The short answer.

AI did not stick in your company because the first push gave people an assistant and stopped there: no piece of work was named, nobody owned that work, and nothing was built that the next person could pick up. The restart is a plan, not another round of licences or training. Find the few people still using AI and what they use it for, turn the best of that into one team’s shared workflow with one owner by name, and set the date you will look at it again.

A restart is the first of three steps, done properly this time: plan it, develop the workflows, deploy them. Why isn’t AI saving us time? explains where the time in a business sits. This piece is for the company that already tried once and watched the effort fade.

Why a first AI push fades.

A first AI push fades for three reasons, and most companies hit all three at once.

  1. No named work.The push said “use AI” and left each person to decide where. Each person tried it on whatever was in front of them that day, got a mixed result and went back to the old way. Nobody wrote down “AI prepares the weekly schedule” or “AI drafts the estimate scope”.
  2. No owner.When nobody owns a piece of AI-assisted work, nobody fixes it when it misses. The first poor output ends the experiment, because there is no one whose job is to adjust the instructions and try again.
  3. Nothing built to be reused.Every good prompt lived in one person’s chat history. When that person got busy, moved roles or left, the good result left with them, and the next person started from a blank box.

None of the three is about the assistant. ChatGPT, Claude, Copilot and Gemini all do the work well when the work is named, owned and written down. The push faded because it was undirected, and an undirected effort runs on enthusiasm, which runs out.

Experimenting was the right start.

Experimenting was the right way to start, and the first push was not wasted. Your people learned what an assistant is good at and where it stumbles. Some of them found real uses. Some leaders now know which of their team’s work could move onto AI and which could not.

What experimenting cannot do is spread. A good discovery stays with the person who made it until someone decides it belongs to the team. That decision is what the restart adds: the move from trying AI to running it on purpose, on named work, in a focused order.

Start the restart from what survived.

The best starting point for a restart is the AI use that survived the first push. If a few people still open an assistant every week for the same job, that job has already passed a real test: it was useful enough to keep doing after the excitement ended.

Ask each team lead one question: who on your team still uses AI, and for what, every week? Write the answers down word for word. A short list is normal. Three people doing the same job with an assistant every week is a better start than thirty people who tried it once.

A use that survived points at the work, the person who knows how it is done and the inputs it starts from. Those are the three things the first push never wrote down.

How to restart with a plan.

A directed restart takes five steps, in this order.

  1. Collect what survived.Each team lead lists who still uses AI, for which job, how often.
  2. Choose one team and one piece of work.Pick the surviving use that repeats most and starts from records someone can export. One team, one piece of work.
  3. Name the owner and the reviewer.One person keeps the workflow current. One person checks every output before it leaves the team, as someone checks that work today. A person decides; the workflow prepares.
  4. Write it down as a shared workflow.Turn the person’s private prompt into written steps, inputs and checks that anyone in the role runs the same way, in your own assistant on your own account.
  5. Put it on the plan with a review date.Write the team, the workflow, the owner and the date you will look at it. Add the next piece of work for that team underneath.

How to write an AI plan in one sitting turns step five into a single meeting with your team leads.

A restart sheet to copy.

Copy these headings onto one page. Fill in one sheet per team. Write “Nobody” or “Unknown” where that is the answer.

  1. Team and lead.The team, named for its work, and its lead.
  2. What the first push gave this team.Accounts, a training session, a list of ideas.
  3. Who still uses AI here, and for what.Names and jobs, word for word.
  4. Which of those jobs repeats every week or month.
  5. Where its inputs live.The system, file or inbox it starts from.
  6. The first shared workflow.One piece of work.
  7. Owner.One name.
  8. Reviewer.One name.
  9. Where the written workflow will live.One shared place, one current version.
  10. Review date.The day you look at it again, and what you will look at.

A landscaping company restarts.

Take a 55-person landscaping company with four teams: maintenance crews, installation crews, estimating and the office, which handles scheduling and billing. Earlier in the year the owner bought assistant accounts for everyone in the office and estimating and ran a lunchtime session. By late summer the owner rarely heard AI mentioned.

Illustrative example.

  • What survived.The owner asked each lead the one question. Two estimators still used an assistant every week to turn site-visit notes and photos into the scope section of an estimate. The office manager used it now and then to reword a late-payment email. Nobody on the crews used it.
  • The first team and the work.Estimating, and the scope section of every estimate. It repeats every week through the season and starts from the site-visit notes the estimators already keep.
  • Owner and reviewer.The senior estimator owns the workflow. The estimating lead reads every scope before it goes to a customer, as the estimating lead does today.
  • Written down.The two estimators’ prompts became one written workflow: the inputs, the headings a scope always carries, the plant and material terms the company uses, and what to flag for the lead. The company started from a free workflow template in the Workflows Directory and adapted it to its own estimates.
  • On the plan.Next for estimating: the follow-up note after an estimate is sent. Next for the office: the late-payment email, owned by the office manager. The crews come later, starting with the end-of-visit summary a maintenance customer receives.

The owner did not relaunch AI to the whole company. The owner relaunched one piece of work with one owner, and wrote the next three on the plan.

What is different in a company of under 200 people.

In a company of under 200 people, you find out what survived the first push by asking, this week. A large company surveys usage across departments. You can ask every team lead in person and have the full list in a few conversations.

  • The people who still use AI are easy to find.Often it is two or three people, and you already know who they are.
  • One person often holds the know-how.The estimator who built the good prompt is also the person who does the work. Writing it down protects the company when that person is on holiday or moves on.
  • A restart is visible.When one team runs a shared workflow every week, the other teams see it on their desks. That does more than another training session.

When you want help building the first workflow, we develop it with you in a 3-Day Workflow Sprint, in your own assistant, run by your own people.

Restart with a plan this time. FullyPowered Planner is where the plan lives: you set out your teams, their roles and the work between them, choose what to build first, and your colleagues add their own teams. How organizations plan their use of AI covers the whole method.