Inventory optimisation: holding the right stock, in the right place

Get inventory optimisation wrong and you end up with two expensive problems at once: shelves full of stock nobody wants, and empty ones where the demand actually is.

Inventory optimisation is the practice of setting stock levels, reorder points, and safety stock to match real demand and supplier lead times, rather than habit or gut feel. It balances service levels against holding costs across every SKU and location, so a business ties up less cash without running out of the products customers actually want.

What inventory optimisation actually means

Inventory optimisation means setting stock levels, reorder points, and safety stock by SKU and location, based on measured demand variability and supplier lead times, rather than by habit or gut feel. Most businesses already do some version of this, usually with spreadsheets and reorder rules that haven't been revisited in years. Slow stock sits there because nobody flagged it. Fast movers run out because the reorder point was set for last year's demand pattern, not this year's.

Done well, this runs as a recurring process, not a one-off stocktake or clearance sale. It touches purchasing, warehousing, demand planning, and finance, because a decision to hold three weeks of safety stock on a slow-moving line has a direct cost attached, and someone owns that cost even if they weren't in the room when the decision was made.

It also isn't the same thing as inventory management, which is the day-to-day mechanics of ordering, receiving, and tracking stock. Optimisation sits a level above that. You can run tight inventory management and still be badly optimised, if the targets you're managing to are wrong.

Safety stock and service levels

Safety stock is the buffer held above expected demand to cover uncertainty in both demand and supply. A supplier whose lead time is usually seven days but occasionally stretches to fourteen, combined with a sales pattern that spikes around promotions, needs a buffer sized for that variability, not for the average case. Plenty of businesses calculate safety stock from averages alone and then can't work out why they still run out mid-month.

Service level is the target sitting behind the buffer: the percentage of demand you want to satisfy directly from stock on hand, without a customer waiting for replenishment. A 95 percent target looks close enough to 99 percent on paper. The stock required to hit it usually isn't close at all, because the relationship isn't linear. Pushing from 95 to 99 percent typically costs far more in extra stock than pushing from 90 to 95, for the same product.

This is where most of the practical work happens: setting different service level targets for different products, rather than one blanket rule across the whole catalogue. A core line that drives repeat orders might justify a 99 percent target and the stock cost that comes with it. A slow-moving accessory almost certainly doesn't, and treating it the same way just ties up cash for no benefit.

The role of forecasting and AI

Forecasting accuracy sets the ceiling on how well any of this can work. A safety stock calculation built on a forecast that's routinely wrong by 40 percent produces buffers that are either too thin to matter or padded so heavily that the whole exercise stops paying for itself.

Traditional forecasting methods, moving averages and simple seasonal indices, cope reasonably well with steady, high-volume lines. They struggle with new products, promotional spikes, and anything affected by weather, local events, or supplier disruption, because they can only see the pattern in their own sales history and nothing else.

This is one of the more genuinely useful applications of AI for supply chain problems: models that pull in promotional calendars, weather data, competitor pricing, and supplier performance alongside historical sales, then re-forecast as new information arrives instead of waiting for the next quarterly review. The improvement isn't dramatic across the board. It tends to be largest on the products that are hardest to forecast by hand, which are usually the same ones causing the most stockouts or write-offs in the first place.

None of this replaces judgement. A planner who knows a major customer is about to place a one-off order, or that a competitor just went out of stock, still needs to override the model. The point of the AI layer is to handle the routine 90 percent of SKUs well enough that people have time left for the 10 percent that actually needs a human decision.

The benefits when it's done properly

The direct benefit is cash. Stock sitting in a warehouse is money that isn't available for anything else, and for many businesses inventory is one of the largest lines on the balance sheet. Reducing excess stock on slow movers while protecting availability on fast movers frees up working capital without touching revenue, which is a rarer combination than it sounds.

The less visible benefit is fewer of the two failure modes that quietly cost businesses the most: stockouts on products customers actually want, and markdowns or write-offs on the ones they don't. Both hit gross margin, just in different line items, and both are hard to see clearly without SKU-level analysis rather than category-level averages.

There's a supplier-side benefit too. More reliable demand signals and steadier reorder patterns make it easier to negotiate lead times and minimum order quantities, because the supplier is dealing with a predictable customer rather than one placing panic orders every few months and expecting priority treatment.

How to implement inventory optimisation

Start with the data, not the software. Inventory optimisation needs accurate, current data on stock levels, sales history, lead times, and unit costs, by SKU and location. If the master data is wrong, or three systems disagree on how much stock is actually on hand, no amount of modelling fixes that underneath.

Segment the catalogue before setting any rules. An ABC or ABC-XYZ split, by revenue contribution and demand variability, shows where a tight service level genuinely earns its cost and where it doesn't. Treating every SKU the same way is probably the single most common reason these projects underdeliver.

Set service level targets by segment, then let the safety stock calculations follow from those targets and the measured variability in demand and lead time, not the other way round. Pilot on one category or one warehouse first. It surfaces the data problems and the internal disagreements about who owns which decision before they get expensive across the whole business.

Only then pick the tooling, whether that's a purpose-built inventory planning system or a forecasting layer sitting on top of the existing ERP. The tool matters far less than the process and the data feeding it, and a good tool bolted onto bad data just produces confident, wrong numbers faster.

Common pitfalls

The most common pitfall is treating this as a one-off project instead of an ongoing process. Demand patterns shift, suppliers change lead times without much notice, and a model tuned six months ago drifts out of date quietly, until someone finally notices a category running out every month and asks why.

Another is chasing a single high service level target across the whole catalogue because it feels like the safe option. That approach is expensive rather than safe, and it hides the fact that most of the value in these projects sits in a fairly small number of SKUs.

A third is skipping the finance conversation. Working capital tied up in stock belongs to somebody's budget, and if operations sets safety stock levels without involving whoever owns that budget, the numbers get quietly overridden later anyway, and the whole exercise gets repeated a year on with the same argument.

A fourth, smaller but persistent, is confusing a system implementation with the optimisation itself. Buying a planning tool and importing last year's reorder rules into it changes the interface, not the underlying decisions. The stock levels need rebuilding from the data, not just relocating from one screen to another.

Frequently asked questions

Want a straight view of where AI can help your business first? Talk to us and start with a clear picture instead of a vendor demo.