Every contractor I've worked with has the same problem. Signed delivery orders (DOs) pile up, someone has to match each one to the right purchase order (PO), key it into the project system, and file the paper. It's slow, it falls behind, and when it falls behind nobody knows what's actually been delivered.
I've now automated PO/DO matching for two Malaysian contractors. Here's how it works, and the problems you'll hit that nobody warns you about.
The basic flow
- Scan. Signed DOs are scanned into one shared folder, sorted by project. This is the only manual step, and it's one your document control team already does.
- Read. AI reads each DO: supplier, DO number, date, items, quantities and units, including handwriting.
- Match. Each DO line is matched to an open PO line: same project, same supplier, same item, quantity within what's left on the PO.
- Check before keying. Before anything is entered, the DO is checked against what's already recorded, so nothing is keyed twice.
- Key. Matched DOs go into your system with the signed scan attached.
- Flag the rest. Anything unclear goes on a short check list for purchasing. Nothing is guessed.
The problems nobody warns you about
- Staff typos in your existing records. In one batch of 76 scans, the AI held back 8. Seven of them had already been keyed by staff months before, with a digit added or swapped in the DO number. A strict "same DO number" check would have created seven duplicates. You need a fuzzy match: same supplier, same date, same quantity, and a DO number that's one character off.
- Suppliers print more than one number. Many dockets have a DO number and a separate form serial or plant code. You have to tell the AI which one your system uses, supplier by supplier.
- DOs with no PO. Rentals, urgent purchases, items ordered by phone. These can't be matched and shouldn't be forced. They go to purchasing to raise or link a PO.
- Closed or used-up POs. A DO that arrives after a PO is fully delivered needs a person to decide, not the AI.
- Unreadable fields. A smudged quantity, a date covered by a stamp, a missing signature. The AI must flag these, not "best guess" them.
What it takes
At one piling contractor, the system has read 1,300+ delivery orders and checked them against 18,000+ PO lines and 41,000+ existing records. Each morning, purchasing gets a short check list instead of a pile of paper, and document control gets a list of DOs that were logged but haven't been scanned yet.
Building PO/DO matching is the biggest single workflow I do: usually 4–10 sessions (RM1,600–4,000), because it has to connect to your accounting or project system and handle your suppliers' quirks. It's also the one that saves the most time.
Where to start
- Pick one project and one supplier type (concrete is a good start: high volume, consistent dockets).
- Get your document control team scanning into one folder.
- Run it for two weeks alongside your current process before switching over.
If you want to know what PO/DO matching would look like for your company, the 30-minute task check is free, and you'll get a personalised automation plan.
Pang Jun Han: AI automation for Malaysian construction companies and SMEs. How PO/DO matching works and what it costs: https://pangjunhan.vercel.app/automate-delivery-orders · All services: https://pangjunhan.vercel.app/ai-automation
Top comments (0)