The sponsor wants another launch campaign. Employees know the AI tool exists, but few keep using it. Before spending more on awareness, find out what happens when someone tries to do a real task.
This case is about choosing the next investigation. The obvious action is more communication. The better action depends on where people leave the journey and what they experience there.
Put every adoption number on the same base
In this fictional company, 1,000 employees are eligible for an internal AI assistant. The same launch cohort produces:
| Stage | Employees | Share of eligible population |
|---|---|---|
| Aware of the tool | 900 | 90% |
| Logged in at least once | 600 | 60% |
| Used it again in week two | 150 | 15% |
Assume the groups are nested: everyone counted as a repeat user is among the original logins, and those logins are among employees aware of the tool.
The repeat-use rate among initial users is 150 ÷ 600 = 25%. Three-quarters of those who tried it did not return in the measured second week.
That is the first investigation I would prioritise. Awareness is already broad; the larger visible drop occurs after trial.
Explain why another campaign is not the first move
You do not need to claim that communication has no value. There are still 100 unaware employees and 300 who know about the tool but have not tried it.
The question is what another campaign buys before the first-use experience improves. If the remaining 100 unaware employees followed the current awareness-to-login and login-to-repeat rates, reaching them would produce roughly:
100 × (600 ÷ 900) × 25% = 17 additional repeat users.
That is a simple benchmark, not a forecast of identical behaviour. It helps frame why fixing the experience among the 600 initial users may deserve attention before expanding the top of the funnel.
Say:
“I would investigate the first-task experience before increasing awareness spend. Only a quarter of initial users return in week two, so bringing more people into the same experience may reproduce the drop-off.”
Quick Prioritisation Drill
A hospital launched a shift-swap app for its 1,000 nurses. Usage is low, and the chief nursing officer wants a second launch campaign.
- Heard of the app
- 950 nurses
- Tried it once
- 700 nurses
- Use it weekly
- 140 nurses
Where should the next investigation focus before any more campaign spending?
Split the journey by work people are trying to do
Ask for repeat use by workflow, alongside task completion, output acceptance and time spent reviewing or rewriting.
Suppose the interviewer provides:
| First workflow tried | Initial users | Week-two repeat users | Repeat rate |
|---|---|---|---|
| Routine summaries | 200 | 120 | 60% |
| Complex client drafts | 400 | 30 | 7.5% |
| Total | 600 | 150 | 25% |
The total hides a large difference. The summary workflow has meaningful repeat use. The client-draft workflow is where most of the abandonment sits.
This changes the intervention. A company-wide reminder treats two different experiences as if they were the same. A workflow-specific investigation can reveal what to fix and what to preserve.
Measure the complete task, including the cleanup
Suppose both workflows previously took 20 minutes per task. The AI-assisted process now looks like this:
| Workflow | Drafting or generation | Review and rework | Total time |
|---|---|---|---|
| Routine summary | 5 minutes | 4 minutes | 9 minutes |
| Complex client draft | 4 minutes | 19 minutes | 23 minutes |
The summary saves 11 minutes. The complex draft takes three minutes longer than the old process, despite producing its first draft quickly.
The relevant experience is time to an acceptable result. A four-minute generation time can look impressive on a product dashboard while making the employee’s overall task worse.
Use repeat activity to diagnose adoption, then measure completed work, quality and how any recovered time is used. That is how you establish business value.
Quick Answer Repair Drill
An insurer gave claims handlers an AI tool for customer letters. An analyst now recommends rolling it out to every letter type.
- Analyst's reasoning
- Both letters drafted in three minutes or less
| Letter type | AI drafting | Checking and fixes | Old process |
|---|---|---|---|
| Payment confirmation | 3 minutes | 2 minutes | 12 minutes |
| Coverage dispute | 2 minutes | 26 minutes | 25 minutes |
What is the main flaw in the analyst's reasoning?
Choose a fix that addresses the observed burden
For complex drafts, examine what employees rewrite. Is context missing? Is the output format wrong? Does the tool make factual errors? Does it attempt an entire document when it would be more useful for a smaller task?
Those causes imply different changes. Adding relevant approved context may help one workflow. A tighter output template may help another. Limiting the tool to an outline or first-pass extraction may be better than asking it to write a complete client-ready draft.
I would choose one recurring failure and test a correction with employees doing that workflow. Keep the task definition and acceptance standard stable so the comparison reflects useful work rather than an easier test.
Give the pilot a clear decision rule
Suppose the proposed correction cuts complex-draft review and rework from 19 minutes to ten, while generation remains four. Total time becomes 14 minutes, saving six minutes against the old 20-minute process.
That is the operating result to verify. Also check whether output acceptance meets the supplied quality standard and whether users return when they next have an eligible task.
Compare the corrected workflow with the current one among comparable users and tasks. Follow assigned work through completion, including abandoned attempts.
