Smart warehouse / Article
Goods-to-Person vs Person-to-Goods: The Picking Model That Fits Your SKUs
Goods-to-person picking means the picker stands still and the system brings the tote to them. Person-to-goods means the picker walks to the item. The choice is decided by one measurement on your own floor: how much of a picker's shift is travel. In a conventional warehouse that figure sits around 60%, which is why goods-to-person stations reach several times the pick rate of a person with a trolley. But it only pays where order lines are many and units per line are few, and where volume keeps the station fed. This article gives you the test, the numbers on both sides, and the cases where person-to-goods is still the correct answer.
The numbers at a glance
01 / The definition
Two models, and only one of them walks
Person-to-goods is the model almost every warehouse starts with. A picker gets a list, walks the aisles, and collects items location by location into a trolley, a cart or a cage. Every variation of it, paper pick lists, handheld scanners, pick-to-light, voice picking, batch and zone picking, is an attempt to make that walk shorter or more accurate. The walk itself stays.
Goods-to-person inverts it. In goods-to-person automation, robots bring products to a mostly stationary associate, nearly eliminating walking and unproductive time, while in person-to-goods, associates travel through the aisles to find and pick individual SKUs. The storage system retrieves the tote, a robot or conveyor delivers it, the operator picks what the screen asks for, and the tote returns to storage.
That is the whole architectural difference, and everything else follows from it. Goods-to-person converts a labour problem into a machine and software problem. It buys pick rate and accuracy, and it costs capital, floor discipline and a control layer that has to keep every station fed. Person-to-goods keeps flexibility and low capital, and pays for it in shift hours spent walking.
Why the model matters more than the hardware
Vendors will sell you a shuttle, a bin robot or a station. None of those is the decision. The decision is whether your operation should be built around a stationary picker or a moving one. Get that wrong and no amount of good hardware fixes it: a goods-to-person station that idles is worse than the trolley it replaced, and a person-to-goods floor that has outgrown itself burns labour every shift no matter how good the scanners are.
02 / The deciding number
Measure travel time before you measure anything else
Start with why picking is worth this much attention at all. The peer-reviewed literature is blunt about it: order picking has long been identified as the most labour-intensive and costly activity for almost every warehouse, with its cost estimated at as much as 55% of total warehouse operating expense. More than half your warehouse cost sits in one process, which is why the picking model is a bigger decision than the racking.
There is one number that decides this, and it is not throughput. It is the share of a picker's paid time spent moving without picking. Industry accounts put it high: pickers spend approximately 60% of their day travelling and only 40% actually picking, and travel time is the single largest controllable waste variable in manual distribution centres. Other operators report similar ranges, with associates spending as much as half of their work time walking during order picking.
Removing that travel is the entire economic case for goods-to-person. So the case is only as strong as your travel share. Measure it before you shortlist a system: follow two pickers for a full shift, log time in three buckets, travelling, picking, and everything else. If travel is 55% or more, goods-to-person has a real target to attack. If it is 25%, the machine has almost nothing to remove and the money belongs somewhere else.
That is also why the pick-rate comparisons look so dramatic. In a conventional person-to-goods environment, skilled pickers typically achieve between 60 and 120 lines per hour, with travel consuming as much as 50 to 70% of each picker's shift. Take the travel away and the same person, doing the same physical picking motion, produces far more lines. Goods-to-person systems are reported at 300 to 600 units per hour against 50 to 100 in traditional systems, and cart-based person-to-goods automation lands in between, around 100 to 200 units per hour, while goods-to-person solutions generally average around 350.
Read those ranges as directional, not as a promise. The top of a vendor range is a peak-hour figure from a tuned installation with a favourable order profile. Your planning number comes from your own lines per order, units per line and station design, which is what simulation is for.
03 / The models side by side
Four ways to pick, and what each one is for
Most operations are not choosing between two options but four, because both models have an automated middle step. Find your order profile in the last column.
| Model | Who moves | Indicative rate | Capital | Flexibility | Fits when |
|---|---|---|---|---|---|
| Manual PTG | Picker walks, trolley or cage | 50 to 120 lines per hour | Lowest | Highest, change the list and you are done | Low volume, bulky items, narrow SKU range |
| PTG with cart robot | Picker walks less, robot carries and meets them | 100 to 200 units per hour | Moderate, no rack rebuild | High, redeployable | Growing volume, existing racking, no appetite for a rebuild |
| Goods-to-person | Tote moves, picker stays | 300 to 600 units per hour | Highest | Lower, the layout is the system | Many order lines, small parts, sustained volume |
| Hybrid split | Both, by SKU class | Per zone | Staged | High if the control layer is right | Fast movers plus a long slow tail, which is most floors |
Scroll the table sideways on a phone / Rates are published industry ranges, not a CODETRACE guarantee
Notice that flexibility runs opposite to rate. That is the real trade, not cost against speed. A goods-to-person installation is a commitment to a layout and an order profile. A trolley is a commitment to nothing. The middle row exists because many warehouses need more rate without that commitment, which is the same argument made in AMR against AGV against conveyor: infrastructure buys throughput, onboard intelligence buys the ability to change your mind.
