.jpg&w=3840&q=75)
A wrong item is rarely one picker's mistake. By the time a customer opens a 250 g pack instead of the 500 g one she paid for, the error has usually passed through three or four earlier decisions: a catalogue entry that put two variants under one barcode, a putaway that slotted lookalikes side by side, a pick step that never asked anyone to prove what they had picked. The picker is simply the last person to touch it.
That distinction matters more in Quick Commerce fulfilment than almost anywhere else. A brand that moves from one central warehouse to a network of dark stores does not just add capacity. It copies every upstream decision into every node, and each copy can drift on its own. This piece maps where wrong items enter a distributed network, why the math turns against multi-item baskets, what one mis-pick costs after the refund, and which controls hold up when volume peaks.
The short answer. Wrong items get delivered when an identification error passes through the fulfilment system without a check designed to catch it. Warehouse researchers describe these as human-system errors rather than human errors, because the root cause usually sits in how the work is designed. In distributed fulfilment the risk compounds: every dark store inherits the same catalogue and slotting decisions, packs a dense assortment into a small footprint, and runs with its own team and its own habits.
Most wrong items start before the pick, even though the pick is where they become visible. A 2022 study in the International Journal of Production Research reviewed 46 papers and interviewed 38 order pickers from 14 companies in four countries. It identified eight design-linked failure modes, including mental fatigue, memory demand, complexity and vision, and the interviews surfaced two the literature had missed: communication and supervision. For an operator, the useful reading is that accuracy is designed upstream and only tested at the shelf.
In a dark store network, it helps to treat every order as passing through five gates, each with its own failure mode and its own owner.
Table 1: The five gates where a wrong item can enter a quick commerce order
| Gate | What goes wrong | What the customer receives | Control that catches it | Usual owner |
| 1. Catalogue and barcode master | Two variants share one barcode, or a SKU is mapped to the wrong product ID at onboarding | The right product family in the wrong variant, often repeatedly | Variant-level barcode audit before go-live | Brand, with partner validation |
| 2. Receiving and putaway | Cartons of similar variants are binned in each other's locations | The wrong variant until the bin is recounted | Scan-to-bin at putaway and receiving counts by variant | Fulfilment partner |
| 3. Slotting | Lookalike SKUs sit in adjacent bins or at the same eye level | A neighboring shade, flavor or pack size | Separate lookalikes by distance or level and flag them on the pick screen | Fulfilment partner |
| 4. Pick | The picker confirms a location but not the unit, or confirms from memory | The wrong item or the wrong quantity | Unit-level scan confirmation | Fulfilment partner |
| 5. Pack and handover | Two orders swap at a staging bin or packing station | Someone else's complete order | Order-level scan at pack and at rider handover | Fulfilment partner and last-mile team |
Gate 1 belongs to the brand, which is why it is so often missed. GS1's GTIN rules state that different sizes of a product must carry different GTINs, and name the consequence of ignoring this: the sizes become impossible to tell apart. A brand that reuses one barcode across a shade or flavor range has built a mis-pick into every dark store it ships to. Scan confirmation cannot catch it, because the scanner will accept the wrong unit as correct.
A central warehouse has one catalogue load, one slotting plan and one team whose habits a single manager can see. A dark store network multiplies all three, and four structural pressures follow.
Redseer and Google's 2026 quick commerce report describes a metro model in which a cluster of 3,000 to 5,000 sq ft dark stores gives each pin code access to 35,000 to 40,000 SKUs. That is a large assortment in small rooms, and in small rooms variants end up as neighbours. The same report counts the difficulty of exchanging a wrong shade among the friction points holding beauty and personal care back on quick commerce. Shade, flavour and pack-size ranges are precisely what D2C brands sell, so the categories most exposed to lookalike errors are the ones brands most want to move quickly.
In D2C fulfilment, a network-wide accuracy figure is an average of very different stores. A node that opened last month, a node that has just absorbed a slotting change and a node on its third store manager this year will not perform alike, and a blended number hides all three. We made the same point about on-time delivery when breaking down whether a fulfilment partner's SLA is actually good. It applies with more force to accuracy, because one weak store can generate most of a brand's wrong-item complaints while the network average still looks healthy.
Accuracy is learned on the job. In a 2025 laboratory comparison of picking methods, inexperienced participants picking from paper lists got 6.86% of picks wrong, while a participant with nine years of picking experience made no errors at all. The study used students and simple items, so its rates do not transfer directly to a dark store. The direction does. A network that hires ahead of a festive peak adds its least experienced pickers at the moment of highest volume, which is exactly when the system, not the individual, has to carry accuracy.
