Beyond Dashboards: How Grupo Bimbo Delivers Insights with Databricks AgentBricks

Esteban Ramirez, Mauricio Imenez17 Jun 2026

Problem

Grupo Bimbo has more than 1,700 sales centers and 54,000 delivery routes. The team computed indicators of missed sales opportunities (skipped client visits, wrong prices, sales recorded far from the client's location) and put them on dashboards. Usage was low and nobody acted on them. Leaders did not know where to start, and supervisors spend one day a week in the office and the rest in the field, so they rarely open BI tools.

Example from the article: a route with 5 extremely risky tickets and 95 normal ones can have the same average risk as a route with 50 medium-risk and 50 clean tickets, so a plain average sends supervisors to the wrong route.

Solution in brief

Detect anomalies in context, roll them up into one risk score per route, and deliver the result where each role works. One Isolation Forest per sales center, pre-sales type and channel gives context-specific thresholds. Supervisors get a Genie natural-language chat on the phone, and executives get a weekly personalized report written by an LLM through ai_query.

Questions this case answers
  • How to detect contextual anomalies in visit records across many segments with one model per context, and roll them up into one risk score per unit that does not hide extreme cases
  • How to deliver analytic findings to staff who work in the field and do not open BI tools, through a phone chat and an LLM-written weekly report
  • How to detect contextual anomalies in field-sales route visits and rank routes by risk
  • How to get field supervisors to act on analytics when they do not use dashboards

Tools and methods

  • Isolation Forest, hundreds of models — one per context
  • Weighted route risk score — anomalous and normal tickets weighted separately
  • Databricks AI/BI Genie — natural-language questions over a curated dataset
  • ai_query — SQL-callable LLM function (Claude, per the session abstract)
  • Unity Catalog, Lakeflow Jobs, medallion layers — governance, orchestration, data preparation
Continue reading — it's free

Google or a one-time email code. No password.