How Grupo Casas Bahia Automated Customer Feedback Analysis 14x Faster with Databricks
Grupo Casas Bahia is one of Brazil’s largest omnichannel retailers, serving over 100 million customers through more than 1,000 stores and a national logistics network. The company deployed Databricks Agent Bricks with Meta’s Llama 3.3 70B model to automate the classification of customer reviews from six distinct channels. Monthly review classification jumped from 1,500 to 33,500, model accuracy reached 90%, and the company saves over 4,000 person-hours annually — equivalent to nearly R$480,000.
Impact
14x
Productivity gain in comment analysis
33,500
Monthly reviews automatically classified
90%
Model accuracy in detecting complaint journeys
4,000+
Annual person-hours saved
91
Distinct problem types tracked
Challenge
Manual classification of customer reviews across six feedback channels could only process 100 comments per hour, forcing analysts to rely on small samples while thousands of daily reviews went unanalyzed, leaving departments without timely insight into customer pain points.
Solution
Grupo Casas Bahia deployed Databricks Agent Bricks with Meta’s Llama 3.3 70B model via Databricks AI Functions, using Unity Catalog for multi-source data governance to automatically classify 33,500 reviews per month across 91 problem types.
Tools & Technologies
What Leaders Say
“Manual classification was a bottleneck: it took about an hour just to process 100 comments, so we often relied on small samples, leaving gaps in our insights.”
“By automating comment analysis with Agent Bricks, we’ve achieved a 14X productivity gain and an efficiency improvement of more than 9 hours for every 1,000 comments. The model saves more than 4,000 person-hours per year — equivalent to nearly R$480,000.”
“Databricks is always by our side, accelerating our analysis, making governance transparent, and helping us adapt solutions for maximum business impact.”
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.
Full Story
In Brazilian retail, where consumer protection agencies like Procon and platforms like Reclame Aqui give customers a loud public voice, the ability to act on feedback quickly is a competitive necessity. Grupo Casas Bahia receives reviews across six channels — NPS surveys, CES scores, Reclame Aqui, app stores, Procon complaints, and Consumidor.gov — and each channel speaks a different language in terms of format, tone, and urgency. Understanding what’s actually breaking for customers required reading all of it.
Before automation, that reading was done manually. One analyst could process roughly 100 comments per hour. At that rate, the team worked with small samples, leaving most of the feedback unanalyzed. Trends were slow to surface, root causes were guessed rather than measured, and the departments that needed to act — product, UX, logistics, operations — worked from incomplete pictures. Manual classification was not just slow; it was a structural ceiling on customer intelligence.
Grupo Casas Bahia built the solution inside their existing Databricks environment. Unity Catalog provided governance for consolidating feedback data from all six sources into a secure lakehouse. A custom prompt for Meta’s Llama 3.3 70B model, accessed via Databricks AI Functions, handled end-to-end comment classification — mapping each review to one of 91 problem types, a customer journey, and a root cause group. The architecture was deliberately collaborative: data engineers, scientists, analysts, and UX specialists worked together in one platform, enabling rapid benchmarking of different models before settling on the production approach.
The throughput shift was dramatic. Monthly classified reviews grew from 1,500 to 33,500 — a 22x increase in coverage. The model achieves 90% accuracy in identifying the customer journey behind each complaint, processing what previously took an analyst over 14 hours in under one hour. Over 4,000 person-hours are saved annually, estimated at nearly R$480,000. More importantly, corporate dashboards now give every department — from product managers to executives — a real-time, 360-degree view of where the customer experience breaks down, enabling faster prioritization and measurably better outcomes.
The infrastructure also enables proactive strategy. Budget planning, UX improvements, and operational initiatives now start from data rather than instinct. For a retailer serving over 100 million customers, the ability to hear every voice and act on the patterns matters at scale.
