The short answer.

A skeptical team comes along with AI when the ask stops being vague. Pick the team’s own most tedious piece of repeating work, let the team write down what good output for that work looks like, run the workflow on cases the team has already finished so they can compare it with their own answers, and make the team’s lead the owner who decides when it goes live. The skeptics become the people who set the standard, and a team trusts a standard it wrote.

Bringing a team along is part of deploying a plan, the last of three steps: plan it, develop the workflows, deploy them. How do we get people to actually use AI at work? covers a willing team that has not changed its habits. This piece is for the team that has heard the pitch and does not believe it.

Skepticism is usually about a vague ask.

Most team skepticism about AI is a reasonable answer to a vague request. “Use AI more” and “find ways AI can help” leave the team to guess which work, to what standard, and who carries the blame when the output is wrong.

A specific ask answers the three things a skeptical team wants to know:

  • Which piece of work.One named piece of the team’s week, not “your job”.
  • Who judges the output.The team, against a standard the team wrote.
  • Who decides.A person on the team checks and approves every output. The workflow prepares; a person decides.

Treat the skeptics as your best reviewers. The person who doubts a workflow most will find its misses fastest, and a workflow tested by its harshest critic is ready for everyone else.

Start from the team’s most tedious repeating work.

The first workflow for a skeptical team is the piece of work the team itself names as the most tedious and repetitive. Ask the team, not the leadership. Leadership tends to pick the work that matters most to the company; the team picks the work it would hand over tomorrow, and a team is far more generous with a tool that takes away work it dislikes.

Good first candidates share three traits:

  1. It repeats.The team does it the same way every week, many times over.
  2. Someone already checks it.A senior person on the team reviews the result today, so the decision point is already in place.
  3. It is low on judgment and high on assembly.Gathering, copying, matching and formatting, with the judgment calls left to the person.

Let the team write what good looks like.

The team, not the person building the workflow, writes the standard the output must meet. Ask the most experienced people on the team what they check when they review the work today, and write each answer down as a rule the output passes or fails. “Every figure traces to the source document” is a rule. “High quality” is not.

Write the skeptics’ objections into the standard. When someone says “it will get the special cases wrong”, ask for three special cases and add them to the test. Each objection becomes a case the workflow must handle before anyone relies on it, and the person who raised it sees their concern in the test itself.

Show the workflow on the team’s own past cases.

A skeptical team believes what it sees on its own work. Take ten or so cases the team finished recently, with the answers the team gave, run the workflow on the same inputs, and put both answers side by side. The team marks each output against its own standard: pass, fail, or pass with a correction.

  • Use finished cases.The team already knows the right answer, so nobody has to trust the workflow to judge it.
  • Include the hard ones.The special cases from the skeptics belong in the set.
  • Log every miss.Each fail goes on a list with what the right answer was, and the workflow is changed until it passes.

Build the test set that proves an AI workflow is ready sets out how to assemble those cases.

Make the team’s lead the owner.

The team’s lead owns the workflow: the lead keeps it current, reads the list of misses and decides when it goes live and when it changes. The lead also has the final say on going live, which is the strongest reassurance a skeptical team can get. Who should own AI in a company your size? separates the owner of the plan from the owner of each workflow.

Questions to ask a skeptical team.

Copy these into one page and work through them with the team in one meeting. Write the answers in the team’s own words.

  1. Which piece of your week would you hand over first?One piece of repeating work, named.
  2. What does it start from?The documents, records or messages the work begins with.
  3. What do you check when you review it today?Each answer becomes a pass-or-fail rule.
  4. Where would AI go wrong?Each answer becomes a test case.
  5. Which recent cases can we test on?Ten or so finished cases, including the hard ones.
  6. Who approves each output?One person, by name.
  7. What would make you say no to going live?The lead’s go-live condition, written down.

A furniture maker’s estimating and customer service teams.

Take a 100-person furniture maker that builds custom pieces to order. Its estimating team and its customer service team were both skeptical. Each had been told to “try AI” and neither had found a use for it.

Illustrative example.

  • Estimating named its work.The estimators chose turning a customer’s drawings and spec sheet into a materials list priced against the current price list. The senior estimator doubted the workflow would read dimensions and finishes correctly, so the team wrote that rule first: every line on the list traces to a line on the spec, and anything outside the standard range of sizes, woods or finishes is flagged for a person, never priced.
  • Estimating tested on its own quotes.The team picked recent finished quotes, including the awkward ones the senior estimator named, and compared the workflow’s materials lists with its own. Every miss went on the list, and the workflow changed until it met the team’s rule. The estimating lead owns the workflow and approves every quote before it goes out.
  • Customer service named its work.The team chose answering customers who ask when their order will arrive. The workflow reads the order record and the production schedule and prepares a reply for an agent to check and send. The team’s rule: the date in the reply matches the schedule, and any order running late goes to a person to write.
  • Customer service tested on past replies.Recent questions and the team’s real answers went side by side with the drafts. The customer service lead owns the workflow and set the go-live condition.

Both teams started from workflow templates in the Workflows Directory, adapted to the company’s own price list, schedule and wording, and running in the company’s own AI assistant.

What is different in a company of under 200 people.

In a company of under 200 people, you bring a skeptical team along in the room with them, not through a change programme. A large company runs a change campaign; you sit down with the team and its lead and work through the questions in one meeting.

  • The skeptic is often the expert.In a small team, the person who doubts AI most is often the one whose know-how the work depends on. Put that person in charge of the standard.
  • Word travels fast.One team’s workflow that holds up on its own past cases is the best argument the next team hears.
  • The lead does the work too.The lead who owns the workflow often runs it, so the lead sees every miss first-hand.

Put each team’s first workflow and its owner on the plan. 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 rest of the method.