Customer-service teams made AI part of their work and kept using it after we left.
A building-supplies distributor needed fewer order corrections and more consistent quote follow-up. Its AI assistant had made little difference to frontline work. We helped teams choose useful applications, test them on the job, and stop workflows that did not earn continued use.
weekly use two months after the engagement
Why it matters
The investment began supporting everyday customer work, and teams continued using it after the engagement ended. Quote follow-up improved and correction callbacks declined alongside changes to team workflows, giving the business evidence of value beyond license purchases and training attendance.
The starting point
Access to the tool had not made it useful to frontline teams.
Seven months after the company licensed an AI assistant for everyone, weekly active use stood at 57% in corporate functions, 8% in customer service, and 3% among warehouse supervisors. An inaccurate AI response had damaged trust in customer service.
Regional operations managers promoted the tool even though they had not used it in 30 days. Teams needed useful workflows, a meaningful say in what continued, and leaders who demonstrated the practices they expected from others.
What the business needed
Build useful work practices that people would continue using beyond training and the engagement itself.
What we changed
We started with how people worked and gave teams a meaningful role in deciding what would continue.
Choose workflows from the work people needed to do
A six-week diagnostic included warehouse-floor shadowing and ride-alongs. Skeptics helped design applications around observed problems, including drafting customer replies from order history. Behavior at week 20 became the main measure of adoption.
Give teams a meaningful decision about what continued
Every workflow had a six-week team decision point, supported by weekly implementation reviews. The organization committed to no quota increases tied to saved time during the year, giving people room to test whether the work helped.
Make leaders practice what they asked of others
Regional managers edited AI-drafted summaries live in Monday operations reviews. Customer-service representatives reviewed and changed drafts before sending replies, keeping judgment with the person doing the work.
What the client could use next
- Workflows shaped by frontline teams
- Applications built around observed customer-service tasks, including draft-from-order-history, with representatives editing replies before sending them.
- A way to improve or stop each pilot
- Weekly implementation reviews and six-week team decisions about whether a workflow deserved to continue.
- Measures that followed behavior beyond training
- A week-20 check on use in daily work and a week-32 follow-up, alongside live editing of drafts in leadership reviews.
What the evidence showed
More quotes received follow-up within 48 hours.
Share of quotes receiving follow-up within 48 hours
- At baseline
- 39%
- Reported result
- 81%
AI-assisted workflows, changes to team practices, and management follow-through shared credit for this result. The figures do not isolate AI’s contribution.
Customer-service work changed in observable ways
At week 20, 66% of service representatives used draft-from-order-history. Fifty-eight percent of outbound replies began as drafts that representatives changed before sending. Callbacks to correct order details fell 24% during the wider change effort.
A rejected pilot stopped, and demand for other work grew
Teams voted down the handoff-notes pilot 13 to four, and it stopped publicly. Nine teams requested the second wave, giving the rollout a source of demand beyond the original mandate.
Leadership practice changed; supervisor use remained weak
All six regional managers used live-edited drafts in Monday reviews. Warehouse-supervisor use rose from 3% to 9%, leaving a clear gap that the customer-service results did not resolve.
A useful result still has limits.
The baseline figures describe different roles, and the week-20 figure measures one specific workflow. They are not a single before-and-after adoption rate. The 63% follow-up does not establish organization-wide adoption; the stopped pilot and weak supervisor uptake remain part of the account.
Use continued after the engagement ended
Weekly use stood at 63% at week 32, two months after the twenty-four-week engagement. That follow-up checked whether the practices continued beyond delivery, rather than counting licenses or training attendance.
Our teams helped decide what deserved to continue. We stopped the pilot that wasn’t useful and kept the workflows that improved daily work. People were still using them after the engagement ended.