OZ Leather: one operating picture, one week on the floor.
CASE STUDY — Leather goods · Vietnam | MIDAS by NBR Intelligence
OZ Leather, a leather-goods manufacturer and exporter in Ho Chi Minh City and Hanoi, ran 1,000+ SKUs on roughly fourteen disconnected spreadsheets. NBR Intelligence forward-deployed an engineer into the workshop for one week and wired sales, warehouse, and manufacturing onto MIDAS. By day six the company ran on one live operating picture — the floor on auto-assigned task cards, leadership on MIDAS chat.
Book an operating assessment · See the platform
How did NBR deploy MIDAS at OZ Leather?
NBR Intelligence forward-deployed MIDAS across OZ Leather, a 55-person Vietnamese leather-goods exporter with 1,000+ SKUs, in six business days on the company's real data. The engagement connected sales, warehouse, and factory into one operating picture, automated fulfillment planning, and let leadership run the operation by asking MIDAS chat in plain language. Inventory accuracy rose from about 71% to about 98%, order entry fell from about 12 minutes to about 1 minute, and gross margin became visible live instead of estimated at month-end.
Query fit: forward-deployed MIDAS case study; Vietnam manufacturing operating picture; AI for leather-goods export operations.
The challenge: effort wasn't the problem — one trusted picture was.
OZ makes wallets, belts, and handbags for buyers who do not wait. Sales, stock, and production lived in roughly fourteen disconnected spreadsheets kept current by hand. The cost showed up four ways: a spreadsheet tax paid in overtime and late nights; no access control and no audit trail; visibility that always lagged a day behind; and an expansion to the United States that would have scaled the chaos, not the business.
Only about 14% of Vietnamese SMEs run any kind of enterprise software — OZ was in the 86% on Excel. Vietnam is the world's second-largest exporter of leather goods and footwear, a sector that shipped roughly $14 billion in the first half of 2025 alone, with the United States its single largest market.
The approach: one week, forward-deployed.
NBR does not sell software and leave you to fit your operation around it. An engineer worked inside the workshop — in Vietnamese, on OZ's real data and its Monday-to-Saturday rhythm — and shipped working software in weeks, not the year a document-first integrator takes. The week moved in four phases:
- Map (Days 1–2) — learn the operation as it actually runs.
- Model and integrate (Days 2–4) — real catalog, bills of materials, and FIFO cost, so margin is true.
- Configure (Days 4–5) — orders, per-customer pricing, transfers, production builds, and automated fulfillment.
- Onboard (Days 5–6) — managers on the web, the floor on the mobile app in Vietnamese, and leadership on MIDAS chat.
One operating picture: fourteen spreadsheets became one screen.
Inventory state on one side, the plan-to-ship flow on the other — what's low, what's in transit, which orders wait on a plan, and which are ready to confirm. Every order, stock move, and build is a recorded action with an owner and a timestamp: the audit trail the spreadsheets never had, now automatic.
Automated fulfillment: take the order you can't fill yet — and let the system plan it.
A rep can take a 100-tote order against three on the shelf. MIDAS reserves the three, works out everything the other 97 need — pull leather and trim, buy the leather that's short, build the totes, move them back — stages it in dependency order, and routes each step to the right person with a due date. Nobody rebuilds that plan in a spreadsheet, and nobody has to remember who owed what.
The floor: the plan turns into each worker's next job.
The plan splits itself into task cards for the right people, in Vietnamese: Lan creates the PO for the missing zippers, Tú sends one trip of leather, rivets, thread, and lining to the workshop, Hoa receives it, and Minh starts the build once the parts are in. Everything on the same trip is one card, so nobody taps item by item, and a button only unlocks when the step before it is done. Workers record their work however is easiest: tap the button on the card, tell MIDAS in a sentence like “got 380 hides, 20 short,” or snap a photo of the delivery note. The system shows what needs doing; people decide and do it.
The management surface: leadership runs the business by giving commands.
The Chairman and CEO give MIDAS chat commands in plain language — pull a number, assign a task, create an order — and get the result back with the reason and the next move — they know earlier and act faster. MIDAS chat runs on NBR's secure hosting on Microsoft Azure, and the AI models run on Microsoft Azure AI Foundry, Microsoft's enterprise AI platform. Every answer uses only the data the person asking is allowed to see, and every action is logged.
The impact, two weeks in.
- Inventory accuracy: ~71% → ~98% — AI checks data before it's saved.
- Order entry & confirmation: ~12 min → ~1 min — Sales tell the AI the order details in chat; the AI enters it.
- Getting operating status: 1–2 days → seconds — With all the data in one place, the AI answers owners in seconds.
- Month-end overtime: A week → down ~90% — The AI updates records as people work and talk to it.
- Margin: Month-end → live — Calculated live, automatically.
Why it worked.
MIT's NANDA initiative found that 95% of enterprise AI pilots deliver no measurable return — not because the models are weak, but because deployment fails. OZ avoided the 95% by taking away the team's worst job first (the nightly re-keying), deploying into the operation rather than around it, and building on the MIDAS product underneath the field work. And OZ's data stays OZ's: MIDAS chat and its AI models run on Microsoft Azure, and Azure AI Foundry guarantees in writing that nobody can use OZ's data to train an AI model without OZ's permission — not Microsoft, not OpenAI, not us.
Frequently asked questions.
How long did the OZ Leather MIDAS deployment take?
Six business days from kickoff to go-live, on the company's real data. NBR forward-deployed an engineer into the workshop in Vietnam; the week moved through four phases — map, model and integrate, configure, and onboard all 55 people.
What results did OZ Leather see?
Inventory accuracy rose from about 71% to about 98%, order entry fell from roughly 12 minutes to about 1 minute, month-end overtime dropped about 90%, and gross margin became visible live per order, SKU, and customer instead of estimated at month-end.
What is forward-deployed engineering?
An NBR engineer works inside the real operation — mapping how it runs, connecting systems and data, modeling owners and risks, and shipping deployed software on the company's own data in weeks, rather than handing over a document and leaving the business to fit itself around the tool.
Is OZ's data used to train AI?
No. MIDAS chat runs on NBR's secure hosting on Microsoft Azure, and the AI models run on Microsoft Azure AI Foundry, Microsoft's enterprise AI platform. It guarantees in writing that nobody can use your data to train an AI model without your permission — not Microsoft, not OpenAI, not us. MIDAS also keeps permissions and an audit trail: every answer uses only the data the person asking is allowed to see.
Website: https://nbrintelligence.com | Contact: [email protected]