Parse, then check against reality
Reading the document is the easy half. The half that matters is checking what it says against your customer records, your pricing and your stock, and refusing to guess when something does not line up.
Order intake
Orders arrive as attachments. Somebody retypes them.
High volume, low judgement, expensive when a digit moves. This is the clearest case for building something — and the easiest one to over-build.
What actually happens
Nobody decided to let this happen.
Customers send orders the way they always have: a PDF, a spreadsheet, an email body, a scan of a fax that was itself a scan. Someone opens each one and keys it into the system of record. It is not difficult work, which is exactly why it is usually being done by somebody whose time is worth a great deal more.
The cost is not really the typing. It is the transposed digit that ships the wrong quantity, the order that sat unopened over a weekend, and the fact that the whole thing stops when that one person is away.
Work out your own number
What is this costing you?
This is the workflow where the numbers are easiest to get and hardest to argue with.
-
1
Orders per week arriving as an attachment or email body rather than through a portal or EDI.
-
2
Minutes to key one in, times that volume. That is the floor.
-
3
What one keying error cost you the last time it happened — the reship, the credit, the customer conversation.
If the volume is there, this workflow usually pays for itself faster than anything else on the list. If the volume is not there, no amount of clever building makes it worth doing.
The honest version
Buy a tool, or build it?
Buy the tool if
- Your customers' documents are consistent — same layouts, same handful of senders.
- You run a mainstream ERP with a supported order-import path.
- You are content to review every parsed order before it posts, permanently.
- Volume justifies a per-document fee at the vendor's pricing.
Build it if
- Every customer sends a different layout and no parser is trained on yours.
- The order needs enriching before it can post — customer codes mapped, pricing checked, stock confirmed — and that logic lives in your head or a spreadsheet.
- You need the exceptions routed to different people depending on what is wrong with them.
- Your system of record has no clean import path, which is common with older or heavily customised installs.
Document parsing has genuinely got good, and there are solid products in this space. If your documents are consistent and your ERP is mainstream, buy one. Building here earns its keep in the messy middle: the enrichment and exception logic around the parse, not the parse itself.
If it is the second one
What a build actually has to do
Exceptions to the right person
An unknown part number and a pricing mismatch are different problems for different people. Routing them separately is what stops an exception queue turning into the same manual job with extra steps.
Nothing posts silently at first
This starts at 'it drafts, you approve' without exception. It graduates to posting on its own only once you have watched it be right for long enough to be bored, and even then with a switch you own.
When to leave this alone
Do not build this yet if…
Under a handful of orders a day, or if your customers are already moving to a portal or EDI, leave this alone. The fix may be arriving on its own.
The part most people worry about
Nothing gets authority by accident.
Every workflow I build has a setting for how much it is allowed to do on its own, and you pick it per workflow rather than once for the whole business.
It drafts, you send
The system does the assembling and the remembering. A person reads it and presses the button. Where most work should start, and where a lot of it should stay.
It runs, you get told
It acts on the rules you agreed and reports what it did. Suitable once a workflow has proven itself and the cost of a wrong move is small.
It runs, you can stop it
Fully hands-off with a switch you own and a record of every action. Appropriate for high-volume, low-judgement work, and not much else.
Straight answers
Questions this one always raises
We already have an ERP. Does that make this pointless?
Usually the opposite. The expensive work is almost never inside the system of record — it is in the gap between the customer's email and the record. That gap is where this build lives, and connecting to what you already pay for beats replacing it.
What happens when it misreads something?
It should refuse rather than guess, and that is a design decision made before anything is built. A system that posts a confident wrong number is far worse than one that puts the document in front of a person and says it could not resolve the part code.
Where this comes up
Most common in these areas
Not a rule, just where I see it most. I work across south and east King County.
Start at the bottom of the staircase
Sixty seconds, and you will know whether this is worth another minute.
No account, no calendar, no waiting for a reply. Run the Fit Check and you get the read on your own screen. If it says the work is worth mapping, the next step is a free 20-minute call. If it says leave it alone, you have lost a minute and I have lost a customer I should not have had.
Or just write to me: bryan@perseidechocreations.com · or call 425-243-3473