Paperless picking: how a warehouse scanning app cuts delivery errors by 80%

Right order, wrong box

In a warehouse shipping 150-400 orders a day off printed pick lists, the line-level error rate almost always lands between 1.5% and 3%. That sounds small. At 900 lines picked daily it means 13-27 wrong lines every day: a neighbouring part number, a quantity counted by eye, the expiring lot left on the shelf, a line ticked on paper but never loaded on the truck.

What a single wrong line actually costs:

• return freight plus re-delivery: 12-25 EUR
• 25-40 minutes of admin work — credit note, new delivery note, the call with the customer
• goods that come back damaged or unsellable: in 5-15% of cases
• stock drifting out of sync in the ERP, discovered only at month-end count

At 20 wrong lines a day, the direct cost sits around 90,000-150,000 EUR per year. The indirect cost — the customer who quietly moves 30% of their orders to another distributor after the third return — shows up in no report, but is usually the bigger number.

The point worth making: this is not a problem of careless people. A picker walking 180 lines per shift, holding part number, quantity and location in their head, ticking a sheet of paper that cannot answer back, will make mistakes. Paper has no way of telling them they grabbed the wrong item.

What a scanner-based picking app actually does

It is not a full WMS. It is a thin layer on a handheld device that validates every action at the moment it happens:

1. Assigns the work. Orders are grouped into waves — by carrier, by zone, by cut-off time — and allocated automatically, instead of being shouted across the floor.

2. Dictates the route. Lines are sequenced by physical location, not by the order they were entered in. That alone cuts walking distance by 20-35%.

3. Confirms the location. The shelf label is scanned before the product. This catches the case where the same item sits in three locations and the picker goes to the wrong one.

4. Confirms the item. The EAN or internal code is scanned. If it does not match the line, the device will not move on. This is where confused-part-number errors disappear.

5. Confirms the quantity. Entered manually or scanned N times. For high counts, validate by weight or by case, not by piece.

6. Captures lot, serial and expiry where it matters. For food, cosmetics or medical devices, the app enforces FEFO and refuses a longer-dated lot while a shorter-dated one is in stock.

7. Validates packing. At pack-out the contents are scanned once more, the SSCC label and carrier label are generated, and shipping an incomplete order is blocked.

8. Writes into the ERP. Stock movements and the delivery note flow into the ERP in real time or in batches of a few minutes — not by re-typing paper at the end of the day.

Point 8 is where most projects fall over. Scanning is worth nothing if the result gets keyed into the ERP manually in the evening.

Why printed lists and the WMS module in your ERP don't fix this

The printed list has a structural flaw: validation happens after the fact — at the dispatch check, or at the customer's door. A 100% check at the loading bay costs 2-4 minutes per order and catches 60-70% of mistakes, not all of them.

The WMS module inside an ERP typically fails for different reasons:

• It is built for desktop. On a 4-inch handheld, a form with 14 fields is unusable with gloves on.
• It assumes permanent connectivity. In warehouses full of steel racking, Wi-Fi coverage has holes; the app has to work offline and sync when it reconnects.
• Per-user licensing in an environment with high turnover and seasonal staff turns into a cost problem.
• It imposes a slotting model you do not have: many real warehouses run on semi-fixed zones, not strictly managed bin locations.

The right decision comes from a simple test: if your ERP already holds accurate stock locations and exposes a movements API, extend what you have. If it does not, build a thin mobile app on top of an integration layer and let the ERP remain the source of truth for stock.

Case study: electrical components distributor, 1,100 lines a day

A company with 62 employees, 11,000 SKUs and two warehouses was picking off printed lists with a visual check at the bay. Measured over 6 weeks before the project: 2.4% wrong lines, 38 monthly returns caused by picking errors, 41 minutes average per order from allocation to dispatch.

