AI automation projects in service operations have a particularity that distinguishes them from AI projects in data analysis: the value is not in the model — it is in integrating the model within a real workflow, with complex business rules, multiple actors and traceability requirements that no model solves on its own.
In this project, the AI component was the most visible but not the most difficult. The assignment engine — with its geographic coverage rules by neighborhood in Bogotá, availability by shift, specialty profiles and maximum workload per analyst — was the component that required the most careful modeling and the longest iterations with the client to accurately reflect how the business operated.
The AI cost architecture was a differentiating element: designing the system to use lightweight models for low-complexity tasks and more powerful models only when the case warrants it, with semantic caching to avoid redundant processing, turned AI from an unpredictable variable cost into a manageable component within the operating model.