The robots arrived, picking got faster, and a few weeks later the fast movers started running out at the pick face.
Nothing about the stock changed. What changed is the rate at which the forward locations empty. Robots pick faster than the people they replaced, so a location that used to last a shift now lasts a fraction of one, and the process that refills it still runs at the old pace, on whoever happens to be free. The result is a robot waiting at an empty location, or a picker walking to reserve to fetch stock by hand, the exact journey the automation was bought to remove.
The short answer
Pick-face stock-outs increase after automation because the robots empty forward locations faster than people did, while the replenishment process that refills those locations still runs at its old pace. The minimum levels, the task priorities and the people assigned to replenishment were all set for a manual picking rate that no longer exists.
The drain rate went up and the trigger did not. A minimum level is a bet about how long a location lasts once it reaches that level. At manual pick rates that might be most of a shift. At robot pick rates the same minimum can be reached and emptied inside the hour, so by the time the task is raised the location is already short.
The fast movers go first. The stock-keeping units (SKUs) in the forward area are by definition the lines that sell most, so they are the ones the robots visit most and empty soonest. The stock-out lands on precisely the lines where a missed pick costs the most.
Our piece on receiving and putaway in an automated warehouse covers getting stock off the dock and into reserve. This piece starts where that ends.
Replenishment competes with picking because on an automated floor both jobs are done by the same fleet. Every robot moving a tote from reserve to the pick face is a robot not completing a pick, so the system allocating work has to choose. Without a rule, the choice falls to whichever task is loudest, which is almost always the pick.
The reason is the shape of the deadlines. A pick has a cut-off attached to it and a client behind the cut-off. A replenishment has no deadline of its own, so it fails quietly, minutes later, when a robot arrives at the location it would have filled. The pattern is the same one that plays out between inbound and outbound at the dock, and so is the fix: decide in advance how much of the fleet replenishment is entitled to, rather than leaving it whatever is left over.
That decision has to live in the allocation layer, the only place where picks and replenishments are visible side by side. A warehouse management system (WMS) raises the replenishment task and a single supplier's fleet manager runs its own robots, but neither weighs a pick against a replenishment for the same machine at the same moment. Our guide to what a warehouse orchestration layer is explains where that weighing happens.
A pick-face stock-out costs you robot time and the labour you automated away. A robot that reaches an empty location either waits or is re-routed, and either way the capacity you paid for is idle. If a person covers the gap by walking to reserve, you are back to paying for the travel the automation was bought to remove.
It is not an accuracy problem. The right item is picked late, or not at all, and reducing picking errors in an automated warehouse is a different job with different fixes. The cost shows up in four forms:
Both. Minimum and maximum levels are the floor: a location that drops below its minimum must be refilled to its maximum regardless of anything else. The order profile for the shift is the plan: it shows which locations will run dry in the next few hours, so they can be filled before the picks that empty them are released.
Min and max alone is reactive: the trigger fires when the location is already low, and at robot pick rates "low" and "empty" are close together. A fixed minimum also cannot tell an ordinary Tuesday from the day a promotion lands. The order profile can, because the work for a shift is known before it is released, so a location the next wave will empty can be filled ahead of the wave rather than halfway through it. On a shared floor the profile changes with the client mix, so the plan is rebuilt per shift. Our piece on multi-client warehouse automation covers why the mix matters.
In practice the rule set looks like this:
You make replenishment a first-class task by giving it the three things every pick already has: a priority, a deadline and a share of the fleet. A replenishment with a deadline derived from when its location will empty can be weighed against a pick on equal terms. A task with no deadline will always lose.
Give every replenishment a deadline. Not "when free". The deadline is the predicted moment the location runs dry, derived from the pick rate against it, so the allocation layer can sequence it against picks on the same basis.
Reserve a share of the fleet ahead of known waves. If the next wave will drain a set of locations, the robots needed to fill them are allocated before the wave, not after the first robot stalls.
Chain tasks so replenishment rides on empty travel. A robot finishing a pick near reserve carries a replenishment on its way back towards the pick face.
Handle replenishment exceptions live. The reserve location is empty, the tote is missing, the pallet is in the wrong slot. Each stalls a replenishment, and a stalled replenishment becomes a stock-out at the pick face unless it is reassigned straight away. Our piece on real-time exception handling covers how.
This is where a coordination layer earns its place. FloxMind sits between your warehouse systems (WMS, warehouse execution system or enterprise resource planning) and the robot controllers, and allocates and prioritises tasks in real time across picks and replenishment, with exception handling and a live view of every machine. It is not a slotting tool and it does not replace your WMS. It decides what the robots do next.
Slotting decides where stock lives. Replenishment keeps it fed. Better slotting shortens the replenishment journey and reduces how often a location needs filling, so it helps, but it does not remove the task. The fast movers in a forward area are by definition the lines that empty most often, wherever you put them.
Your WMS or a slotting tool decides where things go. Getting stock there before the pick arrives is the coordination problem that remains after slotting has been done well, because the robot carrying the refill is still a robot not picking.
Three numbers tell you whether replenishment is the constraint: pick-face stock-out events, replenishment lateness, and the share of fleet time spent on replenishment at each hour of the shift. None of them sits on a standard fleet dashboard, and all three should be baselined for a normal week before any rule is changed.
Our guide to warehouse automation KPIs sets out how to baseline.
Coordinating an existing fleet across picks and replenishment tends to produce throughput uplift of 20 to 40% and return on investment in 4 to 12 months, with decision-making around 3x faster when allocation runs live. In one anonymous third-party logistics (3PL) e-commerce deployment using goods-to-person automation, picking throughput rose by 40%. These are ranges rather than promises, and how much you recover depends on how much of your lost throughput is robots waiting at empty locations.
A coordination layer works across 100+ robot models from different suppliers, keeps your existing WMS in place and needs no in-house robotics team. If your fast movers are running dry at the pick face, book a technical demo and we will look at your stock-out and replenishment numbers with you.