Solution
Document & Finance
Document and Finance Operations, Read and Reconciled
Invoices, purchase orders, delivery notes, statements, claims, contracts. The information is already structured — it is just trapped in a layout. Extraction is the easy half; the valuable half is validation and the exception queue.
“Qualified people spend their week retyping numbers off PDFs.”
How you know
You probably have this problem if…
- Someone's job title says analyst and their day says data entry
- Month-end close slips because documents arrive in twelve formats
- Errors are found downstream by the customer rather than upstream by you
- You cannot answer 'how much did we commit to this supplier' without a manual trawl
The change
What the workflow looks like before and after.
Amber marks a step the system performs. Solid ink marks one a person still performs. The count above each column is how many steps still need someone's attention.
Today
5 human steps- Performed by System
Invoice arrives by email
- Performed by A person
Downloaded and opened
- Performed by A person
Fields typed into the ledger
Line by line
- Performed by A person
Cross-checked against the PO
If there is time
- Performed by A person
Filed to a shared drive
- Performed by A person
Discrepancy found at month end
After
1 human step- Performed by System
Invoice arrives by email
- Performed by Automated
Fields extracted and classified
- Performed by Automated
Validated against PO and tolerance
- Performed by System
Posts straight through
When every check passes
- Performed by A person
Clears the exception queue
Only the failures
- Performed by System
Archived, indexed and searchable
- Automated
- A person
- System
- Customer
The system
What this looks like running in a business.
The same components we build with.
- Processed
- 1,842
- Straight through
- 1,536
- Exception queue
- 264
83.4%
held for a person
- Supplier
- Whitrow Logistics
- Invoice
- INV-88216
- Net
- £4,087.60
- VAT
- £817.52
- Total extracted
- £4,905.12
- Purchase order
- £4,837.12
- PO total differs by £68.001Held — three-way match failed
- Supplier not on file14Possible new vendor, needs approval
- VAT inconsistent with line items9Field highlighted for correction
- Duplicate reference detected3Blocked before posting
27 items awaiting a person, grouped by cause so one decision clears a class rather than a case.
Nothing reaches the ledger unchecked. A document failing validation stops; it is never posted on a best guess.
Invoice arrives06:02
Email attachment, one of two hundred this morning.
Fields extracted06:02
Tuned against this practice's own document corpus.
Matched to a purchase order06:02
Supplier recognised, PO located, terms retrieved.
Validation fails — stopped06:03
High confidence and still held. Accuracy is not the control.
Cleared from the exception queue09:15
Field highlighted, cause stated. Seconds, not reopening the PDF.
Posted, indexed, searchable09:15
Filed against supplier, period and purchase order automatically.
State
- From
- Whitrow Logistics
- Format
- PDF, scanned
- Reference
- INV-88216
- Touched by a person
- Not yet
Follows a single record through the system, including the point where it stops and waits for a person.
The work does not disappear. It moves — and the steps left are the ones that needed a person.
Koderead · scope of authority
What we build
Four things, in this order.
Extraction tuned to your document set
Generic OCR fails on the formats that matter most to you. We tune against your real corpus, including the supplier who still faxes.
Validation rules before posting
Totals reconcile, tax is checked, the PO matches, the supplier exists, the amount sits inside tolerance. Nothing reaches the ledger unchecked.
A triaged exception queue
Documents that fail validation are routed by reason and severity, with the field in question highlighted, so a person resolves it in seconds rather than reopening the PDF.
Straight-through rate as the metric
The number that matters is what share posts without a human touch. We report it weekly and work it upward.
Boundaries
What stays with a person.
Agreed in writing before the build, not discovered afterwards.
- Payment release always requires human authorisation — we automate preparation, never disbursement
- Statutory filings are prepared for review, never submitted unattended
- Any document failing validation stops; it is never posted on a best guess
- Confidence thresholds are set by your finance lead, not by us
Integrations
We connect to what you already run.
Replacing a working system mid-transformation adds risk without adding value. If the existing tool genuinely cannot support the workflow, we say so at assessment rather than three months into a build.
- Xero
- QuickBooks
- Sage
- NetSuite
- SAP
- Tally
- Dext
- SharePoint
- Google Drive
Questions we get
Asked and answered.
How accurate is the extraction?
On a tuned document set, high — but the honest answer is that accuracy alone is the wrong question. What protects you is the validation layer: a wrong number that fails a reconciliation check never reaches your ledger. We design for caught errors, not for a claimed accuracy percentage.
Our documents are genuinely messy. Scans, handwriting, twelve templates.
That is normal and it is why we start by processing a real sample before quoting. Messy corpora change the effort, and you should know that before signing rather than after.
Is our financial data used to train a model?
No. We deploy with data-processing terms that exclude training use, and where the sensitivity warrants it we run extraction inside your own cloud tenancy. This gets specified in writing at design stage.
Where this lands hardest
Industries feeling this most.
Next step
Find out whether this is actually your highest-value fix.
Eleven questions, scored in your browser, with the arithmetic shown. It may well tell you that a different area matters more.
Start the assessment