Document understanding
Understand documents that do not fit a fixed template.
Pull out the useful details, summarise what the document says and keep unclear meaning visible so a person can review it before the information is used.
Some of the meaning is clear.Some of it still needs a person.
Example document
Customer requirement
Apex ComponentsReceived 21 Oct
Need 200 units of 6204 or equivalent before month end. Material certificate required. Existing drawing BRG-17 may still apply.
What the document says
- Product
- 6204 or equivalent
- Quantity
- 200
- Delivery
- Before month end
- Requirement
- Material certificate
Needs review
Drawing reference
BRG-17 may still apply
A fixed map works — until the document changes.
Document data capture
- A known document type
- Expected fields
- A configured mapping
- Stops at uncertainty
Document understanding — this page
- A document that varies
- Free text matters
- Meaning has to be interpreted
- A person reviews what is unclear
Neither of these is the advanced version of the other. A business with a steady supplier bill wants the first; a business reading customer requirements written in prose wants the second.
Document Data CaptureStart with what the document actually says.
Example source
Need 200 units of 6204 or equivalent before month end. Material certificate required. Existing drawing BRG-17 may still apply.
- 200 units
- Quantity
- 6204 or equivalent
- Product, with a substitution allowed
- before month end
- Delivery, in the customer’s words
- Material certificate required
- A requirement, not an item
- BRG-17 may still apply
- Cannot be settled by reading it
Four useful facts and one open question, in one sentence a customer typed without thinking about fields. No fixed template would have caught the certificate, and none would have known what to do with “may still apply”.
Pull out what is clear.
| Field | Value | Read from |
|---|---|---|
| Customer | Apex Components | The document’s sender |
| Product | 6204 or equivalent | “6204 or equivalent” |
| Quantity | 200 | “200 units” |
| Delivery | Before month end | “before month end” |
| Requirement | Material certificate | “Material certificate required” |
Each value can be traced back to the words it came from, which is what makes it checkable. A value nobody can trace is a value nobody can correct.
Keep uncertain meaning visible.
Needs review
“Existing drawing BRG-17 may still apply.”
- What is clear
- A drawing reference exists: BRG-17
- What is not
- Whether it applies to this requirement
- Who can settle it
- The person who knows the account
The temptation is to write “Drawing: BRG-17” and move on. That would turn a question into a fact, and the first time it is wrong the whole result stops being trusted.
There is no percentage here, and there will not be one. “Needs review” is a sentence a person can act on; a confidence figure would be a measurement nobody took.
Summarise the source — do not invent the answer.
The source
A shorter reading of it
- 200 units of 6204, or an equivalent
- Needed before month end
- Material certificate required with the despatch
Still unresolved: whether drawing BRG-17 applies.
A useful summary is shorter than the source and says the same things. It does not answer the question the source left open — if three messages never settled the drawing, the summary cannot settle it either.
What should stay human?
- Deciding whether an ambiguous term is acceptable
- Choosing the business action that follows
- Approving a commercial commitment
- Correcting an interpretation that read the source wrongly
- Deciding what to do when the source itself is unclear
Interpretation can prepare useful information.A person reviews unclear meaning before it enters the business process.
Review before the result enters the process.
Document
However it arrived
Interpreted information
4 clear · 1 needs review
A person reviews
Settles the drawing question
The normal business process continues
An enquiry, a quotation, a record — whatever it was going to be
What happens next depends on your process, and this page does not claim it. Nothing here posts a record, approves anything or decides what the business should do about the requirement.
Document understanding is not document data capture.
If your team already knows exactly which fields belong where, mapped capture is usually the simpler job. If somebody first has to understand what the document means, this page owns that step.
Document Data CaptureWhere understood information goes next
- Document Data CaptureWhen the document type is known and the fields are mapped.
- Business DraftingWhen the next step is writing something back.
- Supplier Quote ComparisonWhen several quotations have to be read against each other.
- AI for BusinessWhere AI helps across reading, finding, drafting and comparing.
Which documents does your team have to read before anybody can act?
Show us a document your team reads by hand today. We’ll help separate what can be interpreted consistently from what still needs somebody to decide.