Our client is a provider of fleet management software, offering a platform that enables logistics companies, delivery services, and mobility providers to track, maintain, and optimize their vehicle fleets. Their clients rely on the platform daily to manage driver activity, vehicle health, route efficiency, fuel usage, and compliance.
The platform served a wide range of users, each with different needs, access levels, and workflows. The complex interface made it hard for users to find or execute the right tools quickly, leading to high onboarding time for new users unfamiliar with the system structure.
We designed a LAM directly into the client’s fleet management platform, enabling users to perform complex operations through natural language commands. This solution was capable of understanding context, resolving intent, and executing backend workflows.
Software Deployment
Mapping Use Cases and Commands
We began by working with the client’s operations and product teams to identify the most common user actions that could be automated through natural language. We analyzed logs, workflows, and existing user flows to create detailed case mapping.
Training Data and Workflow Structuring
Once key actions were identified, we collected relevant UI metadata, API schemas, and knowledge base content to structure a consistent command-to-action logic. We designed sample prompts and labeled datasets for training the LAM on how users phrase requests.
LAM Integration and Action Engine Development
We implemented the core LAM engine using a combination of LLMs and custom logic layers. The assistant was embedded into the main user interface, designed to be accessible but non-intrusive, with both text and voice options available.
Pilot Launch and Feedback Collection
We rolled out the assistant in a controlled environment with a selected group of users. The assistant was initially limited to non-critical tasks to validate understanding, accuracy, and adoption. The purpose was to gather qualitative feedback and usage data.
Expansion and Real-Time Learning
Following the pilot, we released the LAM to the broader user base and began ongoing monitoring. We analyzed interaction data to refine prompts, add synonyms, and expand intent coverage. We also added suggestions based on user history and context.
Users could type or speak full commands, and the assistant would parse intent, gather needed data, and trigger the right workflows.
The assistant dynamically adapted to the user’s role with respect to existing access controls and audit rules, ensuring compliance.
Users interacted with the assistant directly within the fleet management interface through a persistent, floating chat bubble.
Stay updated with the latest case studies
Your request was sent
