AI went live but nobody uses it. Move adoption from 20% to 70%

Three months after an AI assistant launched, we reviewed the data: 240 employees, 43 active weekly users. Adoption was below 20%. The CEO asked directly: "Everyone knew how to use this at launch. Where did they go?"

They didn't leave. The tool never became part of their actual work.

We've encountered this pattern often enough to understand what happens: adoption doesn't stall because the model is weak. It stalls because you haven't identified why employees don't use it. It's not one reason. Four different barriers coexist. Isolate each one and it has a fix and a way to measure. The eight practices below come from field work organized by these four barrier types. Handle one barrier and adoption reaches 40-60%. Handle all four and it reaches 70%. Sequence matters. Build trust first. Without it, the other three just spin.

Barrier 1: Trust, if they don't believe the answer, they won't risk their job on it

Employees aren't lazy. They're worried. An AI provides an answer that sounds reasonable but is actually wrong. The employee is held responsible, not the model. Until trust exists, nothing functions.

Practice 1: Make every answer traceable to its source. Your RAG system should tag each response with links to original documents and exact passages. Users can verify in seconds and know where errors occurred. Measurement: track source-link click-through rates. In a functional system, 30-40% of users click through to verify, showing they're actively using it and checking their work.

Practice 2: Show failures, not just successes. Post regularly in your internal channel: "Three questions our AI got wrong this week" with corrections. Transparency makes people willing to use it because they know the boundaries. Measurement: run a monthly survey asking "would you hand off an AI answer directly to a client or stakeholder?" Track the percentage who say yes.

Barrier 2: Workflow fit, if they have to open another app, it won't get used

This barrier is widely underestimated. A tool can be strong, but if it sits outside the systems employees use every day, it's destined to fade. People don't reshape their workflows for AI. It must be embedded where they already work.

Practice 3: Embed into existing tools, not create new entry points. Sales work in the CRM? Build the Copilot into the sidebar. Support works in the ticket system? Generate suggested replies directly on the ticket page. The principle: users shouldn't have to remember to use the AI. It appears when they need to make a decision.

Practice 4: Set default triggers instead of waiting for manual requests. When someone opens a contract, automatically surface a summary and flag risks. Don't wait for them to ask. Measurement: track two adoption patterns separately. One is active use (they asked for help). The other is triggered adoption (the AI appeared automatically and they used it). Growth in triggered adoption is where real adoption gains come from.

Barrier 3: Incentives, if using it doesn't benefit them or threatens them, it won't get used

Consider a senior employee whose value comes from being the only person who understands a particular process. AI democratizes that knowledge. It threatens their job security. They won't say it openly, but they'll quietly suppress the tool. If incentives aren't aligned, no tool survives.

Practice 5: Tie adoption to managers' goals, not individual contributors' goals. The most effective lever is making team adoption a metric that managers own rather than mandating it top-down. Pressure from a direct manager works better than pressure from executives.

Practice 6: Make early adopters visible by showcasing the time they've saved. Every month, calculate the hours AI freed up for each person and mention the top users by name in meetings. Measurement: track seed-user adoption. When 20% of a team uses the tool heavily, the rest usually cross the adoption threshold within three months.

Barrier 4: Training, it's not that they won't use it, they don't know what to use it for

Most people think of AI as a chatting tool. They don't know it can rewrite reports, search across ten documents, or draft responses. The ceiling on imagination is the ceiling on adoption.

Practice 7: Teach scenarios, not features. Don't run a system training session. Run a "here are the three things that slow you down most about your job, watch us solve them with AI" session. Measurement: track retention two weeks after training, not attendance on day one.

Practice 8: Station someone on-site, not a FAQ document. For the first month after launch, put an engineer in the customer's office. When someone hits a problem, they get immediate help. This is the most expensive move and the most effective. It simultaneously solves trust, workflow fit, and training barriers.

Eight practices and their impact

Here's how adoption changed at several manufacturing and financial services customers after implementing this playbook (weekly active users / total headcount):

Handle one barrier alone and adoption reaches 40-60%. Handle all four together and it reaches 70%. Sequence matters. Build trust first. Without it, the other three just spin.

We insist on stationing engineers on-site rather than handing over a system and leaving because adoption isn't something you design. It's something you build over weeks and months in the customer's environment. A polished demo doesn't matter. What matters is whether 70% of your people use it every day three months after launch.

Frequently asked questions

Q: How long until adoption changes show? Trust and training improvements usually appear within a month. Incentive changes take a quarter because they require organizational behavior to shift.

Q: Do you absolutely need someone on-site? Not necessarily. But that first month of on-site support is the highest-return investment you can make. It solves three barriers at once.