Data Lake in retail: how 120 stores stopped searching for information

When you have 120 stores, the problem isn't a lack of data. It's the opposite: you have too much, in too many places, and nobody can find what they need when they need it.

This is the story of an Italian retail chain that turned information chaos into a competitive advantage — without changing any of the systems they were already using.

The company

• Sector: B2C retail, clothing and accessories
• Stores: 120 across Italy
• Employees: 850+
• Systems: Cegid POS, Shopify e-commerce, Salesforce CRM, SAP warehouse, Microsoft 365 email

The "before": five systems that don't talk to each other

Every Monday morning, the commercial director received the same question from the board: "How was the weekend?"

To answer, he needed to:

1. Extract sales from Cegid POS (120 stores, individual export per location)
2. Add e-commerce data from Shopify
3. Cross-reference with Salesforce CRM for the loyalty program
4. Check SAP stock availability for best-selling products
5. Read zone managers' emails for qualitative feedback

Time: 3-4 hours every Monday. And the report reached the board after lunch, when decisions for the week had already been made.

But the Monday report was just the surface. Underneath, there were daily problems:

• Store managers didn't know which products were running low in their store (the data was in SAP, but they only had POS access)
• Marketing couldn't figure out which campaigns worked in-store (CRM data vs. POS data not connected)
• Logistics replenished based on static rules, not actual demand
• Zone managers wrote weekly reports via email that nobody ever aggregated

Company know-how — the best sellers' intuitions, seasonal patterns, local preferences — was all in people's heads or buried in emails and Excel spreadsheets.

What we did

Phase 1: Audit (2 days)

We spent two days at headquarters and in 3 sample stores. We observed how information was searched for, requested, and manually reconstructed.

Audit results:

• 14 critical information flows identified
• Average time to get a cross-system answer: 18 minutes
• 5 main data sources not connected to each other
• Estimated 420 hours/month of manual search and data aggregation work across the organization

Phase 2: Unified Data Lake + Aitaky Brain (6 weeks)

We installed Aitaky Brain in the headquarters' server room and connected through our integration solutions:

• Cegid POS → real-time sales from all 120 stores
• Shopify → e-commerce orders, abandoned carts, traffic
• Salesforce CRM → loyalty customers, purchase history, campaigns
• SAP → warehouse stock, supplier orders, lead times
• Microsoft 365 → zone managers' emails, shared documents, qualitative reports

All read-only. No modifications to existing systems. Data never leaves the corporate network.

On this foundation, we implemented:

• RAG indexing (Retrieval-Augmented Generation) across all documents and emails
• Local on-premise LLM for natural language queries
• Automatic dashboards generated nightly with aggregated KPIs
• Proactive alerts for anomalies (sales drops, stock-outs, target deviations)

Phase 3: Progressive rollout

• Weeks 1-2: Board and commercial director
• Weeks 3-4: Zone managers (8 people)
• Weeks 5-6: Store managers (120 people, tablet access in-store)

The "after": the numbers

The surprises we didn't expect

1. Store managers started asking questions

They didn't before because they knew the answer would take days. Now they ask: "What's the highest-margin product in my women's category?", "How much of this item did other stores in the zone sell?". Access to information created curiosity.

2. Zone managers' reports became useful

Before, they were narrative emails that ended up archived. Now the data lake indexes them and RAG makes them searchable. The board can ask: "What are zone managers saying about the new collection launch?" and get a summary in 15 seconds.

3. Marketing discovered invisible patterns

By cross-referencing POS + CRM + e-commerce data, they discovered that 23% of customers who bought online picked up in-store and then purchased a second item. No single system showed this pattern.

The quote

"Before we had data. Now we have answers. The difference seems subtle, but when you manage 120 stores it's enormous. Monday morning I no longer waste time reconstructing the weekend — I spend it deciding what to do in the week ahead."

— Commercial Director

The lesson learned

With 120 stores and 5 different systems, the real cost isn't technology. It's the time people spend doing the work technology should be doing: searching, copying, aggregating, formatting.

The data lake didn't replace any system. It connected them. And the on-premise RAG made all that information wealth — including emails, reports, notes — queryable in natural language, without a single byte leaving the company.

Want to find out how much time your organization wastes searching for information? Let's talk →

Related articles:

• What it actually costs to run AI inside your company: three builds with prices and real throughput
• AI in manufacturing: from half a day to 30 seconds for the weekly report