Summary
Warehouse picking performance is how fast, how accurately and at what cost order lines get from their storage locations to the pack station. Five KPIs measure it: pick rate, picking accuracy, travel per line, order cycle time and cost per line. Most of what moves those numbers is decided before a picker sets off: where each item is slotted, which orders share a trolley and the route the trolley takes.
Order picking is expensive. A 2024 paper in the journal Mathematics puts it at up to 55% of all warehouse operating costs, so a difference of a few percent is visible in the budget.
Nobody on the floor can decide to pick 20% more lines per hour, and telling people to walk faster does not change that. Pick rate follows from the work each pick list hands out: how far apart two picks are, how many orders ride on one trolley, how long someone waits at an empty pick face. Those are the things a manager can change.
Which KPIs measure warehouse picking performance?
Five picking KPIs cover the job: pick rate, picking accuracy, travel per line, order cycle time and cost per line. Each one needs to be cut by zone, by shift and by order type, because a total for the whole building hides where the time goes.
The worked examples all use one sample day in one pick zone: 1,600 orders, 4,800 order lines and 7,200 units, picked in 60 picking hours at a loaded labor cost of $30 per hour. 24 lines were picked wrong, in 24 different orders. We made these figures up to show the arithmetic, so yours will differ.
Picking hours are the hours booked to picking: walking, searching, taking the item, confirming, and waiting inside a pick tour. Breaks, packing, receiving and put-away are left out. Where exactly a site draws that line matters less than keeping it in the same place every week, because otherwise the weeks cannot be compared.
1. Pick rate
Pick rate is how much picking work gets done per hour of picking labor. You can count it in order lines or in units, and the two numbers answer different questions.
Lines per hour = order lines picked ÷ picking hours
Units per hour = units picked ÷ picking hours
On the sample day that gives 4,800 ÷ 60 = 80 lines per hour and 7,200 ÷ 60 = 120 units per hour, for the same shift and the same team.
Lines per hour is the one to follow week to week. Every line means a walk to a location, so the line rate tracks the walking, the searching and the waiting. The unit rate mostly tracks how much customers ordered per line.
How far apart the two numbers sit depends on the order profile, and the order profile depends on the business:
- A fashion e-commerce order for one shirt in size M is one line and one unit. In operations like this, lines per hour and units per hour come out close together.
- A store replenishment order for six bottles of the same shampoo is one line and six units. The walk is the same, and the unit count is six times higher.
- A grocery order line for ten cases of water, taken from a pallet location, is one line and ten cases. Here the time goes into lifting more than into walking.
The MedTech distribution center described later in this article reports its volume in order lines: 30,000 a day. On any site, a drop in lines per hour on a day full of small orders says more about the orders than about the people picking them.
2. Order picking accuracy
Order picking accuracy is the share of picks that are right: the right item, in the right quantity, in the right order. You can count it per line or per order, and the same shift gives two different results.
Line accuracy = (lines picked minus lines with an error) ÷ lines picked × 100
Order accuracy = (orders picked minus orders with an error) ÷ orders picked × 100
On the sample day, 24 of the 4,800 lines were wrong, so line accuracy is 99.5%. The 24 errors sat in 24 of the 1,600 orders, so order accuracy is 98.5%.
Order accuracy falls as orders get longer, even when every line gets the same care. If errors are spread evenly, the chance that a whole order is right is the line accuracy multiplied by itself once for every line. At 99.5% per line, a 30-line order arrives fully right only 86% of the time.
That changes how a target reads. ASCM sets the order accuracy target for leading operations at 99.5 to 99.9%. For a 10-line order, hitting that target takes line accuracy of 99.95 to 99.99%.
So use line accuracy to follow the picking process, and compare order accuracy only between operations with similar order sizes. A pack station accuracy check works at order level, which makes it an order accuracy figure too.
The same error also costs different amounts in different businesses. A wrong size in a fashion order turns into a return and a second shipment. A missing case on a store delivery is an empty shelf until the next truck.
