AI inventory forecasting: 50% to 90% accuracy
How Super-Pharm used AI to improve inventory accuracy from 50% to 90% and make demand forecasting 10x more efficient across 290 stores.
Quick answer
Super-Pharm, Israel's largest pharmacy chain with 290 branches, used AI to improve inventory prediction accuracy from 50% to 90% and make demand forecasting 10 times more efficient.
Most AI case studies focus on customer-facing tools: chatbots, email automation, voice agents. But some of the biggest ROI comes from behind-the-scenes operational AI that customers never see.
Super-Pharm is Israel's leading pharmacy and beauty retailer. 290 branches, 80 stores in Poland, and an ecommerce website where online sales recently overtook brick-and-mortar revenue. Their click-and-collect service, where customers buy online and pick up in-store, was growing fast.
The problem: inventory predictions were wrong half the time.
When customers can't find what they ordered
Click-and-collect only works if the product is actually at the local store when the customer arrives. If someone orders vitamins online for pickup at their nearest branch and the shelf is empty, that's a failed experience.
Super-Pharm has tens of thousands of products across hundreds of locations. Predicting which products will sell, in what quantities, and at which branch is a massive data problem. Their on-premise tools couldn't process enough data to make accurate predictions.
The result: 50% inventory accuracy. Half the time, the demand prediction was wrong.
What they built
AI demand forecasting models trained on years of data:
- Historical sales patterns by product, location, and time period
- Purchase order history
- Inventory levels and transfer records
- Seasonal trends and promotional impact
The system predicts what will sell, where, and when. It also recommends inventory transfers between locations to match predicted demand.
Before the AI system, demand forecasting relied on an eight-person team plus in-store managers making manual estimates. The process was slow, limited by how much data humans could analyse, and produced inconsistent results across locations.
The numbers
| Metric | Before | After | |---|---|---| | Demand forecasting efficiency | Baseline | 10x improvement | | Inventory accuracy | 50% | 90% |
"Our demand forecasting is 10 times more efficient now. Our accuracy has improved to 90%, when it was previously about 50%." — Elran Aharonee, Data Manager at Super-Pharm
What "10x more efficient" actually means
It's not just that predictions are more accurate. It's that the same team can now forecast across all 290 locations with better accuracy than they could previously forecast for a fraction of stores.
The manual process involved staff pulling data from multiple systems, building spreadsheets, and making educated guesses. Each store's forecast took time, so the team prioritised high-volume locations and estimated the rest.
With AI handling the data processing, the entire chain gets equally accurate forecasts. Lower-volume stores benefit the most because they were previously getting the least analytical attention.
The marketplace extension
Beyond forecasting, Super-Pharm uses AI to automatically categorise products on their online marketplace. When a new product is listed, the AI determines what category it belongs in and where to place it on the "virtual shelves" to maximise sales.
This is a small detail, but it matters. Manual product categorisation across tens of thousands of SKUs creates inconsistencies. Some products end up in the wrong category, some get buried. The AI ensures consistent placement based on what data shows sells well in each position.
Source: Google Cloud case study
Does this apply to smaller businesses?
Super-Pharm is a large retailer, but the principle scales down. Any business that holds inventory and needs to predict demand can benefit from this approach. The tools are more accessible than ever.
Even a 10-store retail chain or an ecommerce business with a few hundred SKUs can use demand forecasting models to reduce overstock (which ties up cash) and stockouts (which lose sales).
The core question is the same regardless of scale: are you making inventory decisions based on data or gut feeling? If it's gut feeling across more than a handful of products, there's room for AI.
Key Takeaways
- Inventory accuracy: 50% to 90% using AI demand forecasting.
- Demand forecasting 10x more efficient across 290 stores.
- The biggest wins come from stores that previously got the least analytical attention.
- AI product categorisation ensures consistent online merchandising.
- The principle applies to any inventory-holding business, not just large retailers.
Holding inventory and guessing at demand?
Whether you run a retail chain, an ecommerce store, or a wholesale business, if you're making stock decisions based on spreadsheets and intuition, there's probably room to improve. I build AI automation pipelines that turn historical data into actionable forecasts. Let's talk about what's possible.
Related reading: