Michelin Connected Fleet is the Michelin Group business that delivers fleet management, telematics, and vehicle tracking to corporate clients across Brazil. Its contact center spans roughly ten customer service teams, a dedicated retention unit, and a telesales operation, supporting fleet operators whose drivers and vehicles depend on continuous service. Two distinct teams within that operation, retention and sales quality monitoring, went live with NiCE Copilot for Agents (Automated Summary) on a pilot basis. The teams configured the tool separately for each context. Retention agents now operate against a real-time read of every conversation, including the sentiment behind it. The quality team scaled its sales call review from roughly 232 interactions a month to approximately 1,200 in two days. Across both teams, 100 percent of customer interactions are now analyzed and summarized automatically.
01 BEFORE
Listening as hard as manual effort allowed
Michelin Connected Fleet is the Michelin Group business that delivers telematics, GPS tracking, and fleet management to corporate operators across Brazil. Its clients run trucks, vans, and equipment whose drivers depend on continuous visibility into where the vehicle is, how it is being used, and whether the hardware on board is working. The contact center supporting that service comprises roughly ten customer service teams, a dedicated retention unit, a telesales operation, and an implementation team responsible for supporting customers with the installation of the hardware and software infrastructure the solution requires. The Customer Care operation is led by Jucielle Ramos do Nascimento, Customer Care Manager, while the telesales operation is led by Elaine Cristina Franquini Pimentel. The entire structure runs on the NiCE CXone platform.Like most operations of its size, Michelin Connected Fleet read its customer conversations the way the industry had always read them: by hand. Two quality analysts could review around 232 telesales interactions per month, working through calls one at a time. Retention agents listened call by call to understand why a fleet client was leaving and what might bring them back. Sentiment was inferred from tone and judgment. Documentation varied agent by agent depending on who picked up the next call. The recordings held more signal than manual review could capture, but at the scale of human effort, that was the ceiling.The retention picture in particular sat with a quiet structural challenge. When a fleet operator called to cancel telematics service, the stated reason was not always the real reason. A driver who had moved to a competitor often said they had sold the truck, because that closes the conversation. Without a scalable way to read what the call actually contained, retention agents worked from instinct and from whatever notes the previous agent had typed during their own divided-attention call. Vitor Augusto Torres Ramos, who runs Customer Service Excellence at Michelin Connected Fleet and led the retention side of the pilot, described the workflow plainly: the team needed to listen to every call, every time, and there was no way around that.
“Before, we had to listen to every call one by one. Now, if I want to summarize the feeling of our last 10,000 customers, I can do it in a few clicks. It is really fast.”
Lilian Beatriz Goncalves da Costa
Telesales Training Coordinator Michelin Connected Fleet
02 DESIRE TO CHANGE
A pilot built around two different listening problems
The mandate took shape inside the contact center itself. The team had access to NiCE CXone and could see what the platform’s AI capabilities offered. Rather than chase a single broad rollout, Michelin Connected Fleet scoped a pilot around two specific problems where listening was the bottleneck. The first sat inside the retention unit, where the prize was understanding why customers were really leaving. The second sat inside the telesales operation, where Lilian Beatriz Gonçalves da Costa, Telesales Training Coordinator, was running quality monitoring and coaching against a sampling rate that could only ever cover a fraction of what her agents were doing on the phone.Vitor and Lilian led the program. The team stood up the configurations and ran the technical enablement on the NiCE side. The brief they set was deliberately tight: stand up NiCE Copilot for Agents with the Automated Summary capability, configure it specifically for the two contexts the team had selected, and let each team prove the value on its own terms before scaling further. The retention configuration would need to capture sentiment and surface root causes the team had been chasing for years through inference. The sales QA configuration would need to produce structured monitoring output the quality analysts could use directly, at a volume human review could not match.The discipline was the design. Michelin Connected Fleet did not assume one configuration would serve both teams, and did not start the pilot until each side knew what good output looked like for its specific work.
