use figures I have given you from our own records. Do not invent time savings, percentages or benchmarks. Where a figure is missing, say it needs measuring in the baseline. - Keep it simple enough that a busy site manager will actually fill the log in. - Use UK spelling and plain English.
What it produced
A short trial plan, a baseline checklist for the first few days, a ready-to-copy daily log table, a handful of end-of-week questions for the people doing the work, and a one-page decision sheet with keep, change or stop and the evidence behind each. Where you had no figures of your own, it marks them as things to measure rather than filling them in.
What you still check
- The task is one that happens often enough to give you two weeks of real examples.
- The baseline is measured the same way as the trial, by the same people, on similar work.
- Nothing confidential goes into a personal AI account. Use your company's approved account and follow your own data rules.
- The day 14 decision is based on the log and the feedback, not on the one example that went really well or really badly.
Steps
- Choose one task and one or two people, and run the prompt with what you already know.
- Measure the baseline for two or three days, then use AI for the rest of the fortnight and fill the log in daily.
- Sit down on day 14 with the log and the feedback, and make the keep, change or stop decision.
Go further
- Every workflow for this role: AI for Construction Directors
- More ideas like this: daily ideas for Construction Directors