The short answer.

The first three things to build with AI come from one team’s list and pass four tests: the work happens every week, its steps can be written down in order, one named person already checks it, and it needs no more from your business systems than a file someone already exports. Among the pieces of work that pass, choose three that start from the same records and go to the same checker, and build them in order. The first runs in daily use before the second is built.

Choosing the work is part of planning it, the first of three steps: plan it, develop the workflows, deploy them. Which team should go first with AI? picks the team. Choose your first AI wins scores the long list across the company. This piece takes one team’s list and turns it into a short sequence of three.

Why three, and why from one team.

Three pieces of work from one team give that team a pattern instead of a one-off. One workflow on its own depends on the person who built it and is easy to treat as an experiment. Three in a row, sharing the same records and the same checker, become the way the team works, and the third is built faster than the first because the inputs and the review are already in place.

More than three spreads the checker thin. The person who reviews the output also reads every miss while a workflow is tested on real cases, and that attention is the scarcest thing in the plan.

Four tests for each piece of work.

Put every candidate piece of the team’s work through the same four tests, with the people who do the work in the room.

  1. How often does the work happen?Weekly or more passes. Monthly passes if it is heavy each time. A piece of work that comes round once a quarter waits, because nobody remembers the misses from last time.
  2. How well are the steps known?Ask someone other than the usual person to say the steps in order. Written down already passes. Known by the whole team passes. Held by one person waits until that person’s method is captured; When the know-how sits with one person covers how.
  3. Who checks it?One person, by name, who reads the work before it leaves the team today. That person stays the decision point: a workflow prepares, a person decides. Work that goes out unchecked today waits until someone is named.
  4. What does it need from your business systems?The next section sets out four levels. For the first three, the answer is a file or an export at most.

Work that fails a test stays on the plan, with the test it failed and who will clear it.

What the work needs from your business systems.

Every piece of work sits at one of four levels of what it needs from your business systems, and the first three things you build sit at the first two.

  • Nothing.The inputs are notes, photos, emails or documents a person already has in hand and gives to the assistant.
  • A file someone already exports.A report from your accounting system, a list from your project system, a spreadsheet the team keeps up to date.
  • A live read.The workflow pulls records from a system as it runs.
  • A write back.The workflow puts data back into a system, with a person approving each change.

Live reads and write backs belong in the plan from the start and come after the first three, which is where the team learns how its workflows behave. How AI workflows connect to your business systems covers the later levels.

Choose three that fit together, in order.

The three pieces of work that go first share at least one input or one checker, and each one is built only once the one before it is in daily use. Choose them in this order.

  1. First: the surest.The piece of work that passes all four tests most clearly, with written steps and inputs at the level of nothing or a file. It proves the team’s way of working with a workflow.
  2. Second: the one that reuses the first.Work that starts from the same records or goes to the same checker. The checker already knows how to read a prepared draft, and the inputs are already gathered.
  3. Third: the one that stretches.Work that is heavier, or that needs an export the team has not used yet. By the third, the team has two workflows behind it and room to take on more.

Writing the steps down for each one follows the same five headings; How to hand a piece of work to AI sets them out.

A shortlist sheet to copy.

Copy these headings, one row per piece of work, for every candidate on the team’s list. Write “Unknown” where nobody knows, with who will find out.

  1. Piece of work.Named for what it produces, such as “weekly client update”.
  2. How often.Weekly or more, monthly, or less.
  3. Steps.Written down, known by the team, or held by one person (name them).
  4. Checker.One name, or “nobody yet”.
  5. From your systems.Nothing, a file, a live read, or a write back.
  6. Starts from.The records or documents the work begins with.
  7. Shares with.Other pieces of work on the sheet with the same inputs or checker.
  8. The decision.First, second, third, or later with the test it failed and who clears it.

The rule for the sheet: a piece of work is one of the three only if it passes all four tests. Rows six and seven decide the order among those that pass.

A civil engineering firm chooses its three.

Take a 45-person civil engineering firm whose project delivery team had chosen to go first. Its head filled in the sheet with two project managers and a project engineer.

Illustrative example.

  • Site visit reports.Weekly on every active project, written from field notes and photos, to the firm’s report format, checked by the project engineer. Needs nothing from the firm’s systems. The decision: first.
  • Weekly client status updates.Weekly, steps known by both project managers, checked by the project manager on each job. Starts from the project list export and draws on that week’s site visit reports. The decision: second, because it reuses the first.
  • Contractor question log and draft replies.Several a week during construction, steps known by the team, every reply checked by the engineer responsible for the design. Needs an export from the document system the team has not used before. The decision: third.
  • Permit application packages.Heavy and regular, but the steps sit with one senior engineer. The decision: later, once that engineer’s method is written down.
  • Drainage design narratives.The calculations run in the design software and the engineer of record signs off. The decision: later, once the firm decides how the narrative draws on the calculation output.
  • Fee proposals.Every proposal differs and the principal writes each one. The decision: later.

The three share a thread: the site visit report feeds the client update, and the contractor questions arise on the same projects. The site visit report starts from a free workflow template in the Workflows Directory, adapted to the firm’s own report format and tested on recent visits before any client sees a report.

What is different in a company of under 200 people.

In a company of under 200 people, the checker on all three pieces of work is often the same person, and that person also does the work. Plan around that person’s week first.

  • The list is short and known.The team lead can name every repeating piece of work from memory. The whole sheet fits one conversation.
  • Know-how sits with fewer people.More work fails the second test. That is a reason to capture it, and FullyPowered Studio captures existing know-how as workflows.

Write your three down where the rest of the plan lives. FullyPowered Planner is where you set out your teams, their roles and the work between them and choose what to build first, and your colleagues add their own teams. When you want the three built with you, the 3-Day Workflow Sprint develops the first. How organizations plan their use of AI covers the rest of the method.