Employees won't use your Copilot? 7 tactics to move adoption from 20% to 70%
You deploy an AI Copilot. Three months in: 19% weekly active users. The problem is rarely the model itself. What kills adoption is that nobody made it their job to get people actually using it. We moved adoption from 20% to 70% on seven projects by embedding it into existing workflows, developing power users, and systematically removing friction points.
By
Tenten AI FDE 團隊
前線部署工程
Published
May 26, 2026
Read time
6 分鐘
Last month I visited an insurance company. They had launched an AI Copilot six months earlier for their sales team. The tool could look up policy language, draft customer scripts, and organize visit notes. On launch day, 3,000+ people got accounts. Three months later, I checked the backend: 19% weekly active users.
The tool worked. The problem was nobody had ownership of getting people to actually use it.
Adoption is not a training problem. It is an engineering problem. Think of it like tuning a production line: find where people get stuck, fix it, measure the change, and iterate. We used seven tactics to move projects stuck at 20% adoption up to 60 to 70% weekly active users. Most of the impact comes from details that have nothing to do with the model itself.
Map your friction points
Figure out where your people are actually stuck. We divide the user journey into four stages, ask one diagnostic question at each stage, and pick the right tactic for each. A misdiagnosis wastes everything that comes after.
| Where they get stuck | What it looks like | What to do |
|---|---|---|
| Forget it exists | Low weekly active, only opens when reminded | Embed in existing workflows, manager demo |
| Open it, don't know what to do | High login count, low task completion | Killer use cases, template library |
| Try once and quit | Churn within 7 days of first use | Remove friction, power-user pairing |
| Use it but don't trust it | Only searches, won't share with customers | Feedback loops, source attribution |
Match your situation to the table above, then apply the corresponding tactic.
Tactic 1: embed it in the windows they already have open
The biggest friction is one more place to look. One more tab, one more login, one more URL to remember, each adds friction. Put the Copilot where they already spend their time: in the CRM sidebar, in the customer service console, in Teams or Slack. The insurance company moved their policy-lookup feature from a standalone website into the CRM sidebar. Usage tripled within two weeks. People did not change how they worked. The path got shorter.
Tactic 2: grow 5 to 8 power users first, don't rush company-wide launch
Company-wide launch is a mistake. Give 3,000 people access simultaneously and nobody owns making it work. Pick 5 to 8 power users first, typically people burdened by repetitive work. Spend two weeks with them, document their actual usage patterns, and turn it into internal case studies. Once they can say in a team meeting 'I saved two hours on that one task,' expand to the rest. That is how you build advocates: through visible results, not announcements.
Tactic 3: perfect the first three killer use cases, leave the rest for later
Generic demos that try to do everything usually do nothing well. Sit with the team and pick three high-frequency, high-pain tasks. Refine them with real data until they work immediately after launch. Instead of handing them a blank chat window, give three 'click once, get answer' entry points. Keep the use cases narrow and tied to actual daily work, because specificity protects quality.
Tactic 4: cut every unnecessary step
Friction kills adoption quietly. Single sign-on cuts the need to re-login. Pre-populate common prompts. Make frequent actions one-click. Avoid making people type if you can. Grab a stopwatch: if a task takes more than three clicks from 'I need to use this' to 'I have my answer,' cut something. Fewer clicks and faster decision-making boost adoption.
Tactic 5: build a template library to turn good prompts into repeatable assets
The prompts that power users develop need to become repeatable. Capture them in a template library: ready-to-use scripts, report frameworks, standard openers for common situations. New people do not learn how to talk to AI from scratch. They click and go. This cuts churn from the 'opened it but didn't know what to do' stage significantly, because you have replaced a blank page with an actual starting point.
Tactic 6: measure behavior, not logins
Login counts do not tell you much. Track weekly active users (WAU), task completion rate, and 7-day retention from first use. More importantly, talk to people who are not using it every week. Their reasons for not using it reveal more than praise from active users. One conversation revealed a critical issue: 'I cannot paste what it gives me to a customer, I am afraid it is wrong.' This led us to add source citations for every sentence. Trust and adoption recovered together.
Tactic 7: make it part of the workflow, not optional add-on
When using the Copilot is optional (treated as 'extra credit if you have time'), it stays at the bottom of the to-do list. The turning point comes when managers use it in meetings and it becomes part of the delivery workflow. For instance: 'visit notes are drafted by Copilot, then edited by hand.' Once using it becomes part of the job itself, not a side task, adoption stabilizes at 70% instead of declining month to month.
The insurance company reached 68% weekly active users after about ten weeks, up from 19%.
At Tenten, we practice FDE (front-line deployment engineering). The engineer does not hand off the system and disappear. If adoption falls short of targets, the work is not finished. A beautiful demo is not success. People using it every day is.
One stuck workflow
is enough to begin
Tell us what the team does today, where it breaks down, and what a better working day should look like.