What we implemented together with the team at NEXVA SYSTEM:

• An offline-first Android app on 14 handhelds, syncing every 30 seconds with a local queue when signal drops
• An integration layer between app and ERP for stock reservations, movements and delivery note generation — no direct writes into the ERP database
• Physical labelling of 2,800 locations, plus a labelling station for the 19% of SKUs that had no usable barcode
• Zone-optimised routing and wave grouping by carrier
• Pack-out validation that blocks incomplete orders
• A pilot in one zone for 5 weeks, then rollout across the warehouse

Results after 4 months:

• Wrong lines: from 2.4% to 0.45% — an 81% reduction
• Monthly returns from picking errors: from 38 to 7
• Average time per order: from 41 to 26 minutes
• Bay checks: cut from 100% to a 15% sample, with no increase in errors reaching customers
• Inventory accuracy: from 94.1% to 99.2%, with cycle counts replacing a full stock freeze
• Training a seasonal picker: from 3 days to 4 hours — the device tells them what to do

What it costs and when it pays back

| Component | Cost |

|-----------|------|

| Flow analysis, location mapping, barcode inventory | 2,500-4,500 EUR |

| Offline-first mobile picking app | 8,000-14,000 EUR |

| ERP integration: reservations, stock movements, delivery notes | 5,000-9,000 EUR |

| Packing, SSCC labels, carrier integration | 3,000-6,000 EUR |

| Rugged handhelds (400-700 EUR each) or phones with ring scanners (180-260 EUR per set) | 2,500-10,000 EUR |

| Physical labelling of locations and unbarcoded SKUs | 1,500-4,000 EUR |

| Hosting and maintenance | 200-450 EUR/month |

Payback is calculated on three lines, in order of size.

First, errors: 900 lines a day, 250 working days, dropping from 2.4% to 0.5% means 4,200 wrong lines avoided per year. At a conservative all-in cost of 20 EUR per line, that is 84,000 EUR a year.

Second, productivity: 15 minutes saved per order across 220 orders a day is 55 hours daily only if every minute had been fully occupied. In practice it translates into 3-5 pickers you no longer need to hire as volume grows — 45,000-75,000 EUR a year at a loaded cost of 15,000 EUR per person.

Third, inventory: five percentage points of accuracy means fewer phantom stockouts, fewer emergency purchase orders, and the end of the two days a year the warehouse shuts down for a full count.

On a total budget of 25,000-45,000 EUR, payback in the cases we have seen arrives in 4-8 months.

The mistakes that ruin it

• Going live before labelling is finished. If 15% of locations have no label, pickers will look for a way around scanning — and they will find one.
• The "continue anyway" button. An override added for exceptional cases ends up used on 30% of lines within two months. Exceptions belong behind a reason code and a report, not a free button.
• Duplicate barcodes. The same EAN on two different SKUs turns validation into noise. Clean this up before, not after.
• Ignoring radio coverage. A warehouse with metal walls has dead spots. If the app cannot run offline, pickers are back on paper inside a week.
• No baseline. Write it down now, over 4 weeks: percentage of wrong lines, returns caused by picking errors, minutes per order. Without the before numbers you can prove nothing later.
• Starting across the whole warehouse. A pilot in one zone with 3-4 pickers surfaces every process problem within two weeks — the same problems that would otherwise halt shipping.

How to start

1. Measure your real error rate over 4 weeks, from returns and complaints, split by cause. Most companies estimate it at half of what it actually is.

2. Inventory your barcodes. Export the SKU list and check a 200-item sample for a scannable code. The share that is missing determines your labelling cost.

3. Check what your ERP can write through its API: reservations, stock movements, delivery note generation. If there is no API, the integration layer becomes phase one.

4. Pick a pilot zone — ideally the one with high volume and easily confused SKUs, not the easiest one.

5. Run the pilot for 4-5 weeks with mandatory validation and reason codes for exceptions. Only then roll out and reduce the bay check.

The difference between a warehouse at 2.4% errors and one at 0.45% is not how attentive the people are. It is that the second one moved validation from the moment of the complaint to the moment the product leaves the shelf.

Want to see what picking errors actually cost you, and what scan-based validation would change, using your own numbers? Book a free consultation.