Case study · Retail

An optical retail chain that predicts demand at each store with AI

The data from every store in a single brain. Machine learning models that anticipate what will sell and where. Stock and assortment optimized store by store.

Assortment and stock optimized

Demand forecasting with machine learning

Network data centralized

Where it all started

Dozens of stores, dozens of data islands

An optical retail chain with a wide network of stores was selling well, but each point of sale operated as an island. Sales, stock and customer data lived separately in each store, with no picture of the whole.

Without that central view, stock and assortment decisions were made by intuition or by average. The same assortment for very different stores, stockouts (OOS) in some and excess immobilized inventory (capital tied up) in others.

And without centralized, organized data, applying AI to anticipate demand was simply impossible.

The challenge

Sales, stock and customer data trapped in each store, with no single view of the network.

Assortment and stock decided by intuition or by average, the same for stores with very different demand.

Stockouts (OOS) in some stores and immobilized overstock in others.

Reactive replenishment, with no anticipation of demand by store and SKU.

Without centralized data, AI and machine learning were out of reach.

What we proposed

First, centralize. Synchronize data from every store into a modern data platform on Azure, creating a single source of truth for the whole network. On top of that, machine learning models that predict demand by store and by SKU, capturing seasonality and the behavior of each point of sale, and optimize stock and assortment store by store. And a Power BI dashboard so the business can steer it.

The shift

From replenishing blind to anticipating demand store by store

Stores stop being islands. Data from the whole network converges into a single brain, and AI turns that brain into a forecast of what will sell, where and when.

Before

Sales and stock data isolated in each store

Assortment and stock decided by intuition or by average

Stockouts in some stores, immobilized excess in others

Reactive replenishment, with no anticipation

Now

A single source of truth with the whole network synchronized

ML models that predict demand by store and SKU

Assortment and stock optimized for each point of sale

Replenishment anticipating what each store is about to sell

stores synchronized

+ 1

single source of truth

forecast by store and SKU

The value generated

What changes when each store stops deciding blind

A single view of the network

Every store, synchronized into a single source of truth.

Demand anticipated by store and SKU

ML models predict what will sell at each point of sale before it happens.

Fewer stockouts, less excess

Stock adjusts to the real demand of each store, with less OOS and less capital tied up.

Assortment tailored to each store

An assortment based on store profile, not the same for every store.

Business in charge

A Power BI dashboard puts the forecast in the hands of whoever decides.

Ready for more AI

With the data centralized, the next AI use cases (recommendation, personalization) are already the next step.

What the client says

Before, each store was its own world and we replenished by eye. Now AI tells us what is going to sell at each point of sale and we adjust the assortment accordingly. It is a different way of running the network.

Retail Operations Management

Optical retail chain / Retail sector

F&Q

Frequently asked questions

What does it mean to centralize data across a store network?

It means synchronizing sales, stock and customer information from every point of sale, today scattered store by store, into a single platform. At BertIA, an AI and data consultancy based in Barcelona, we build it as the foundation for applying AI across the whole retail network.

How does AI predict demand for each store?

Machine learning models learn from the sales history of each point of sale (seasonality, location, buying behavior) to anticipate which SKUs will sell and in what quantity, and the forecast is refined as new data arrives. Each store gets its own forecast, by SKU.

Does this work for any store chain?

Yes. The approach, centralizing the data and predicting demand by point of sale, is replicable in any retail network with multiple stores, inside and outside the optical sector.

Do the store systems need to change?

No. The platform connects to each store’s existing systems and synchronizes their data without replacing anything or interrupting the operation of the point of sale.

What can be done next with this foundation?

With the data from the whole network centralized, more AI use cases open up: product recommendation, offer personalization or detection of emerging trends. Demand forecasting is the first step, not the last.