AI Without the Hype · Part 8 of 10
If you tell people a tool can do their work faster, they may reasonably wonder what you plan to do with the time it saves.
A cheerful presentation about innovation will not answer that question. Your employees need to understand the purpose of the change, what you are asking them to try, and how you will judge the result.
My preference is to start with work the team wants help getting through. Ask what consumes their time, what gets in the way of serving customers, and which tasks could use a better first draft.
Then be honest about what the business intends to do.
State the purpose in terms people can assess
“We’re adopting AI” gives staff an announcement. “We’re testing whether draft job summaries can reduce the office backlog while keeping you responsible for approving them” gives them something concrete to evaluate.
Say who is participating, what will change, how long the test runs, and what remains under human control. Explain what measures you will collect and how they will be used.
If your intent is to free time for overdue customer calls, say so. If workforce changes are genuinely under consideration, do not promise they are impossible. Trust depends on the promises you can keep.
Ask the experienced employee where the demonstration breaks
Imagine a dispatcher who has spent years handling schedule changes. A vendor shows an assistant that proposes a tidy daily schedule. The dispatcher immediately notices that one technician cannot do that job, another needs different equipment, and the travel time ignores a recurring road closure.
That reaction is valuable input. Ask which constraints the demonstration missed and how a correct schedule would handle them.
The employee’s job knowledge helps you decide what belongs in the system, what needs current information, and what should remain a person’s decision. Make room for that knowledge before buying a tool around an oversimplified version of the work.
Give people time to learn on the clock
Telling staff to experiment between everything else you already expect can turn the trial into extra work.
Set aside practice time. Use approved or fictional examples. Show the team how to give a useful brief, check source information, recognize missing details, and report a result that should not be used.
One employee may need help starting a task. Another may need help knowing when to stop trusting a polished answer. Training should cover both.
Save examples of acceptable outputs and important failures. They make better teaching material than a collection of clever prompts without any context.
Make reporting problems part of the job
If every mistake is treated as resistance or incompetence, people have a reason to hide problems. That leaves you with an inaccurate picture of the tool.
During a trial, ask staff to record material corrections, failed attempts, and work they chose to complete manually. Make it clear that choosing the manual route can be appropriate when the output is uncertain or the task is outside scope.
The person who catches a false statement before it reaches a customer has performed useful quality control. Include that effort when assessing the result.
Avoid judging adoption by prompt counts or logins. Those measures can show activity without showing whether the work improved.
Decide what happens to recovered capacity
Suppose an office assistant saves three hours a week after review and corrections. Agree on what those hours should support: faster billing, fewer overdue callbacks, better records, or a backlog that has been waiting for months.
Ask the employee whether the expected capacity actually appeared. It may have been consumed by training, interruptions, or maintenance. You need that information before adding more work.
OECD survey research on SMEs examines AI use, worker preparation, and changes to skill and staffing needs. It is evidence from surveyed businesses across seven countries, not a forecast for your team. Your own decisions should come from your work, your staff, and the results of the trial.
Keep people capable of judging the work
When a tool drafts a summary or suggests a response, the employee still needs to understand what makes it correct. Keep source documents accessible and make the manual process available when the tool fails.
Have staff explain why they accepted or changed an output during early training. That tells you more than asking whether they liked it.
You are building a working practice your team can understand and maintain. Their ability to question the tool is part of that practice.
The conversation to have first
Ask: “What part of your week would you most like to make easier, and what would we need to protect if we changed it?”
Listen to both halves. The frustrating task may be a useful starting point. The answer about what needs protecting may determine whether the project works at all.
Source
OECD: Generative AI and the SME Workforce. The dispatcher and office-assistant scenarios are illustrative; recommendations reflect the approach described in this article.