Most articles on this subject explain why you should automate. This one does the opposite. There are warehouses where automation is the wrong move for now - too small to carry the business case, wrongly diagnosed, or better served by fixing the manual process first - and pretending otherwise costs a six-figure sum and a year of runway. Here are the honest downsides, the five situations where automation does not pay, and the alternatives worth trying before it.
The case for warehouse automation is well documented, and most of it is sound: labor is scarce, wages rise every year, and manual picking has a hard productivity ceiling. What that literature rarely says is where the argument stops. Automation is a tool with a fit window, and outside it the honest recommendation is to wait, optimize, or buy differently than a vendor would like. The barriers that keep most warehouses manual are covered in warehouse automation challenges; this article asks the harder question of whether automating is the right call at all. Nothing below is a disqualifier for other people and an exception for us - the same lines apply to NEO.
The real downsides of warehouse automation
Every automation architecture buys throughput with commitment, and the commitment is easy to underweight while a demo is running. The first cost is capital and the lock-in that comes with it. Classic goods-to-person systems - AS/RS, shuttle, cube-based storage - are a multi-million investment, often in the tens of millions once construction is included. That capital is tied up for years before it earns back: shuttle systems typically reach payback in 3-5 years and AS/RS installations in 4-7, and the depreciation runs whether or not volume holds.
The second cost is rigidity. A fixed installation is optimized for one layout, one product mix, and one throughput profile at the moment it is commissioned, and it does not follow easily when the SKU mix shifts, a channel changes, or volume moves in either direction. Reconfiguring a shuttle or extending an AS/RS often rivals the original outlay, which is why operators end up running partly obsolete systems rather than paying to change them.
The third cost is dependency. Proprietary hardware and integrator-specific control software mean switching vendors after installation ranges from expensive to impractical, and with grid-based systems it is close to starting over. That is a reason to ask, before signing, what a change in five years would cost and which components survive it.
Two more costs are routinely treated as afterthoughts and routinely become the critical path. WMS integration is one slide in most pitches and one of the top reasons projects overrun in practice - a missing real-time API or an unplanned sub-WMS can slip a schedule by months. And automation changes jobs, which makes it a workforce project as much as a technical one: pickers, supervisors and, where applicable, the works council belong in the evaluation, not in a briefing after the contract is signed.
None of this argues against automation. It argues for checking the fit before committing - which is the next question.
Five situations where automation does not pay off
1. Below the volume threshold
Automation economics are driven by how much work the system absorbs, and below a certain daily volume there is not enough of it to amortize the cost. As a working threshold, a goods-to-person retrofit becomes economical from around 5,000 picks per day in the area being automated. Under that line the per-pick math rarely closes, whatever the payment model, because the fixed or minimum costs are spread too thin.
The right move below the threshold is not to force a smaller system in but to improve the manual operation first - the alternatives further down often lift productivity enough that the automation question can wait until volume genuinely justifies it.
2. No shelf-based picking to automate
A retrofit solves a specific problem: too much walking in a manual, person-to-goods shelf operation. An operation that is already high-bay AS/RS end to end, with no manual picking area, does not have that problem for a retrofit to solve.
If your bottleneck is storage density in an automated hall rather than travel time on foot, a shelf retrofit is the wrong tool, and a look at how the architectures differ - for instance compared to cube-based AS/RS - will point you somewhere else.
3. A pure new build with no existing facility
Retrofit automation earns its keep by working inside a building you already run, without construction and without downtime. On a greenfield site those advantages have nothing to bite on: when you are pouring a floor and designing a hall from scratch, the constraints that make a retrofit attractive - keep the racking, keep operating, go live in weeks - simply do not apply.
Greenfield projects follow a different logic. With a clear multi-year forecast, a capital budget, and a building shaped around the technology, a purchased fixed system can deliver the lowest unit cost over its life. That is a legitimate outcome, and a retrofit-first vendor is the wrong first call for it.
4. Stable full load for many years, with capital to spend
This is the situation where buying beats subscribing, and it deserves an honest hearing rather than a sidestep. A warehouse that runs at consistently high utilization for many years, on its own premises, with an investment budget available and a horizon of seven years or more, is the textbook case for CapEx. Bought and depreciated over its full life at full load, a fixed system delivers a lower cost per pick than any variable model, because a pay-per-pick price necessarily includes the provider's margin for carrying technology and performance risk.