Pick accuracy is usually quoted per line. Customers experience it per order. If each line is picked correctly with probability p, an order of n lines arrives fully correct with probability p to the power of n, and the gap widens quickly.
Table 2: What per-line pick accuracy means at order level (illustrative calculation)
| Per-line accuracy | 1-item order | 3-item order | 5-item order |
| 99.0% | 99.0% | 97.0% | 95.1% |
| 99.5% | 99.5% | 98.5% | 97.5% |
| 99.9% | 99.9% | 99.7% | 99.5% |
Calculation assumes errors on each line are independent. Real errors are not, so use the table to translate a quoted per-line figure into customer terms, not as a forecast.
A partner quoting 99.5% line accuracy to a brand whose typical order carries three items is describing a customer experience in which roughly 15 orders in every 1,000 contain an error. At 1,000 orders a day, that is about 15 households a day, every day. It is also why perfect order rate, one of the fulfilment KPIs that actually predict repeat purchase, is so sensitive to basket size: bundles and multi-item offers raise order value and lower order accuracy at the same time.
The visible cost is easy to list: a reverse pickup, a second delivery, a unit that may not be resaleable and a support ticket. If the return then stalls at the store, the damage compounds, a failure mode we covered in why dark stores reject returns customers have already initiated. The larger cost sits in the inventory system.
When a picker sends variant B against an order for variant A, the system records A as sold. A's shelf count is now one higher than its record, and B's is one lower. One error has broken two records. B now shows stock the store does not have, so the next order for B can be routed to that node, meet an empty bin and end as a cancellation. In a study of nearly 370,000 inventory records across 37 stores of one retailer, 65% were inaccurate, and more complex store environments made inaccuracy worse. That research covered physical retail rather than dark stores, but a dense, fast-turning dark store is a close cousin of the complex environment it describes. In Quick Commerce logistics, a node-level stockout carries its own penalty, which we unpacked in why a product can vanish from the shelf after one stockout.
Then there is trust. In Redseer and Google's consumer interviews, one quick commerce user explaining why she still does her monthly stock-up offline said: “I got a wrong order once, and getting the refund was such a hassle.” One wrong order was enough to keep a ₹3,000-plus basket off the app. Wider surveys point the same way. A LocalCircles survey reported by Moneylife in January 2025, drawing more than 1 lakh responses, found that 48% of online shoppers had been stuck with a wrong product at least once in the previous 12 months. That figure spans all of e-commerce and blends genuine mis-picks with seller substitutions and restrictive return policies, so read it as a measure of customer exposure, not of dark store error rates.
The research agrees on one point: confirmation beats instruction. Telling a picker where to go saves search time. Making the picker prove what they picked is what catches the error.
In the same 2025 laboratory study, a light-guided method paired with a wrist scanner, which would not move to the next location until the correct one was scanned, produced errors on 1.18% of picks against 6.86% for paper lists. A 2018 study summarized in that paper found that adding scanner confirmation to paper picking cut errors from 161 to 59 across the same tasks, a reduction of roughly 63%. Guidance alone has limits: in the 2025 study, most pick-to-light errors came from participants misjudging which bin a light referred to.
For a dark store, that translates into five controls that map back to the gates above:
An accuracy figure without a definition is not a figure. Before comparing partners, or renewing one, get written answers to these six questions:
If you are moving between partners, the catalogue load is the highest-risk moment for gate 1 errors. Our migration runbook for switching fulfilment partners covers how to sequence it without breaking the delivery promise.
Zippee runs Quick Commerce fulfilment as infrastructure for a brand's own channel. Inventory sits in Zippee's dark stores, and pick and pack runs to the brand's own SLAs and SOPs. Across 100+ brands, 140+ dark stores and 21+ cities, that design choice is what lets a brand set the standard for its lookalike variants and hold the network to it, store by store. Delivery partners are full-time Zippee employees rather than gig workers, which keeps the final gate, handover, inside one accountable team.
The contrast with listing on a quick commerce platform is sharp. There, a brand's packing and verification SOPs become suggestions rather than standards, applied by an operation built around someone else's catalogue. On its own channel, running on Quick Commerce-as-a-Service, the brand writes the rule for every variant and can audit whether it holds. That is what Quick Commerce for D2C brands should mean: the speed of a dark store network with the standards still set by the brand.
Wrong items are rarely random and rarely one person's fault. They enter at five identifiable gates, they multiply with every node a brand adds, and each one leaves behind two broken inventory records that set up the next failure. The brands that fix this stop asking pickers to be more careful and start asking their network to prove every pick, at every node, against a definition agreed in writing.
If you're ready to turn your fulfilment into a competitive advantage, join our waitlist.