Introduction

Most peak plans are traffic plans. Load tests in September, autoscaling configured, CDN capacity checked, a code freeze two weeks before Black Friday. All of that is sensible, and most of the time it works. The site stays up.

And yet, on the Monday after Black Friday, customer service is buried in "where is my order?" tickets.

The reason is simple: your website scales. Your fulfillment network doesn't. Software capacity is elastic. You add servers, and the bottleneck disappears. Fulfillment capacity is not. A store has a fixed number of people who can pick, a fixed amount of stock, a fixed amount of packing space and a fixed number of carrier pickups. When peak breaks, it breaks where elasticity ends: in the layer between the order and the location that has to fulfill it.

To see how that happens, follow one order through one evening.

‍

Black Friday, 8:04 pm

The scenario below is a composite, not a specific retailer. But anyone who has run peak operations will recognize every step.

A customer orders a winter jacket and two pairs of gloves. The system shows both items in stock at a store four kilometers from her home. The order is routed there. At checkout, she sees: delivery on Tuesday.

So far, nothing has gone wrong. The site is fast, the payment goes through, the confirmation email arrives. On every dashboard that tracks the front end, this is a successful order.

‍

Break 1: The stock in the system isn't the stock on the shelf

At 8:40 pm, a picker goes to the rail. The jacket in size M isn't there. Maybe it was sold at the till an hour ago and the stock update hadn't come through yet. Maybe it ended up on the wrong rail during the afternoon rush. Maybe it was never there.

On a normal Tuesday, this is a minor annoyance. Someone finds a substitute or the order gets moved, and nobody outside the store notices. At peak, the same gap hits hundreds of orders in parallel, and each one needs a new home within minutes.

Inventory accuracy is the least glamorous topic in commerce, and the one everything else depends on. In a recent piece on store fulfillment at Portuguese grocer Continente, missing items on the shelf and gaps between system and reality are named as typical failure points, with the conclusion that fast delivery promises depend on clean store processes rather than the last mile.

But our order has already left the store. It is being rerouted.

‍

Break 2: The network has capacity. The node doesn't.

The routing rule says: next nearest store with stock. That's the flagship in the city center. It is also the next nearest store for most other orders in the area that the first store couldn't fill. And it was already the busiest location in the region before any of them arrived.

By 10 pm, the flagship's pick queue is hours behind. Meanwhile, two stores on the edge of town have the same jacket in stock and staff with time on their hands.

Nothing here is a system failure. The rule did exactly what it was set up to do, in October, based on average load. Static routing rules treat a location's capacity as a constant. At peak, it changes by the hour. A rule that doesn't ask whether a location can handle one more order doesn't spread the load. It concentrates it.

This is why the store's role in fulfillment is being rethought. EHI's study on retail logistics in 2036 argues that a store's future role will depend on how well it is connected to data and logistics networks, not on its assortment alone. A store whose capacity the network can't see is a store that will eventually be overloaded.

‍

Break 3: The promise doesn't know any of this

The customer still sees "delivery on Tuesday". That promise was calculated at checkout, from carrier transit times and the assumption that the store would pick within its usual window. It knew nothing about the missing jacket, the reroute or the queue at the flagship. And nothing updated it afterwards.

The jacket ships on Thursday. The gloves, still with the original store, ship separately. Two parcels, two shipping costs, one late delivery. On Wednesday, the customer opens a ticket. On Friday, she writes a review.

This is the moment an operational problem becomes a trust problem. It surfaces days after the actual failure, which is why it rarely gets traced back to its cause. In the post-peak review, it shows up as "customer service volume", not as "a routing rule that ignored store capacity".

‍

Why the chain matters more than the links

Each of these breaks is manageable on its own. A stock mismatch can be fixed. A busy store can be relieved. A late promise can be corrected with an honest update. What makes peak dangerous is that the breaks are coupled. The stock mismatch triggers the reroute. The reroute overloads the store. The overloaded store breaks the promise. And because peak means this happens hundreds of times in parallel, small inaccuracies stack up into a network-wide problem. The traffic graph looked great all night.

‍

What a network that holds at peak looks like

If peak breaks in the order layer, that's also where it has to be held. Three things make the difference.

• ‍Stock you can trust in real time. Not a nightly sync, but availability that reflects what was sold at the till, reserved online or returned in the last few minutes. Every decision further down the chain is only as good as this number.‍
• Routing that knows capacity. Orders go to locations based on what each one can handle right now, not only on distance and stock. Load gets spread before a store tips over, not after its queue is already hours long.‍
• Promises calculated from reality. A delivery date that takes the actual state of the network into account, and gets updated when that state changes. A customer told on Saturday that her parcel will arrive Thursday is far less likely to open a ticket than one who finds out on Wednesday.

None of this is new thinking. What is new is that it's becoming the foundation for everything else. As AI agents move into shopping and operations, they rely on exactly this data. An agent that makes promises based on wrong stock simply fails faster and at larger scale. Gartner expects that by 2030, only 5% of organizations will make at least 10% of their supply chain planning decisions autonomously, and points to data quality, system architecture and organizational maturity as the reasons. Peak is where those foundations get tested first.

This is the layer we work on at fulfillmenttools: the operational ground truth between order and location.

‍

[cg_inject-el="card"]