InvoiceIQ
Clean invoices flow through untouched. Exceptions arrive with a reason.
InvoiceIQ turns inbound supplier invoices — email bodies, PDFs, images, Office files and forwarded chains — into validated entries in your ERP. A bounded AI decision layer decides, invoice by invoice, whether to post it or query it. Your AP team only touches the risky minority, and each one comes with an explicit reason. We built it in production for a UK business and tailor it to your controls.
Who it's for
- Construction and civil engineering
- Interiors and fit-out
- Manufacturing
- Field services
- Distribution and wholesale
- Any mid-market business on an ERP with a real AP team
Accounts payable is interpretation, not just data entry.
In a project-driven or high-purchasing-volume business, an invoice is rarely self-contained. The purchase order number lives in the email, not the invoice. A supplier serves several jobs at once. One document mixes goods, freight and a discount. Some supplier documents behave more like confirmations or receipts than trade invoices. Line items need checking against purchase orders and goods receipts, not merely capturing as text.
OCR reads the page; a generic AI extractor guesses at the fields. Neither decides, safely, whether an invoice is genuinely ready to enter the finance system. That decision — post it, or hold it and say why — is where the work and the risk actually sit, and it is what InvoiceIQ is built to do.
What it does
What InvoiceIQ does
An asynchronous service that sits between inbound invoice material and your ERP posting process.
Multi-source understanding
It reasons across the whole submission — email subject and body, cleaned HTML, text and images from PDFs, and parsed Office documents — because the references you need are fragmented across them. A project code in the email body and the invoice number in the attachment are read together.
Schema-constrained extraction
Messy real-world invoice material is converted into a strict, structured schema and normalised to line-level financial data, so what reaches your controls is consistent rather than free text.
Deterministic finance controls
Price-tolerance checks, VAT reconciliation, duplicate detection, PO and goods-receipt matching, and currency and date checks run as deterministic rules — not as something the model is trusted to remember.
Post it, or query it with a reason
A bounded AI decision layer either auto-posts a clean invoice into the ERP — attaching the source document and recording the reference — or returns a structured query with an explicit reason code from a defined catalogue: no PO, price over tolerance, VAT mismatch, duplicate, goods not received, and so on.
Orchestration, export and feedback
Durable job orchestration and status tracking let your team see where any request sits. A daily CSV and summary export keeps finance in control, and the exceptions surface upstream process problems worth fixing at source.
How it's governed
Built for a finance team's standards
The point of InvoiceIQ is confidence, not just extraction. The engineering reflects that.
A reason for every exception
Nothing is held without an explicit, catalogued reason. Your AP team spends its time on the invoices that genuinely need a human, each one arriving with the reason it stopped.
Deterministic controls, not model judgement
Tolerances, VAT, duplicates and matching are enforced by deterministic finance rules. The AI decides post-or-query within those bounds; it does not get to overrule the controls.
Idempotency against double-posting
Strict idempotency guards mean the same invoice cannot be posted twice, even if a submission is retried or arrives more than once.
Strict prod and test separation
Production and test are kept apart by design, so validating a change never risks a live posting.
ERP-agnostic core
SAP Business One is the first supported ERP integration. The extraction and validation core is ERP-agnostic, so the same discipline can be pointed at another finance system.
The evidence
What we can and can't claim
We hold ourselves to evidence, so we will not put an auto-post rate on this page.
InvoiceIQ runs in production, but no independently audited performance percentages exist, so we make qualitative claims only: clean invoices flow through untouched, and exceptions arrive with a reason code. During a build we measure the real rates against your invoices and your controls.
Questions
InvoiceIQ, answered
See InvoiceIQ decide on your invoices.
A short conversation with the people who built it about where a bounded AI decision layer fits your accounts payable — and how we would measure it on your invoices.