Trade magazines describe picking as a ladder. You start with paper pick lists. Then you move to scanners, then to pick-to-light, then to goods-to-person systems, and finally to a fully automated line. You climb when your order volume allows it. Manual picking sits at the bottom — the thing you do until you can afford something better.
That ladder is useful for anyone who sells equipment. It also builds the decision on the wrong number. Volume tells you whether you can afford automation. It tells you nothing about whether the machine will have work to do. That depends on how similar your orders are, and on how long they stay similar.
The fair case for automation
In a typical distribution centre, a manual picker handles about 60 to 100 picks per hour. Voice-directed picking raises that to a little over 100. For systems with mobile robots or goods-to-person storage, suppliers quote 200 to 400 picks per hour at each station.
The accuracy figures point the same way. The usual figures are 95 to 97% for manual picking with no support tools, against 99.9% or better for a well set-up automated system. And picking takes 55 to 65% of the total cost of running a warehouse.
So you triple the speed, make ten times fewer errors, and apply both to two thirds of your costs. That is a large gain, not a small one.
Two warnings belong next to these numbers.
First, most of them come from the companies that sell automation. That is not a reason to throw them out. It is a reason to read them as marketing material.
The second warning matters more. Every number above describes a line that is already set up for the job. Such a line assumes four things:
- a product range stable enough to plan storage locations for each SKU;
- order lines that repeat in patterns you can predict;
- standard packaging;
- a volume that stays inside the range the system was built for.
Most ROI models treat these four points as background. In fact, they are the whole argument.
Two sites, same volume, opposite problems
In a single-brand distribution centre, volume and sameness grow together. A larger operation usually has a settled product range, settled packaging and orders you can forecast. Growth makes the site more uniform, not less. Here the ladder works.
Third-party fulfilment breaks that link, and it does so by design.
Picture two European sites. Each one ships 3,000 orders per shift. The first belongs to a single brand: one catalogue, one range of cartons, one marketing calendar, one peak. Any consultant would approve a line here. The second is a multi-client site that runs forty brands through the same doors. Same volume, similar headcount, similar floor area. But also forty catalogues, forty packaging specifications, forty marketing calendars, and forty peaks that never fall in the same weeks.
Think of a staff canteen and a restaurant. Both feed 3,000 people a day. The canteen serves one dish, so a production line suits it. The restaurant has a menu of forty dishes, and guests ask for changes to them. No number of guests per day will make a production line right for the restaurant. The order count alone cannot tell you which building you are standing in.
For fulfillment operators that run multi-client sites across Europe, this is the normal situation, not a rare case. Cross-border e-commerce moves volume into shared warehouses for exactly that reason: so that each brand does not have to build its own. The volume is real. The sameness that automation needs never arrives.
Exception rate: the first number to check
The exception rate is the share of orders that contain something an automated line cannot handle without a person.
In practice, that means:
- marketing inserts;
- leaflets and legal documents for a specific market;
- free samples;
- gift wrapping;
- branded outer packaging;
- bundles of several products put together at the pack bench;
- subscription boxes with contents that change every month;
- any order that needs a batch or serial number check.
The mistake is to treat these orders as a small extra group. In consumer goods — supplements, cosmetics, clothing, anything sold direct to the customer with marketing material in the box — the exception rate on a multi-client site can reach half of all orders. Beauty products are a good illustration of this: as any breakdown of how cosmetics warehousing actually works shows, lot tracking, shelf-life limits, and insert kitting quietly turn “simple” fulfillment into a steady stream of edge cases.
At that level, the numbers are no longer close. The line is faster on half the work, but you pay for the whole machine. Next to the line, you still hire, train and plan shifts for the manual pack bench at almost full strength. The line has not replaced those people. You bought a machine and kept the staff.
There is a clear threshold here. Below about 10 to 15% exceptions, a line plus a small manual bench for the rest is a good design. Above one third, the case usually fails, whatever the order count says.
The catalogue will not stand still
Variety weakens the case for an automated line. Time weakens it in the same way.
Add a new SKU to a manual process, and you spend a storage location, a barcode and a short briefing. Add one to an automated line, and you may spend a new slotting plan, changes to tote or carton sizes, new sensor and gripper settings, and in some systems a change to the control software.