3. Travel per line
Travel per line is the time or distance walked for each order line.
Travel per line = travel time ÷ order lines picked
A floor study on the sample day found that half of the 60 picking hours went to walking. That is 30 hours over 4,800 lines, or 22.5 seconds of walking per line. It takes a floor study or scan timestamps to get this number, and it is the KPI that moves first when slotting, batching or routing changes.
4. Order cycle time
Order cycle time runs from order release to pick complete, reported as a median per order type.
Order cycle time = pick complete time minus order release time
The sample day's median is 95 minutes: 70 minutes waiting for the tour to start and 25 minutes on the trolley. Those two parts are worth reporting separately, because the wait before the first pick is a planning matter. Wave release and batch size set it, and no change on the floor will shorten it.
5. Cost per line
Cost per line is what one picked order line costs in labor.
Cost per line = picking hours × loaded hourly cost ÷ order lines picked
On the sample day, 60 hours at $30 is $1,800, and $1,800 over 4,800 lines is 37.5 cents per line. Finance reads this number, so it is the one that belongs in a business case. It falls whenever lines per hour rise.
What is a good pick rate?
There is no single good pick rate. Published figures run from about 71 picks per hour to 175 pieces or cases per hour, and they disagree because they count different work. A more useful target is your own baseline per zone, shift and order type, compared week over week.
The average warehouse pick rate you find online depends on who counted and what they counted. One fulfillment directory reports a typical rate of 71.42 picks per hour from its own survey. It does not say how many sites answered, in which year, or what counted as a pick.
The same ASCM article expects an average picker to handle 120 to 175 pieces or cases per hour, and more than 250 at the top end. It dates from October 2022, carries an editor's note that it has been updated, and cites no source for the range.
Both published figures may be right, and they still cannot be compared. Four things move the figure before anyone works faster or slower:
- The unit counted: order line, unit, piece or case.
- The hours in the sum: with or without waiting, replenishment and walking back to the pack station.
- The order profile: lines per order and units per line.
- The building and the method: pick density, aisle length and how many orders travel on one trolley.
A baseline you can act on comes from a few weeks of your own data, cut by zone, shift, order type and picking method. Leave individual names out of it. A rate per person mostly reflects which orders and which aisles that person was handed.
Picking productivity calculator
The calculator uses the formulas above and needs no sign-up. The basic version asks for three numbers for one zone and one shift: order lines picked, units picked and picking hours. It returns lines per hour, units per hour and units per line.
Two optional steps go further. Add lines with an error and a loaded hourly cost, and it returns line accuracy and cost per line.
The third step estimates what a shorter walk does to the pick rate. Walking is only part of the time per line, so a cut in walking distance moves the pick rate by less than the cut itself:
New lines per hour = current lines per hour ÷ (1 minus walking share × walking reduction)
On the sample day, half of the picking time is walking. A walk that is 15 to 20% shorter saves 7.5 to 10% of the time per line, which lifts 80 lines per hour to between 86.5 and 88.9. That is a gain of 8 to 11% in pick rate, assuming walking time falls in line with distance and everything else stays the same.
The next section explains how to measure the walking share on your own site. Run the calculator per zone and per order type, because a blended figure for the whole building can look stable while one zone slows down and another speeds up.
Where does the picking time go?
Most picking time goes to walking. A 2023 study in the journal Axioms describes travel as the most expensive picking activity, at 55% of picking time. The rest goes to searching, taking the item, confirming, waiting and handling exceptions.
- Travel: walking between pick locations, and to and from the pack station.
- Search: finding the right location, then the right item inside it.
- Pick: taking the item and placing it in the right tote, box or trolley shelf.
- Confirm: scanning or keying the pick.
- Wait: standing at an empty pick face, behind another trolley or at a blocked aisle.
- Exceptions: damaged stock, a short pick, a label that will not scan.