04 / When person-to-goods wins
Sometimes the trolley is still the right machine
An article about goods-to-person that never argues against it is a brochure. Here is the honest set of conditions where person-to-goods stays correct, and where we will tell you so rather than quote a system.
That last one deserves emphasis, because it is the cheapest win in this whole subject. Operational data attributes 30 to 40% of pick walk time to poorly slotted SKUs, with good slotting cutting that waste by 40 to 60% and combined slotting plus route optimisation producing total walk-time reductions of 35 to 60%. Slotting costs analysis and discipline, not capital expenditure. Fix it, run the shift again, measure travel share again, and then decide what is left for a machine to solve. Plenty of warehouses discover the remaining problem is much smaller than the one they were quoting for.
05 / When goods-to-person wins
Many lines, small parts, and a shift that never lets up
The profile that makes goods-to-person obvious is the inverse of the list above. Orders with many lines and few units per line. Thousands of small SKUs. Sustained volume through the day, ideally across more than one shift. A pick face so large that the walk between two items in the same order is measured in tens of metres. E-commerce fulfilment is the textbook case, but so is spare parts distribution, electronics kitting and any production line that consumes hundreds of distinct components.
Accuracy is the benefit that gets undersold. A stationary operator working a guided station makes fewer mistakes than a person reading a list in an aisle, because the system presents one tote and asks for one quantity. Reported accuracy rates reach 99.9%, with workstations guided by screens, indicator lights and colour cues, and training time compressed from weeks to as little as 20 minutes because new hires do not have to learn the warehouse layout. In a labour market where you are recruiting and retraining constantly, that last point is worth as much as the pick rate. It is the same argument as the one behind automating around a labour shortage: the constraint is not just how fast people work, it is how quickly you can get a new person productive.
Density is the third benefit. Because nobody walks into the storage block, the storage block does not need walkable aisles. That is how a tote handling system can hold far more in the same footprint, which is the subject of the comparison of ASRS types, where CODETRACE tote handling runs in an 850mm aisle against the 3.5 metres a forklift needs.
06 / The handover
The tote handover is where real throughput is won or lost
Retrieval gets the attention. Handover decides the outcome. In the CODETRACE configuration, goods-to-person retrieval pulls the tote from storage and a floorbot carries out the handover to the picking station, so the tote arrives at the operator without anyone walking into the storage aisle. Simple to describe, and the part that most often limits the system in practice.
The reason is queueing. A station has a pick cadence. If totes arrive slower than that cadence, the operator waits and your expensive station runs at trolley productivity. If they arrive faster, totes back up and the buffer becomes the constraint. Three variables have to be sized together: retrieval rate from storage, floorbot fleet size and travel distance, and the number of stations. Get one wrong and the other two are wasted.
Peak matters more than average here. A floor that averages 300 lines an hour but takes 700 in the two hours before a cut-off needs to be designed for 700. This is the same lesson the storage side teaches, where shuttle systems miss modelled capacity when too many vehicles compete for one lift. Averages size nothing. Peaks size everything.
What we need from you to size it
A month of order lines with timestamps, the SKU velocity profile, units per line, and your cut-off times. From that we can model station count and fleet size honestly, including the case where the answer is fewer stations than you expected. Station count and fleet size are outputs of that model, not fixed specifications, which is why we will not quote either before we have seen your data.
07 / Match it to your SKUs
Your order profile picks the model, not your ambition
Run your own SKUs through this. It takes an afternoon with a spreadsheet of order lines and it will tell you more than any vendor visit.
Row six is a prerequisite, not a criterion. Automated retrieval trusts your records absolutely: it goes to the location the data names and brings back whatever is there. If your stock count is wrong, goods-to-person industrialises the error. Fix the data layer first, or fix it as part of the same project with the count and the tracking built in.
The hybrid answer is the most common outcome of this exercise. Head of the curve into goods-to-person, tail left manual, bulky items on pallets, one control layer scheduling all three so an order can draw from any zone and consolidate at pack. It is less tidy than a single-model warehouse and it usually costs less and performs better.
