Making stock work harder
In retail and e-commerce, the product lifecycle is plagued by a classic operational headache: managing thousands of distinct items individually is humanly impossible, forcing businesses to rely on imperfect rules of thumb. When inventory management fails, brands face the dual sting of lost sales and costly overstock.
To solve this, ZEOS built an end-to-end solution taking the guesswork out of supply chain management. We sat down with Andreas Syren, Senior Principal Applied Scientist at ZEOS, to unpack the core features, predictive simulation models, and profitability algorithms powering the Replenishment Engine.
The engine's core philosophy
Q: Let's start with the product's primary value proposition. What is the goal of the Replenishment Engine?
Andreas: The engine doesn't just make a Partner's life easier; it actively improves their operational performance. It delivers actionable recommendations that are optimised to increase profit contribution and systematically improve stock health.
Instead of just aiming for a generic "high service level" or “target stock cover”, our system is uniquely built around financial outcomes. It evaluates the end-to-end lifecycle of an article, from replenishment and storage to return handling and offering re-sellable returns, explicitly balancing inventory processing costs and storage fees against the opportunity cost of being out of stock when a customer wants to buy.
Q: How does the software translate those calculations into daily recommendations?
Andreas: If we keep it simple: if a brand has plenty of inventory and we don't see a near-term risk of stocking out, the engine advises them to wait. This prevents them from tying up capital or incurring excess fees. However, if an item is rapidly running out, or if it is an extremely high-value product where a lost sale is highly detrimental, the system automatically factors in higher safety margins. Every recommendation is ultimately a tradeoff between risk of lost opportunity and cost exposure - expressed in euros.
Every recommendation is ultimately a tradeoff between risk of lost opportunity and cost exposure - expressed in euros.
Handling supply chain volatility
Q: Supply chains face constant disruption. How does the product handle variables like logistics delays or customer returns?
Andreas: That is where our simulation framework comes into play. The tool runs thousands of simulations exploring possible futures to map out what the future might look like, and how different replenishment decisions perform in those scenarios. It simultaneously evaluates several critical variables:
Current stock levels
Granular demand forecasts
Return rates, historical return behaviours, and return lead times
Logistics lead times and potential shipping delays
We don't just look at if an item will be returned. We factor in the probability of it being returned after one, two, or three weeks, knowing some customers return items quickly and others slowly.
Q: So the algorithm is designed for a "worst-case scenario"?
Andreas: We are actually designing for robustness. We want to deliver a recommendation that works across a wide distribution of scenarios. If a brand gets unlucky with returns, our built-in safety margins protect them. If they get lucky, the system ensures they do not end up drowning in overstock anyway. The tradeoff comes down to the value of sales compared to anticipated costs.
Managing lifecycles and the "graceful exit"
Q: Seasonal items can completely break standard replenishment algorithms. How does the engine adapt to strict product lifecycles?
Andreas: We look at both hard rules—like inventory season tags—and the organic, historical seasonality of specific assortments to map exactly where the peak demand sits throughout the year.
Our demand forecasts are designed to recognise when an item's lifecycle is nearing its end, signaling that demand will soon dry up. The algorithm then strategically selects a date for the last replenishment and halts further orders. This allows for a "graceful exit" from the market, preventing brands from getting stuck with dead stock at the end of a season.
Cyber Week & the overstock feature
Q: Peak trading events like Cyber Week require completely different retail strategies. How does the feature set adapt?
Andreas: Cyber Week is highly subjective, some partners use it as a liquidation event to clear house, while others see it as a massive demand opportunity to push top sellers.
Because of this, the tool features a dedicated Cyber Week Mode that exposes control back to the brand a few weeks before the event. They can input their specific promotional configurations and manually adjust our demand forecasts to align with their trading strategy, and the engine adapts its recommendations accordingly. This mode is interactive, allowing planners to try out settings and see how they affect the recommendations in real time.
Q: Have there been any unexpected ways partners are interacting with the application?
Andreas: Yes! While the core replenishment recommendation is the main driver, we were surprised by how incredibly popular the Overstock Report is. Instead of telling you what to buy, it explicitly flags how much of the assortment that is in excess, so partners can take fine grained action immediately. It is a very popular feature in the application.
The metrics: proven product impact
Q: Software features sound great on paper, but what does the actual performance data look like?
Andreas: The data speaks for itself. When brands follow the engine's recommendations, we see a 21% increase in demand fill rate. Simultaneously, those exact same brands manage to reduce their overstock rate by two-thirds.
Technical Note: The underlying mathematics driving this engine are highly advanced. For those who want to truly get into the technical weeds, our team actually published a peer-reviewed paper in Nature last year detailing the underlying algorithm.
When brands follow the engine's recommendations, we see a 21% increase in demand fill rate. Simultaneously, those exact same brands manage to reduce their overstock rate by two-thirds.
Keeping people in the loop
Q: With an engine this automated, is the goal to completely remove people from inventory management?
Andreas: Not at all. Our brands are always in the driver's seat. We provide optimised recommendations, but we fully expect and welcome deviations. Partners have unique strategic considerations, localised incidents, or uncommunicated plans that an AI simply cannot know. We provide the data-driven foundation; they make the final call.
Product availability
Existing Partners: The tool is available right now in zDirect under the Fulfilment tab. Educational guides are also hosted on the Partner University platform.
Prospective Partners: Brands not yet partnering with us can reach out via the contact form to learn more.