You can get the split for your own site in two ways. The first is scan timestamps: the gap between two consecutive scans is travel plus search for that line, and gaps inside one aisle are mostly search.
The second is a floor study, where someone times a sample of pick tours across zones, shifts and order types for a week and logs each of the six parts. Before the stopwatch comes out, tell the people on the floor that the study is about the pick list and the layout.
The same split answers the question of how to pick faster in a warehouse. The person holding the scanner controls two of the six parts, the pick and the confirm. The other four are decided before the tour starts, by the slot, the orders grouped on the trolley, the sequence of the list and the timing of replenishment.
| What the data shows | Likely cause | First lever |
|---|---|---|
| Long gaps between scans inside one aisle | Searching: labels that are hard to read, several SKUs in one location, look-alike items side by side | Labels readable from the trolley, one SKU per pick face, look-alikes moved apart |
| Long gaps between aisles, tours that cross the building twice | The orders on one trolley are spread out, or the list follows a fixed sequence | Batching and routing |
| Time lost at empty pick faces | Replenishment runs behind the wave | Replenishment triggers set ahead of release |
| Queues in the same aisles every shift | Fast movers concentrated in one aisle | Slotting that spreads the fast movers |
| Lines per hour drops when the order mix shifts to small orders | One picking method for every order profile | A second method for single-line and small orders |
| Errors cluster on a few locations or SKU groups | Look-alike items, unclear unit of measure | Separation, unit-of-measure labels, scan confirmation |
| Pick rate drops whenever new colleagues start | Routes and locations live in people's memory | A system-directed sequence with a visual pick guide |
View this comparison on desktop for the full data table.
If travel comes out under about a quarter of picking time, routing software is not your next project. That threshold is our own rule of thumb. Below it, the gain sits in replenishment, pick face design or the confirmation step.
How do you improve warehouse picking speed and accuracy?
Work through the options in order of cost: fix the process, match the picking method to the order profile, then shorten the walk with slotting, batching and routing. Taking out walking and searching time makes picking faster without costing accuracy. Asking people to hurry the pick itself is where speed and accuracy pull apart.
Fix the process first
The cheapest gains need no software. On the MHI blog in December 2025, Greg Meyne writes that it is not uncommon to find a 10 to 20% improvement in throughput from redesigning processes instead of introducing technology.
In picking, that means replenishment that runs ahead of the wave, so nobody stands at an empty pick face. Location labels should be readable from the trolley. And with one SKU per pick face, the search ends when the picker arrives.
Onboarding belongs here too. If a new colleague needs weeks to learn where things are, a pick sequence directed by the system puts that knowledge on the screen.
Match the picking method to the order profile
Different order profiles call for different picking methods. Single-line orders, large multi-line orders and many small orders with shared SKUs each travel best in a different way, so a site with a mixed order book often runs more than one method side by side.
| Method | How it works | Best for | Main benefit |
|---|---|---|---|
| Discrete | One picker completes one order at a time | Low volume, simple orders | Easiest to train and manage |
| Batch | One picker collects the same SKU for multiple orders in one trip | Orders sharing common SKUs | Removes repeat visits to the same location |
| Cluster | One picker fills multiple order bins on a single cart per trip | Mixed orders, mid-size warehouses | Fewer return trips to the pack station |
| Zone | Each picker covers an assigned area, and orders pass between zones | Large warehouses, high SKU count | Less congestion, shorter travel per picker |
| Wave | Orders are grouped and released in timed waves aligned to shipping cut-offs | High volume, scheduled dispatch | Picking stays in step with outbound |
View this comparison on desktop. Discrete picking is the simplest; batch, cluster, zone and wave picking each cut travel in a different way, and the right choice depends on the order profile.
If trolleys keep coming back to the pack station with one order on board, or half empty, the method no longer fits the orders.
How do you improve order picking accuracy?
