E-commerce manufacturing · Systems integration
80 orders a day through five systems, entered once
A UK manufacturer had been re-typing every order between five systems for six years. We connected them into one operational backbone.
THE PROBLEMFive systems touched every order. Each one was good at its own job. None of them talked to each other — so every order was copied by hand onto the operations board, checked against stock, turned into a production order, scheduled for delivery, and marked complete once it arrived. Roughly 80 a day, for six years.
SIX SYMPTOMS OF A MISSING BACKBONE
Does this sound familiar?
01.
The same order information is typed into more than one system.
02.
Someone checks three tools to answer “where's my order?”
03.
Stock figures in different systems regularly disagree.
04.
Orders are closed by hand, by cross-checking delivery reports.
05.
The ad platform optimises on checkout data that ignores cancellations and margin.
06.
The team spends the first hour of the day preparing work instead of doing it.
THE SYSTEMOne rule, agreed before anything was connected: each system keeps doing what it's good at, one place shows the full picture, and nobody types the same information twice.
How an order moves
The order arrives
01
Sent to the operations board the moment it's placed. Every product becomes its own line, and kits are expanded into the components the production floor actually works with. Orders that don't look right go to an investigation list instead of flowing silently into production.
Picking list ready at 5am
04
A scheduled workflow gathers the day's orders and builds the picking list, sorted. The warehouse team starts with the work already organised instead of spending the first hour assembling it.
Production from the board
02
Production orders live in the inventory system, but the team manages them from the board. Changing a status there releases or voids the matching production order and the real result is written back.
Delivery is scheduled and tracked
05
Own routes are planned in routing software; larger items go through a pallet carrier network. Both feed the same place, so nobody needs to know which carrier an order went with to find out where it is.
Materials are checked
03
Bill-of-materials requirements are calculated and grouped by planned date, then compared against on-hand stock, so the team can see whether a job is ready to start or will stall halfway through.
The order closes itself
06
Delivered orders are marked complete in the store and the proof of delivery is attached. The ad platform receives what actually happened to the order, margin included, rather than what the website saw at checkout.
FROM CHECKOUT TO DELIVERED
WHAT ACTUALLY MADE IT HARD
Connecting systems is the easy part
Keeping them connected through API changes, busy Mondays and human mistakes is where these projects break. Five things we had to get right.
01
The same order arriving twice
Webhooks get sent more than once, and an update can land before the original has finished processing. Either one creates a duplicate, and duplicates are how a team stops trusting the board. Every workflow that creates a record searches for it first, if it exists, it updates; if it doesn't, it creates.
02
A board that lies
When the system releases a production order, it writes the real outcome back. If the release failed, the board says it failed. A status that is confidently wrong costs more than no status at all, because the team stops checking it.
03
Busy days breaking the sync
Pulling a full day of deliveries in one go hits the routing system's rate limits and fails partway through, usually on the busiest day. Updates run in small batches with deliberate pauses. Slower on purpose. Timeouts retry themselves; problems a retry can't fix go to a person rather than being retried until they disappear from view.
04
Stock changes that shouldn't be automatic
A wrong stock figure costs far more than the few seconds it takes someone to confirm it. Proposed adjustments post to Slack with approve and reject. Nothing moves until a person says so. The automation moves information; it doesn't make judgement calls it isn't qualified to make.
05
The ad platform counting one sale five times
Each order is flagged once its conversion has been sent. Without that flag, every status change looks like a new sale and the platform optimises towards numbers that were never real.
1yr+
Running live in production, used every day
StoreOperations boardInventory & productionDelivery systemsAd platform
The test isn't launch day
Building something that works on the day you demo it is straightforward. The test is whether it's still standing after a year of API changes, busy periods and edge cases nobody predicted.
6 +
Hours of daily manual data entry no longer done by hand.
14 K
Item bill of materials calculated from a single approval
LET'S TALK
If this sounds like your setup
A call is the quickest way to find out whether you need one integration or a backbone. We’ll look at what’s actually happening in your setup and tell you straight whether there’s something worth fixing.