A single brand adds new SKUs at its own speed. A multi-client site adds the new SKUs of every brand it serves, all at the same time. On top of that, it takes in a whole catalogue each time it signs a client, and loses one each time a client leaves. A single-brand site pays this set-up cost once. A third-party operator pays it every quarter.
Machines check identity. People check the condition.
Accuracy comparisons hide this difference, and the difference costs more than the percentage points do.
An automated check confirms identity: the right item, in the right order, in the right quantity. It checks a barcode, a weight, and sometimes an image from a camera. It cannot confirm the condition. A person holding the product sees the crushed corner. They see the seal that has started to leak, the colour that does not match the rest of the batch, the print defect on the sleeve, and the lid that has not been closed straight.
None of this appears in pick accuracy figures, because none of it is a picking error. It appears later, as a return, a replacement parcel and a support ticket. That costs several times more than catching the problem at the bench. When the unboxing is part of the product, this later bill is not a small detail.
Label application runs in the same way whatever sits inside the box. That is what makes it worth automating.
The accuracy gap is smaller than it looks
Accuracy is the one place where automation looks clearly ahead. The lead is smaller than the comparison suggests, because the manual figure everyone quotes describes manual picking with no support tools at all.
Barcode scanning at the pick cuts picking errors by about half, against paper or memory. Then add a checkweigher at the pack bench. It is one of the cheapest checks on the market, and many well-run manual sites still skip it. Add a scan before the parcel leaves, plus a clear plan for where each SKU is stored. Manual picking now reaches the level most e-commerce brands need.
So the honest comparison is not 4% against 0.04%. It compares an improved manual process with an automated one. The gap between them is small enough that capital cost, set-up cost and exception rate decide the answer — not accuracy.
In plenty of buildings, that comparison still favours automation. High volume, little variation, a stable catalogue, standard cartons: install the line and stop reading. This article is about another kind of building. European e-commerce holds more of them than the ladder model suggests.
What to automate in a manual warehouse
Saying no to an automated picking line is not an argument for a warehouse without machines. Operations that read it that way lose money.
Automate every step that stays the same, whatever the order contains. Label application is the clearest case. Every parcel gets a label, the task runs in the same way whatever sits inside the box, and a labelling machine pays back fast without changing any step before it. Sorting parcels by carrier and by destination works the same way.
So does travel. Walking takes most of a picker’s time. Goods-to-person storage that feeds human pack benches removes that walking, and it is the single largest gain a manual operation can make. It also leaves packing in human hands, which is where the variation sits anyway.
Automate the steps that never change. Keep the steps that do change manuals. That decision is different from automating the pick, and most operations should be making it.
Three numbers to check before the ROI model
- Exception rate. What share of orders needs something non-standard at pack — today, and after you sign the next client?
- Catalogue turnover. How many SKUs arrive and leave each year, as a share of the total? And who pays for the changes to the line?
- Peak alignment. If you serve several brands, do their peaks fall in the same weeks? Peaks spread across the year make it harder to choose the right size of machine, not easier.
Volume comes fourth. It settles whether you can afford the line. Whether the line will have enough work to do is a different question — and it is the one worth answering first.
If you are the brand, not the operator
Everything above looks at the decision from inside the warehouse. If you are choosing a fulfillment partner, turn the question around and start with your own orders.
Start with your customization. Count what goes into a typical box besides the product: inserts, samples, gift wrap, leaflets for a specific market, bundles put together at the pack bench. That share is your exception rate, and it does not disappear when you hand the work to a partner.
Ask what happens to a non-standard order. On a good site there is a clear route for it. Ask who does that work, how long it takes and what it costs per order. The answer tells you more than a tour of the equipment.
Ask which steps are automated, not whether the site is automated. Labelling, sorting and travel are worth automating for every client. A partner that has automated those steps and kept packing manuals has made the right choice for a mixed catalogue.
Match the site to your plans, not to your volume today. If you add bundles, market-specific inserts or a subscription box next year, your exception rate goes up, not down.
And drop one idea: that a manual warehouse is a weaker warehouse. On a site that serves forty brands with forty packing standards, manual packing is not a sign that the operator could not afford a machine. It is the design that fits the work.