Pay-per-pick is not always cheaper, and any vendor who claims otherwise should be pressed on the math. Its advantage is situational: it wins where volume is seasonal, growing, or uncertain, or where capital or approval is the binding constraint. Where none of that is true, the variable premium is a cost without a matching benefit. The trade-offs, and how to compare the models per pick, are set out for smaller operators in our piece on financing paths for mid-size operators.
5. A decision made before the diagnosis
The most expensive automation is the kind that automates the wrong thing efficiently. It happens when a warehouse is read through averages instead of distributions. "Five picks per order" hides whether most orders carry one or two lines and a minority carry twenty, and a system optimized for the mean serves neither. The same applies to SKUs: the Pareto shape of the catalog matters more than the count.
A single architecture stretched across the whole assortment is the usual result - long-tail SKUs forced into a system built for fast movers, or vice versa, with unit costs that only surface months later. The correction is cheap relative to the mistake: read order and SKU data as distributions, with median and high percentiles, before shortlisting any technology. If that diagnosis has not been done, the warehouse is not ready to automate yet, regardless of volume - the same discipline separates projects that deliver from the ones examined in why automation projects fail.
What to do instead: the alternatives worth trying first
For many warehouses, the highest-return move is not automation at all but a better manual operation - and the data it produces sharpens any later automation decision. In manual person-to-goods warehouses, research puts roughly 50% of a picker's working time into travel between locations (Tompkins et al.). That is the single largest reserve, and most of it can be reclaimed without capital.
Slotting is the first lever. Placing fast movers near the dispatch point and in the ergonomic pick zone, grouping items that ship together, and re-slotting as demand shifts cuts travel directly. Travel-path optimization in the WMS - sequencing picks so the route through the aisles is short rather than the order they were entered - compounds the effect. Neither requires new hardware, only attention and reasonably current warehouse-management software.
Batching and zoning change how orders are grouped rather than how the building is built. Batch picking serves several orders in one pass and shortens the aggregate walk, provided the orders are similar enough to combine. Zone picking splits the warehouse into areas with a picker each and passes the order between them, parallelizing the work without moving to goods-to-person. Both are process changes on top of existing racking and WMS integration, not capital projects.
Reverse picking is the specialist option, powerful only inside a narrow profile. It inverts the flow - one SKU distributed across many waiting orders in a single pass - and it beats any collecting strategy when the fit is right: an assortment under roughly 5,000 SKUs with order overlap of 50% or more, meaning a given item appears in at least every second order. Outside that profile the number of passes becomes the bottleneck and the advantage disappears, so adopt it on evidence, not enthusiasm.
Between all-manual and full-facility automation sits partial, modular automation - automating one zone or one flow rather than the whole building. The economics favor it more often than the all-or-nothing framing suggests: Prologis research finds that modular automation requires roughly one-third the capital expenditure of full facility automation while delivering approximately 1.5x more throughput gain per dollar invested. Starting with the zone that hurts most and expanding on measured results is often the better path, even for operations that will eventually automate broadly.
And when automation is worth it
The conditions for automating are not hard to state. Daily volume from around 5,000 picks per day in the area under review, giving the system enough work to carry. An existing shelf-based warehouse where travel time, not storage density, is the bottleneck. A labor situation that is tight or spikes hard in peak season. And no contractual commitment to a competing system for the same workflow. Where those hold, the manual process has usually been optimized as far as it reasonably goes, and the remaining gains need automation to reach.
The payment model is a separate decision from the technology. For sites with seasonal, growing, or uncertain volume - or without the budget or approval appetite for a large fixed purchase - a variable model earns its premium by shifting the sizing risk to the provider. That is the ground on which retrofit automation on a pay-per-pick basis is designed to sit: existing racking, go-live in 6-8 weeks against 12-36 months for classical projects, and cost that scales with volume rather than a multi-million commitment sized to the peak. Across NEO deployments the aggregate figures are 70% less picking labor and 2-3× storage capacity on the same floor space. It is one path among several, and the right one only when the situation above describes your warehouse - which is the point of naming where it does not.
Find out which side of the line you are on
If you are weighing automation, the useful first step is an honest read of whether your warehouse is inside the fit window. That is exactly what one session covers - shelf type, aisle width, daily volume and order structure - and it is built to say no when the warehouse does not (yet) fit, including when the right advice is to optimize the manual process first: See NEO in action.
If you would rather start on your own, our resource library compares the automation architectures with their investment, timeline and risk profiles as a market overview for self-diagnosis (in German).