03 NiCE SOLUTION
Two configurations, one listening layer
NiCE Copilot for Agents went live in pilot with two distinct configurations. The retention configuration listens to live conversations between fleet customers and Michelin Connected Fleet retention agents, generates an automated post-call summary, captures the customer’s sentiment across the conversation, and structures the result for the retention team to act on. The sales QA configuration runs against the telesales operation’s recorded calls and produces standardized monitoring output that Lilian’s analysts can review at a pace human listening cannot reach. The same NiCE platform sits underneath both.Vitor took the retention side through a deliberate validation step before the team trusted the output. He picked up the line as a customer himself, simulated a deliberately frustrated call, and watched what the system reported back. The summary captured his sentiment correctly and flagged that the agent on the other end had not shown empathy in response. That was the point at which the retention team knew the tool was reading the conversation, not just transcribing it.Retention work at Michelin Connected Fleet runs against a regulatory backdrop that complicates everything around it. Several conversation types require scripts to be read in full, exactly as written, for legal compliance. Vitor described the constraint in plain terms.That regulatory weight was part of why a generic configuration would not work. The retention configuration had to handle the compliance scripts alongside the open conversation, and the team is still tuning the prompt structure to fit the longest scripts inside the model’s context window. The sales QA configuration sits in a different regulatory context entirely and was tuned independently.On the sales side, the team ran the configuration against the same population of telesales interactions they would normally sample by hand. The rollout was kept quiet on the agent floor on purpose. The sales side wanted to see how the calls landed naturally, without anyone adjusting their behavior because they knew a new system was watching. The point was to measure the real work, not the rehearsed version.Across both teams, what the platform takes off the analyst’s plate is not the listening itself. It is the fixed cost of listening. The retention agent still talks to the customer. The QA analyst still reviews calls and runs coaching. What changes is how many calls each one of those people can reach in the time they have.
“We have scripts here that need to be read from beginning to end, exactly as written. If we do not, we have legal problems.”
Vitor Augusto Torres Ramos
Customer Service Excellence Coordinator Michelin Connected Fleet
04 RESULTS
Listening at the scale of every conversation
The headline number arrived from the sales QA side. With the configuration tuned and the platform running against the team’s recorded calls, two quality analysts moved from reviewing roughly 232 interactions in a month of effort to reviewing approximately 1,200 in two days. The output is structured, consistent, and immediately usable in coaching sessions. The sales QA team has been folding the insights directly into telesales training material, and Michelin Connected Fleet expects the additional review depth to carry into sales results across the year ahead.On the retention side, the change is in what the team can now see. Every retention conversation is summarized automatically, with sentiment captured across the call and the customer’s stated reasons for leaving structured for analysis. Patterns that previously hid inside thousands of unread call recordings now surface in the aggregate. The retention team has been mapping the data to identify what is actually driving churn, including a population of departing customers whose stated reason for leaving turned out, on review, to be pricing perception rather than the surface-level reasons they offered to the agent. That intelligence is now visible to the team in a form it had no way to access before.Across both teams, 100 percent of customer interactions are now analyzed and summarized automatically. Sentiment is reported alongside the substance of the call, validated through the deliberately frustrated test interaction Michelin Connected Fleet ran before going live. Documentation that previously varied call by call now arrives in a consistent, reviewable form.The team is still in pilot, and Michelin Connected Fleet is deliberate about what the data already supports. What the operation will say plainly is that a manual ceiling has been removed. The work that two analysts could do in a month is now possible across two days, and the insight that retention agents could only piece together one conversation at a time now exists in a form the team can study at scale.
05 FUTURE
From pilot to production across coaching, procedures, and supervisor tooling
The roadmap that comes out of the pilot has three threads. The first is a structured taxonomy of cancellation and contact reasons. The team learned during the pilot that the model returns more useful output when it is given a defined set of categories to work against. Michelin Connected Fleet is now building that structure in advance of broader rollout, with a working count of around 200 candidate reasons that need to be reduced to a workable set before the next phase.The second is procedure integration. Michelin Connected Fleet maintains a substantial internal procedure library covering roughly 400 processes and 1,000 procedures, including image-based references for hardware troubleshooting. The team is working through how to bring those references into the agent-facing experience so that Copilot can surface the right procedure at the right moment in a live conversation. Image-based content carries its own design requirements, and the work to describe those references in a form the system can match is part of the next planning cycle.The third is broader Copilot deployment, including supervisor-facing tooling and a structured before-and-during performance comparison the team will run as the pilot transitions to production. The sales QA side has its own forward agenda for coaching, with the structured monitoring output already feeding into how training sessions are designed and what telesales managers prioritize.The team’s posture across the pilot has been deliberately phased. Prove value in the two listening contexts where the gap was clearest. Build the structured taxonomy and procedure linkage the next phase will need. Move into supervisor tooling and broader Copilot deployment as the operation, not the pipeline, decides it is ready. The pilot’s discipline is the discipline Michelin Connected Fleet expects to carry forward.
“At Michelin Connected Fleet, we put the customer at the center of every decision. By combining artificial intelligence with the talent of our teams, we are raising our ability to understand needs, anticipate opportunities, and deliver an experience that is increasingly simple, human, and efficient.”