08 / On your floor
We study the floor first, then design and simulate
We do not sell hardware on day one. A picking model is a decade-long commitment to a layout, so the sequence is a floor study first: travel share measured on a real shift, order lines with timestamps, the SKU velocity curve, unit sizes, cut-off times and where the queue forms today. Then a design, then a simulation against your own order history, and only then a build.
Simulation is where the cheap discoveries happen. That the third station adds nothing because the lift is the ceiling. That fixing slotting removes a third of the walk before any capital is spent. That a cart robot on the existing racking gets you two years of headroom at a fraction of the cost, which is often the right answer for a Malaysian operation growing quickly but not yet sure of its volume floor.
CODETRACE integrates on site from Shah Alam in Selangor and Batu Kawan in Penang, so the team that models your picking is the team that commissions the stations and the team you call in year three. CODETRACE is a member of the NVIDIA Inception program. Send us a month of order lines, and we will tell you which model your floor needs, including when the answer is better slotting and no new machine at all.
Measure the walk. If it is most of the shift, stop the walking. If it is not, fix slotting instead.
FAQ / Goods-to-person picking
Questions, answered.
01What is goods-to-person picking?
Goods-to-person picking means the picker stays at a station and the storage system brings the item to them. A storage and retrieval system pulls the tote or carton from its location, a robot or conveyor delivers it to the workstation, the operator picks the quantity the screen asks for, and the tote goes back. The picker never walks to a location. Person-to-goods is the opposite and the traditional model: the picker travels the aisles with a list, a trolley or a cart, and collects items location by location.
02How much faster is goods-to-person than manual picking?
Published ranges put manual person-to-goods picking at roughly 50 to 120 lines or units per hour per picker, and goods-to-person stations at roughly 300 to 600 units per hour, with vendor case studies claiming more. Treat the top of any range as a peak-hour figure from a well-tuned installation, not a planning number. The honest way to read it is directional: goods-to-person removes travel, and travel is most of the shift, so the multiple is large. Your own number depends on order profile, lines per order and how well the station is designed.
03How much of a picker's day is spent walking?
In a conventional person-to-goods warehouse, travel consumes roughly half to two thirds of the shift. One industry account puts pickers at approximately 60% of the day travelling and 40% picking, and travel time is described as the single largest controllable waste variable in manual distribution centres. That is the number that decides whether goods-to-person pays on your floor. If your pickers walk very little because the pick face is compact and dense, the gain from removing travel is small and the investment is hard to justify.
04When is person-to-goods still the right choice?
When order lines are few and units per line are many, when the product is bulky or heavy enough that pallet-level handling beats piece picking, when the SKU range is narrow and stable, and when volume is too low to keep a station busy. A goods-to-person station only pays when it stays fed. A quiet station is expensive floor space. Many operations get most of the available gain from better slotting, batching and a cart-based robot instead, and should spend the capital there first.
05Should I fix slotting before buying automation?
Almost always yes. Slotting is where a large share of walk time comes from, with operational data attributing 30 to 40 percent of pick walk time to poorly slotted SKUs and good slotting cutting that waste by 40 to 60 percent. Slotting costs analysis and labour, not capital. If you automate on top of a bad slotting plan you buy a system that retrieves the wrong things quickly. Fix the cheap problem first, then measure again, then decide what is left for automation to solve.
06How does the tote reach the picking station?
In the CODETRACE configuration the storage system retrieves the tote and a floorbot carries out the handover to the goods-to-person station, so the tote arrives at the operator without a person walking into the storage aisle. The handover is the part that decides real throughput: if the tote queues, the station idles, and the pick rate on the brochure never appears. That is why the station count, the floorbot fleet size and the retrieval rate have to be sized together against your own order data.
07Can I run both models in the same warehouse?
Yes, and most operations should. The usual split is goods-to-person for the fast-moving small parts that generate most of the order lines, and person-to-goods for bulky items, full-carton picks and the slow tail where automated retrieval would sit idle. One control layer schedules both so an order can draw from either side and be consolidated at pack. CODETRACE designs and simulates that split against your order history before anything is installed.
Sources / Every figure in this article
Where the numbers come from
Sources are listed by what they are rather than by brand name. Only the first entry is peer-reviewed. The pick-rate ranges come from companies that sell picking systems, so treat the top of any range as a tuned installation on a favourable order profile, not as your planning number. Anything in this article that is CODETRACE judgement rather than a measured figure is written as judgement.
Related / Smart warehouse