The AI tool you paid for is not being used because it was introduced as software rather than as a fix to one specific job, and because the first week was left entirely to the person who was meant to adopt it.
Why did staff stop using the AI tool after the first week?
Nobody decides to stop. It just never becomes part of the work.
- A tool is bought because it seemed useful to the owner, in a meeting, in a demo.
- Everyone is told to "try it", with no task attached and no deadline.
- Two or three curious people open it. The rest never do, and that is fine, because nobody is counting.
- The curious ones hit a wrong answer, have nobody to ask, and quietly go back to the way they already know.
- Two months later the subscription is still running and the belief is that "the staff will not use it".
The failure is not at the last step. It is at step two, where no job was ever given to anyone.
How do you tell whether the tool is the problem or the introduction was?
Ask one question at a time and write the answers down. Answers that point at the person mean the introduction failed; answers that point at the output mean the tool failed.
| What you hear | What it usually means |
|---|---|
| "I don't have time to learn it" | No task was ever assigned, so it competes with real work |
| "It gets things wrong" | Nobody was told what to do when it is wrong |
| "I don't trust it" | The outputs are never checked by anyone senior |
| "I did not know it could do that" | The use cases were never explained for their job |
| "We tried that already" | The trial had no success defined and no deadline |
| "Nobody else is using it" | The first users were given no status and no thanks |
| "It needs better internet" | A real but solvable problem, usually not the main one |
| Silence | Nobody was ever asked, so nobody was ever asked again |
The cheapest diagnostic: sit with three users for fifteen minutes each and watch them attempt one real task. That beats any survey.
How do you pick the first use case?
Not the most impressive one. The most repetitive one, done by someone with time to give it a fair go. Score three or four candidates out of five on each line, then pick the highest total that is also low-risk:
| Criterion | The question behind it |
|---|---|
| Frequency | How often does this happen, across the team? |
| Tedium | Would staff happily not do this again? |
| Clarity of input | Is there a consistent starting point, or is it judgement? |
| Error tolerance | Can a wrong draft be reviewed in seconds, or does it go to a customer? |
| Data readiness | Does the person already have the material to hand, usable? |
| Visible win | Would the saving be obvious to the person doing the work? |
Two rules decide most cases. If a wrong output would reach a customer or a safety-critical machine, do not start there. If the intended user has no quiet ten minutes in a week, it will not happen either.
A good first target looks boring: writing up quotes, summarising site notes, drafting follow-ups, turning voice notes into a handover summary.
What does a successful first week with a new tool look like?
One person, one task, fifteen minutes a day, and a named person available to answer questions within minutes. That is the whole programme.
- Day one: they do the task the old way, timed honestly.
- Day two: same task with the tool, timed as well.
- Day three: same task, output checked against the old way.
- Day four and five: same task, no help at all.
- Friday: compare the two timings and ask one question — would you do it this way again next week?
If the answer is yes, you have a result written up by a staff member, not by management. If it is no, you learned something in a week that would otherwise take a quarter, and nothing was lost. Then add the second person.
How do you measure adoption without asking anyone?
Three numbers, none of which require a survey.
Weekly adoption = staff who used the tool in the last seven days ÷ staff who have access × 100. If the denominator grows and the numerator does not, you are adding access, not adoption.
Cost per actual user = monthly subscription ÷ weekly active users. This is what makes the "should we cancel it" conversation honest.
Time returned = (minutes saved per use × uses per person per week × people using it) ÷ 60. Count time returned, not time spent: the first two weeks are slower for almost everyone, so a tool that saves time from week three needs three weeks.
What do you do about staff who think it will replace them?
Do not argue with them, because they are partly right. Separate the two things they are worried about, then answer each.
- "Will my job go?" Answer the real question: what will my work look like in six months? People are often relieved to hear their job is changing upward rather than ending.
- "Will I be the last one able to do the old way?" Their experience is the specification. The person who knows why the process has three odd steps is the one who can say which step the tool will get wrong.
- "Will I look stupid?" Never let anyone be the guinea pig. Have the owner and the most senior person use it on their own work first, in the open, including the failures.
How do you make it work on mobile data, load shedding and a shared phone?
Assume the infrastructure will be interrupted, because it will be. Design for it now.
- Start with tasks that need little data: text, short summaries, structured notes. Not files, images or long documents.
- Put a one-step instruction on paper at the workstation or in the cab, so someone can start without loading an app.
- Agree which staff do what, so the tool is not opened on a metered connection by someone who does not need it that day.
- Treat WhatsApp as the primary channel, not the fallback. If your staff already live there, that is where the tool has to meet them.
- Test peak hours on the worst phone in the building, not the newest.
What do you do in the next 30 days?
- Name one task, one person, one week. Not a tool rollout — a task.
- Watch that person try it for fifteen minutes. Write down every point where they hesitated.
- Time the old way and the new way honestly, setup time included.
- Remove one step from the task if the tool does not need it. Most abandoned tools were asked to do too much.
- Ask the user, not management, whether they would do it again next week.
- Only then add a second person, and only while the first is still using it unasked.
If you would rather find out which processes are quietly still being done by hand than keep paying for an unused tool, book a free strategy session and we will go through it with you.