Picking errors come in four kinds: the wrong location, the wrong item, the wrong quantity, and the right item in the wrong order's tote. Each has a different cause. That is why the useful count is errors per line by location, SKU group, shift and picking method, and why a count per person hides the pattern.
The fix follows the cause. Look-alike items get moved at least one location apart, and unit-of-measure errors get a label that says each, inner or case. Tote mix-ups in cluster picking drop when each pick is tied to its box before the tour starts and confirmed by scan.
How do slotting, batching and routing cut walking time?
Three decisions set the walk: where each item is stored, which orders share a trolley, and the route through the aisles. Slotting moves stock, so it changes slowly. Batching and routing move nothing but information, so they can change with every pick list.
Every slotting improvement is a trade-off
Where an item lives decides how far every pick list travels. Slotting by pick frequency puts the fastest movers closest to the pack station, and slotting by what is ordered together shortens the gap between consecutive picks.
Each of those gains has a cost, and it usually lands on another objective. Push velocity to its limit and the pickers end up in the same few aisles, so the walk gets shorter and the queues get longer. Group by product family and look-alike items end up side by side, which is where mispicks come from.
No slot plan wins on every objective. What an analysis can show is where the current plan sits, and which trade it makes without anyone having chosen it. The weighting also shifts at peak, when waiting in a crowded aisle starts to cost more than walking.
How to re-slot without a shutdown
There are four ways to improve a slot plan, and only one of them is a project. A full re-slot gives every SKU its best location from scratch. It needs a shutdown window or a quiet week, and it starts to drift the day it goes live unless ongoing re-slotting keeps it current.
Ongoing re-slotting scores every possible move by the walking it saves and keeps only the top ten or fifty. One operator walks that list at the end of a shift, and next week the list is rebuilt from fresh order data.
The other two cost almost nothing extra. Every inbound item is a slotting decision, so its location can be chosen on data instead of by whoever holds the scanner. And the locations need housekeeping: merge a slow mover spread over three pallet positions and you get two positions back.
Simulate the gain before anything moves
A slot change is physical, so it pays to see its effect first. A digital twin of the building, built from your layout, your locations and your order history, scores the moves and projects the change in picks per hour before anyone touches a pallet.
Our own modelling puts a ten-move list at roughly 3 to 4% more picks per hour. The curve flattens as the list grows, because the first moves do most of the work, so capping the list costs little. Our guide to smarter warehouse slotting covers the full method.
Batching and routing: two decisions, solved together
Order clustering, a form of batching, decides which orders share a trolley. Grouping orders whose picks sit in the same aisles raises pick density, so one walk to the far aisle serves several orders instead of each order making its own trip. The grouping has to respect trolley capacity, weight limits, cut-off times and order priorities.
Pick path optimization decides the sequence. It calculates the shortest route through each tour and follows the floor's movement rules: one-way aisles, cross-aisles and blocked zones. A WMS that cannot see the distances between locations falls back on a fixed snake path through every aisle.
Solved one after the other, the two decisions limit each other: a tidy batch can force a long route, and a short route can leave the trolley half used. Solving them together avoids both, which is why batching and routing belong in one calculation. The MedTech case in the next section added them in sequence, clustering first and route optimization later.
How do you increase picks per hour without robots or a new WMS?
Keep the WMS and add an optimization layer on top of it. The WMS carries on running stock, locations and transactions, while the layer takes over three decisions the WMS makes with fixed rules: which orders travel together, in what sequence they are picked, and where the fast movers live.
Automation can lift output further, but it takes capital, lead time and a changed layout. What follows is what to do first, or what to do when robots are not on the table. An operation that is already well batched and well slotted is also a better starting point for automation later.
What a WMS does, and where it stops
A WMS is built to record and execute. Its picking rules group orders by arrival time or order number and sort lines by location sequence, but they do not weigh today's orders against the real layout.
Warehouse optimization software fills that gap. Optioryx Pulse, our own product, is one example. It sits on top of the existing WMS, calculates the batches and the routes, and hands the result back to the WMS for execution.