If the workflow saves time but users cannot find it at the right moment, the next intervention may be integration or guidance. If it still adds work, more reminders will not fix the underlying experience.
Connect adoption to the business case
Suppose the corrected workflow handles 1,000 tasks a month and saves six minutes each. That returns 100 hours of capacity per month.
Ask what those hours enable: clearing a backlog, completing more client work or reducing paid overtime. Do not automatically label them cash savings if staffing and spending remain unchanged.
Then compare the realised benefit with licences, support, review and implementation costs. The adoption programme should increase useful work, not simply logins.
Give the sponsor a specific next move
“I would preserve the summary workflow and fix the complex-draft experience before expanding the campaign. The biggest drop follows a workflow that currently takes longer after review. I would test one correction against task time, acceptance and repeat use, then promote the workflow once it consistently earns a place in employees’ work.”
Try Enterprise Sales Productivity Decline to practise selecting an investigation from an adoption or productivity symptom. For how to phrase the request, see asking for useful case data.
Prioritisation drill
How would an MBB interviewer judge your prioritisation?
Enterprise Sales Productivity Decline
Choose what to investigate first and explain why. Get feedback on your choice, supporting evidence and next step.
Frequently asked questions
Reading the adoption funnel
Why should every adoption figure use the same base?
A common base lets you compare stages directly. A stage-to-stage rate can look acceptable while hiding a large absolute loss. Expressing every stage as a share of the 1,000 eligible employees shows where the largest number of people leave, so you are not comparing percentages built on different denominators.
Where is the biggest drop-off in the AI tool funnel?
The biggest drop comes between first login and week-two repeat use. 600 employees tried the tool but only 150 came back, a loss of 450. By comparison, 300 aware employees never logged in and 100 remain unaware. The post-login experience is therefore the larger leak.
Is another awareness campaign pointless?
No, but it buys little until first use improves. There are still 100 unaware employees and 300 who know about the tool without trying it. If the unaware 100 followed current conversion rates, only about 17 would become repeat users. Most new triers would meet the same disappointing experience.
Why break repeat use down by workflow?
An overall repeat rate can hide very different experiences. Routine summaries kept 120 of 200 initial users, a 60% repeat rate. Complex client drafts kept only 30 of 400, or 7.5%. Most first-time users tried the weaker workflow, which explains much of the overall drop.
What data should I request besides login and repeat counts?
Ask for task completion, output acceptance and time spent reviewing or rewriting, split by the workflow people attempted. A login shows that someone opened the tool. These measures show whether it produced usable work. High trial with low acceptance points to a product problem rather than an awareness one.
Measuring the real task and fixing it
Why count review and rework time, not just generation time?
Fast generation can disguise a slower overall task. Both workflows previously took 20 minutes. Routine summaries now take nine minutes in total. Complex client drafts take 23 minutes because review and rework absorb 19 of them. Employees experience the whole task, so a tool that adds time will not earn repeat use.
How do I work out what to fix in the complex-draft workflow?
Look at what employees rewrite and why. Each cause implies a different change. Missing context suggests adding approved context. A wrong format suggests a tighter template. Factual errors and over-ambitious whole-document drafts call for other fixes. Diagnose the cause first, otherwise you may fix a problem users do not actually have.
Should the tool still try to write whole client documents?
Not necessarily. If rework comes from the tool attempting an entire document, it may be more useful when limited to an outline or a first-pass task. Narrowing the scope is a legitimate fix when a smaller, reliable contribution saves more time than an ambitious draft that needs heavy rewriting.
What should the pilot have to show before the fix is rolled out?
It should show a time saving, acceptable quality and repeat use together. If rework falls from 19 to ten minutes and generation stays at four, the task takes 14 minutes, six fewer than the old 20. Output acceptance must also meet the quality standard, and users should return for their next eligible task.
Making the business case to the sponsor
Do 100 hours saved a month mean cash savings?
Not automatically. At 1,000 tasks a month and six minutes saved on each, the workflow returns 100 hours of capacity. That becomes cash only if staffing or spending changes, for example through reduced paid overtime. Otherwise, describe it as capacity for clearing a backlog or completing more client work.
What costs should be weighed against the time saved?
Compare the realised benefit with licence, support, review and implementation costs. Review effort deserves particular attention because the rework figures show it can absorb most of the gain. A workflow that saves capacity on paper but needs costly checking may still fail the business case.
What should I recommend to a sponsor who wants another launch campaign?
Recommend preserving the summary workflow and fixing the complex-draft experience before expanding the campaign. Explain that the biggest drop follows a workflow that currently takes longer once review is included. Propose testing one correction against task time, acceptance and repeat use, and promote it once it consistently earns a place in employees' work.
Should the routine summary workflow be changed too?
No, protect it. Routine summaries already cut a 20-minute task to nine minutes and keep 60% of their initial users, so they are the tool's working proof of value. Concentrate change on complex drafts, where the evidence shows the burden, and avoid disrupting what employees already rely on.