Artificial intelligence moves faster than most teams can keep up with. Every week brings new headlines, new capabilities, and new promises. But for the fleet professionals who actually have to use these tools day in and day out, the question isn't "what can AI do?", it's "can I trust it with the decisions that matter?"
Behind every credible answer to that question is a team that has spent months asking hard questions, testing assumptions, and refusing to ship anything until it genuinely helps. That's the story behind the MICHELIN AI Assistant, and to tell it, we sat down with one of the people closest to the product.
Yanita Dimitrova, Product Manager responsible for GenAI at MICHELIN Connected Fleet, leads the integration of artificial intelligence into the platform that fleet operators and managers, transport coordinators, and operations teams rely on every day. We spoke with her about why the Assistant exists in the first place, the principles it was built on, and the realities of making AI genuinely reliable outside the lab.
Let's start at the beginning. Why did the team decide to build the MICHELIN AI Assistant in the first place?
Yanita Dimitrova: We decided to build the MICHELIN AI Assistant to address a clear gap for our customers: getting to insights and outcomes from fleet data was often too slow and required too much manual effort. Users had to navigate complex data, interpret results, and perform repetitive tasks before they could act. The MICHELIN AI Assistant solves this by making the experience more intuitive and efficient; helping users quickly find answers, surface relevant insights, and automate routine work. Ultimately, it reduces friction, speeds up decision-making, and allows clients to focus on higher-value activities.
What were the non-negotiables when you started designing it?
Yanita Dimitrova: A few principles shaped everything that followed. First, the MICHELIN AI Assistant needed to solve meaningful user problems, not just be a feature for its own sake. Second, accuracy and trust: responses had to be reliable, transparent, and grounded in the right data so users could confidently act on them. Third, seamless integration into the workflow: we designed it to fit naturally into how users already work, reducing friction rather than adding another layer of complexity. And finally, scalability and continuous learning, so it could improve over time and adapt as user needs evolve.
What does the MICHELIN AI Assistant actually help users do, in concrete terms?
Yanita Dimitrova: In day-to-day work, the MICHELIN AI Assistant helps users get answers and take action much faster. It can quickly surface insights from data, answer ad hoc questions, and guide users to the right information without them needing to dig through multiple tools or reports. It also takes on more routine tasks, like summarizing information, highlighting key trends, or helping structure analysis, so users spend less time on manual effort. Overall, it streamlines workflows, reduces friction, and frees users up to focus on higher-value decisions and outcomes.
There's a lot of anxiety around AI replacing people. How did you think about that?
Yanita Dimitrova: The MICHELIN AI Assistant is designed to support all MyConnectedFleet users, not replace them: fleet managers, transport coordinators, and operations teams bring expertise, context, and accountability no system can replicate. Our assistant keeps users in control by being transparent about how outputs are generated and allowing them to validate, refine, or build on the results. Rather than automating decisions, it supports better ones, acting as a guide and accelerator within the workflow, not a substitute for human judgment.
"Trust" gets used a lot in AI conversations. What does it actually mean in your product?
Yanita Dimitrova: We ensure users stay in control by designing the MICHELIN AI Assistant to be assistive, transparent, and editable rather than autonomous. Users can review and refine outputs before taking action, and the assistant clearly shows how responses are generated so there's no "black box" decision-making.
It also operates within the defined boundaries of the MyConnectedFleet platform and data permissions, so it only works with existing permissions. Overall, the design principle is simple: the assistant accelerates the work, but the user always makes the final decision.
How does that translate into how customer data is actually handled?
Yanita Dimitrova: We ensure customer data is protected by building the MICHELIN AI Assistant on a security-first and privacy-by-design foundation. All data access is governed by existing user permissions, so the AI assistant only works with information the user is already authorized to see.
We also apply strict controls around data handling, ensuring inputs and outputs are processed securely and not exposed outside of the approved environment. In addition, we minimize data retention, avoid unnecessary storage, and continuously monitor for compliance with our security standards. This ensures the assistant delivers value while maintaining the same level of trust, governance, and protection customers expect from the platform.
What's been the hardest part of making the assistant reliable outside controlled conditions?
Yanita Dimitrova: The most challenging part was ensuring consistent reliability in real-world, messy usage conditions, where queries are often ambiguous, data can be complex, and expectations are high. In controlled environments, the MICHELIN AI Assistant performs well, but in practice users ask very open-ended questions, switch context frequently, and rely on nuanced business data.
Making sure the AI assistant could handle that variability while still producing accurate, grounded, and relevant responses was the hardest part. A lot of effort went into improving how it interprets intent, retrieves the right context, and stays aligned with trusted data sources, so that even under imperfect inputs it remains dependable and useful.
How was that learning loop set up?
Yanita Dimitrova: The MICHELIN AI Assistant was built entirely in-house by our dedicated data science and engineering teams. We started from zero, with a clearly defined problem to solve.
The learning loop was launched with intensive internal testing of the first prototype, opening it up to the wider organization so colleagues could share immediate feedback. Not long after, we took the leap into private beta with a hand-picked group of just over 20 organizations. These early users were more than testers — they were key partners in the AI assistant initiative.
The beta expanded progressively over the following eight months, and by March 2026, we had onboarded users from over 4,800 customer organizations. Each onboarding wave brought more user feedback, new use cases, and new edge cases we addressed — making the assistant smarter and more reliable with every step. The assistant gets better the more it's used. That's not a side effect. It's by design.
When you cut through all of it, what does the assistant actually change for the users using it?
Yanita Dimitrova: Concretely, the MICHELIN AI Assistant changes how quickly and easily customers can go from question to action. Instead of spending time searching through dashboards, running manual analyses, or stitching together information from multiple places, they can now get relevant answers and insights in a single step.
It reduces the time and effort required to understand data and make decisions. Practically, this means faster reporting cycles, quicker identification of trends or issues, and less dependency on specialized expertise for everyday analysis. It shifts the experience from "finding information" to "acting on it."
The MICHELIN AI Assistant is one piece of a much larger commitment. At MICHELIN Connected Fleet, AI isn't a standalone solution; it's increasingly woven into the tools and workflows fleet professionals rely on every day, from driver behavior monitoring through to platform-level analytics. The ambition is consistent across all of it: integrate AI in a way that simplifies, accelerates, and supports, without adding complexity or new burdens for the users relying on it.
And the work is far from finished. New capabilities are in development, more data sources are constantly being made available. Over time, the assistant will transform from surfacing information to helping users think through decisions; not making those decisions, but providing proactive, data-driven, and reliable support that makes better decisions easier to reach.
The question the team keeps coming back to, iteration after iteration, is a simple one: does this actually help our customers? As long as the answer is yes, the work continues. If you'd like to be part of this journey for our fleet management solutions, reach out to our sales team today.