Two measured results
A MedTech distribution center picks roughly 30,000 order lines a day with 60 operators, all on foot with trolleys. Operators used to choose by hand which orders to combine in the WMS, and the pick sequence followed a general serpentine route.
With Pulse clustering the orders onto each trolley, the first measured result was 15% higher picking productivity, worth up to 9 FTE in the manual picking organization. Six weeks later, with route optimization and double-stacked trolleys added, the figure stood at 27% above the period before any optimization software.
The job on the floor stayed the same, with no retraining and no new equipment. The trolleys were optimized in the background, before they reached the picker.
Bleckmann, a 3PL for fashion and lifestyle brands, started working with us in November 2022. The setup generates a route for each pick list as soon as the picker is ready to start walking, alongside the existing WMS.
The two numbers measure different things. Bleckmann's 15 to 20% is walking distance, from routing, against non-optimized routes. The MedTech 27% is picking productivity, from clustering, routing and fuller trolleys together.
Put through the formula in the calculator section, a 15 to 20% shorter walk at a 50% walking share is worth 8 to 11% in pick rate.
When this will not help
An optimization layer shortens the walk, so where there is little walk to shorten, the gain is small:
- Goods-to-person and other automated sites, where the items travel to the picker.
- Small teams. Our rule of thumb puts the floor at around 10 pickers.
- Sites where travel is already a small share of picking time. Replenishment, pick face design or the confirmation step come first there.
- Simple grid layouts with few path choices, and sites that already batch by hand with care. Both land at the low end of any range.
- Operations with heavy clustering restrictions. The fewer orders may share a trolley, the less there is to bundle.
You can check most of this before buying anything. Pulse runs on uploaded order history in the web app, with no WMS connection, and shows the optimized routes and KPIs next to the current ones. Live operation does need the integration, so the test on your own data comes first and the integration after.
How do you calculate the ROI of picking optimization?
The value of a shorter walk is picking hours × walking share × walking reduction × loaded hourly cost. On the sample day, Bleckmann's measured 15 to 20% applied to 30 walking hours frees 4.5 to 6 hours a day. At $30 an hour that is $135 to $180 a day, or $33,750 to $45,000 over 250 working days for one zone, on made-up inputs.
The sum assumes that walking time falls in line with walking distance. It also only holds if the freed hours are used: they turn into money when they absorb more volume or replace overtime and agency hours.
The order picking ROI calculator runs the same kind of sum with your own picking staff, FTE cost and daily picks, and returns a yearly saving and a payback estimate.
Questions?
There is no single good pick rate, because published figures count different work. One survey reports about 71 picks per hour without defining a pick, while ASCM expects 120 to 175 pieces or cases per hour. A more useful benchmark is your own baseline in order lines per picking hour, tracked per zone, shift and order type.
Picking accuracy is lines picked minus lines with an error, divided by lines picked, times 100. A shift with 4,800 lines and 24 errors scores 99.5% at line level. The order-level version divides error-free orders by total orders and gives a lower figure for the same shift, so an accuracy figure needs to say which level it uses.
Most of what makes a pick fast is decided before the picker starts walking. The slot of each item, the orders grouped on the trolley, the pick sequence and the timing of replenishment set the distance and the waiting. The picker controls the pick and the scan, which together are the smaller share of the time.
Take out walking and searching time. Travel takes about 55% of picking time according to a 2023 study in the journal Axioms, and it is set by slotting, batching and routing before the tour starts. The pick and the scan confirmation keep the time they need, so accuracy is not traded for speed.
Yes. An optimization layer on top of the existing WMS can take over batching, routing and slotting decisions while the WMS keeps executing. A MedTech distribution center picking 30,000 order lines a day raised picking productivity by 27% this way, and Bleckmann cut walking distance by an average of 15 to 20% compared to non-optimized routes.