Inventory optimisation with data and AI
For each item, we calculate safety stock and the reorder point based on demand and lead time. For Athlon, we built predictive models that show which parts are needed and when.
At Athlon, the right parts needed to be available at the right time, without unnecessary capital tied up in stock
Parts for maintenance and repairs were spread across separate systems and spreadsheets, leading to excess stock and missed reservations. We built a central data platform with a reservation system and predictive models that show which parts are needed and when.
Why is the warehouse full, yet the item the customer wants is out of stock?
Minimum stock levels and reorder points are often still based on figures set years ago, with the same values used for many items. An inventory model calculates how much of each item you need until the next delivery, with a buffer that matches your chosen service level.
A buffer and reorder point for each item, balanced against capital tied up in stock
Safety stock for each item
We calculate the buffer for each item from variations in demand and lead time, rather than applying one fixed margin to everything.
A reorder point that adjusts
The reorder point is the expected demand during lead time plus the buffer. If demand or the supplier changes, the reorder point adjusts too.
Service level versus capital tied up
For each product group, you can see how much extra stock a higher service level requires, so procurement and finance can decide together.
Reserve and plan ahead
For Athlon, we built a reservation system on a central data platform, so every department can see which parts are needed and when.
We test new ordering rules against your historical data before using them in purchasing
-
Items and rules
We categorise items by value and predictability and record the rules: minimum order quantity, packaging, shelf life and fixed delivery day.
-
Testing against historical data
We simulate new buffer levels and reorder points over a past period: how often did an item run out, and how much stock sat idle?
-
An order proposal for the buyer to review
The model proposes an order; the buyer accepts or adjusts it, giving a reason.
Is our product range becoming too large for fixed ordering rules in the ERP?
A good fit if
You have hundreds or thousands of products, demand and lead times vary, and you face both excess stock and shortages. Your ERP or WMS contains historical data on usage, purchasing and deliveries.
Not a good fit if
You have a handful of products with stable demand, or you make everything to order. In that case, correctly set minimum and maximum stock levels in your ERP are enough.
What determines the scope
How many products and locations are involved, how accurately lead times and stock movements have been recorded, whether the recommendation needs to feed back into the ERP or WMS, and who manages the model.
Will buyers follow the recommendation?
Only if they can see where it comes from. That’s why each recommendation shows the demand, lead time and buffer behind it, and we track when a buyer deviates from it and why.
Min-max in your ERP, a buffer for each ABC class or a model for each item?
Min-max rules in your ERP
Choose this if demand and lead times are stable and your product range is small. Review the values regularly, though, or they will fall behind what happens in practice.
A buffer for each ABC class
Choose this if you want to get started quickly: a higher service level for A items and a lower one for the long tail. It is a rough approach, but better than a single rule.
A model for each item
Choose this if demand or lead times vary significantly and capital matters. Our advice: start with the items that account for the most stock value and run a simulation first.
An inventory model depends on the data platform behind it, and we built both for Athlon
Data Science Lab has worked with data and AI since 2016 as a Databricks Partner and Microsoft Partner. For Athlon, we built the data platform, the data pipeline in Azure and predictive models for spare parts, working in short iterations with the operational teams.

Your product list and current ordering rules are the starting point for our first conversation
Stef Stoelwinder
Business DevelopmentStef is your point of contact. In our first conversation, we’ll discuss your product range, your ordering rules and the current state of your inventory data.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which product category ties up the most money right now, or where do you most often run out of stock?
Describe your product range, how you currently place orders and which system holds your stock data. We’ll reply by email or arrange an initial conversation.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
Inventory optimisation means deciding how much of each item to keep in stock, when to order and how much to order. The aim is to achieve a chosen service level while tying up as little capital as possible in stock. Optimising inventory with AI means using a model to estimate demand and lead time for each item, then calculate safety stock and reorder points.
You can read about the forecasting itself under forecasting. It is part of our AI applications.
The reorder point is expected demand during the lead time plus safety stock. That buffer depends on how much demand fluctuates, how reliably the supplier delivers and the service level you want. If lead times vary, that variation also needs to be taken into account.
The quantity you order each time balances ordering costs against holding costs, as in the economic order quantity (EOQ). In practice, pack sizes, minimum order quantities and discounts affect that figure.
The service level per order cycle is the probability of not running out of stock between two deliveries. The fill rate is the proportion of demand you can fulfil immediately from stock. An item can be temporarily out of stock quite often and still have a high fill rate, or vice versa.
Decide in advance which measure matters. The final steps towards a higher service level require proportionally much more stock, so we agree on the target for each product group.
Slow-moving or obsolete stock is inventory that barely moves or no longer moves at all. A model flags items whose stock will last a long time relative to consumption, or whose stock levels have not changed for some time. There are separate methods for items with sporadic, unpredictable demand, such as Croston’s method.
Whether to discount, return or write off that stock remains a business decision. You can read about a pricing model for this under price optimisation.
You need stock movements and consumption or sales by item and location, purchase orders with order and receipt dates so you know the actual lead times, and agreements with suppliers. Also record when a customer requested something that was out of stock: otherwise, the model sees only sales, not demand.
You can read why this matters in our blog post why your forecasting model is right, but your inventory isn’t.
That is up to you. The model makes an order recommendation that follows your business rules: minimum order quantities, packaging, supplier contracts, shelf space and shelf life. The buyer can accept or adjust it and record why.
This lets you see, by product group, whether the recommendation or the buyer’s decision worked out better. Only if you want it to will some orders, such as those for long-tail items, be placed automatically.
Agree in advance which measures matter: service level or fill rate, inventory value, stock turnover and slow-moving or obsolete stock. First simulate the new rules using a past period, then test them on one product group or location alongside a comparable one.
You can track those figures in an inventory dashboard. If demand changes, we retrain the model with MLOps.
When your product range is small and demand is stable, when you make to order, or when lead times and stock movements have not been recorded reliably. In the last case, start with the data foundation.
If you are dealing with orders for individual stores, take a look at Prediqt for sales forecasting and order recommendations. If your main priority is planning capacity or staffing, read about optimising planning with AI.
Forecasting demand or tracking inventory: where can you read more?
- AI applicationsAll the applications we build, from predictive models to language processing and image recognition.
- Forecasting and predictive modelsPredict demand by product and period, providing inputs for every buffer and reorder point.
- Inventory dashboard in Power BIInventory value, turnover and obsolete stock by location, all in one view.