Home / Work / Stores Live in Seven Days
CASE STUDY · MANUFACTURING · INVENTORY & ERP

Asked to go faster.
Stores live in seven days.

A UAE manufacturer that has just moved from R&D into production — composite, metal and electronic parts. Nobody could say what was in stock or where it was. When the CEO asked us to accelerate, the stores went live seven days later, about four weeks ahead of plan, without skipping a single rehearsal.

Client
UAE manufacturer, R&D into production
Parts
Composite, metal and electronic
Scope
Stores, purchasing, drawing vault, handhelds
Hosting
On-premise, the client’s own servers
Planned go-live
14 December
Actual
Stores live 16 November
A rugged handheld scanner aimed at a parts bin in a workshop storeroom
7
Days from request to live stores
257
Engineering parts live
108
Coded store locations
19/19
Changes rehearsed, replays matched
/ 01
The Problem

Moving from R&D into production changes what a stores has to do. Prototypes can be built from whatever is on the shelf; a production line cannot.

Nobody knew how much stock there was, or where, across several storage places. The same part went by different names. Urgent parts bought on petty cash never reached the stock records at all. Purchase approvals by the CEO were on paper. Shortages were found when a build was already waiting for the part.

Engineering IP had to stay in the building, so everything had to run on-premise, on the client’s own servers.

/ 02
The Approach

Discovery on 22 September, kickoff on 1 October. The engineering drawing vault (PDM) server was in service on 11 October, and the ERP was installed on the client’s own test server on 4 November. The plan said go-live on 14 December.

Around 9 November the CEO asked us to accelerate. The baseline stock count ran on 15 November and the stores went live in production on 16 November — seven days after the request. Handhelds, offline and locked down, and purchasing followed on 18–19 November.

The data came first. Every part got an 8-digit number — a class code plus six digits — and a noun-first name, across 18 part classes. We analysed 225 historical purchase orders and normalised their 848 order lines into 719 items, so the catalogue started from what the company actually buys. Every store place got a code: 108 of them, with 225 barcode and QR part labels. Serial or batch tracking is set by how critical the part is, and every issue from stores is charged to a project.

Purchasing runs on 108 vendors. Purchase orders print in the company’s own layout and terms, each order carrying its own terms and liquidated damages. The signatory is chosen by order value, and no order is confirmed until the signed copy is attached.

/ 03
The Hard Parts
  1. Going faster without getting looser Every change was a numbered plan, rehearsed on a scrubbed copy of the database, approved, then applied to test and to production with a replay check. Nineteen plans; every replay matched. One security update broke image handling — caught in rehearsal and rolled back before it reached production. Passwords and keys never appeared in a plan or a chat, and the server change log was written the same day.
  2. A scanner that works without a connection The handheld app, built in Flutter, runs on Zebra rugged Android scanners. Scan a bin, then Take out (charged to a project), Put in, Move or Count. It works fully offline and syncs in one step. Each movement is recorded once however many times it is re-sent, so a retry can never double a stock figure. Kiosk mode with a supervisor PIN keeps the device on the job, and the app carries the client’s brand.
  3. Turning a scan into a proper stock movement A custom server module turns scans into stock transfers, receives goods against a purchase order or against petty cash — so emergency buys land in stock like any other — prints labels and runs counts place by place. It is covered by 50 server tests and 31 app tests.
  4. One part, one name The same part under three names is three stock figures, all wrong. The numbering and naming standard — class code first, noun first — had to be strict enough to stop duplicates before they reach the shelf.
/ 04
The Outcome

The stores went live on 16 November, about four weeks ahead of the original plan, with 257 engineering parts each carrying one number, one name and a coded place. Handhelds and purchasing were ready by 19 November.

Purchasing is controlled in the system: the signatory is set by order value, and no order is confirmed without the signed copy attached. Petty-cash purchases can now be received into stock like any other delivery.

Training was delivered as narrated videos — 13 chapters, 58 steps — and a 41-page guide, and everything runs on the client’s own servers.

Stack ERP implementation Inventory & purchasing Flutter Zebra rugged Android Barcode & QR labelling Offline sync Idempotent movements PDM drawing vault On-premise Rehearsed change control

Faster is only safe
when every change is rehearsed first.

BEFORE THE BUILD WAITS

Find the shortage in the stores,
not on the shop floor.

Contact Us →