Case study · Finance operations
Supplier invoices processed without retyping
Accounts payable teams still key thousands of supplier invoices a month. This is how we used document AI to read invoices from email and PDFs, validate them against purchase orders and post them to the ERP, with people checking only the exceptions.
Project overview
Document AI for accounts payable.
- Industry
- Finance operations
- Client
- Confidential
- Solution
- Invoice automation
- Platform
- Azure AI + ERP
- Timeline
- 8–12 weeks
- Team
- 3 people
- Inputs
- Email, PDF, scanned paper
- Validation
- PO and goods receipt matching
- ERP
- Business Central, F&O or other
Expected business outcomes
What the client set out to achieve
Measured on a sample of real invoices before and after go-live.
-
60–85%
Invoices posted without manual entry
Extracted, matched and posted automatically when all checks pass.
Target range; depends on supplier mix and PO coverage
-
95%+
Field extraction accuracy
Key fields checked against a labelled sample before launch.
Launch threshold on the client's own invoices
-
Days → hours
Invoice processing time
Invoices reach approval the day they arrive.
Seen once exceptions were handled daily
Business challenges
Where the client started
Invoices arrive in every format and are typed in by hand.
Arrival
Invoices by email, portal and post to a shared inbox.
Entry
Staff key supplier, amounts, VAT and lines into the ERP.
Matching
Purchase order matching done by eye.
Errors
Duplicate and incorrect invoices slip through.
Month-end
A backlog builds up just when finance is busiest.
Proposed solution
What the automation covers
Extraction, business rules and a clear exception queue.
-
Capture
Invoices collected from mailboxes and upload points automatically.
-
Extraction
Supplier, dates, totals, tax and lines extracted with confidence scores.
-
Validation
Checks against supplier master, purchase orders, receipts and duplicates.
-
Exception queue
Low-confidence or failed invoices reviewed side by side with the original.
-
ERP posting
Validated invoices created in the ERP with the PDF attached.
-
Reporting
Automation rate, exceptions by reason and processing time.
Project requirements
What the client asked us for
The request was to stop typing invoices. The key is confidence: anything the AI isn't sure about goes to a person, with the fields highlighted.
We measure accuracy on the client's own invoices, because supplier formats vary so much.
What the project had to deliver
- 01Automatic capture
- 02Accurate extraction
- 03Matching to orders
- 04People review only exceptions
Key challenges & considerations
What made it harder, and what we did
-
Poor-quality scans
What we did
Low-quality scans were scored for confidence and routed to a person below the threshold.
-
Invoices without PO numbers
What we did
Invoices without a PO number were matched by supplier, amount and date, or sent for coding.
-
Multi-page, multi-currency invoices
What we did
Multi-page and multi-currency invoices were included in the test set and handled explicitly.
-
Supplier master data quality
What we did
Supplier master data was cleaned before launch, because matching depended on it.
What we watched for
- Test on the client's own invoices
- Route low confidence to people
- Clean supplier master data
- Track accuracy after launch
Development process
Measure on your invoices
Accuracy is proven on a labelled sample first.
-
Sample
A representative sample of invoices labelled.
-
Extraction
Models configured and tested on the sample.
-
Validation rules
PO matching and duplicate checks built.
-
ERP integration
Posting and exception queue connected.
-
Go-live
Live with daily exception reviews and monthly accuracy checks.
Technology stack
Document AI with ERP integration
Azure document models, with a language model for unusual layouts.
AI
- Azure AI Document Intelligence
- Azure OpenAI for edge cases
Workflow
- Power Automate or Azure Functions
- Exception queue app
ERP
- Business Central / F&O / other ERP API
Monitoring
- Accuracy sampling
- Automation dashboards
Quality assurance & deployment
Measured against a test set before launch
AI quality was measured, not judged from a demo. A test set of real examples with agreed answers decided whether the system was ready, and the same set was re-run after every change.
- A test set of real examples with agreed answers
- Accuracy thresholds agreed with the business before launch
- Safety checks: refusals, permissions and prompt-injection tests
- A person reviewing low-confidence output
- Cost and response time measured at real volumes
Deployment
- 01ERP integrationPosting and exception queue connected.
- 02Go-liveLive with daily exception reviews and monthly accuracy checks.
- 03Pilot, then staged rolloutA small group of real users first, with results reviewed weekly before widening access.
- 04Monitoring and warrantyAccuracy, drift, usage and cost tracked after launch, and six months of free bug fixing.
Final deliverables
AP focused on exceptions
Most invoices flow from inbox to ERP without being typed, and the team spends its time on genuine exceptions.
Duplicate and mismatched invoices are caught before payment.
- Automatic capture
- High-accuracy extraction
- PO matching
- Exception-only review
Handed over to the client
- 01Production AI serviceDeployed in the client's own cloud tenant
- 02Source code and modelsIn a repository the client can access
- 03Test set and resultsThe evidence behind the launch decision
- 04MonitoringAccuracy, usage and cost
- 05Six-month warrantyFree bug fixing after delivery
Success metrics
How success is measured
| Measure | Outcome | How it is measured |
|---|---|---|
| Invoices posted without manual entry | 60–85% | Target range; depends on supplier mix and PO coverage |
| Field extraction accuracy | 95%+ | Launch threshold on the client's own invoices |
| Invoice processing time | Days → hours | Seen once exceptions were handled daily |
Verified reviews
What clients say about our custom software work
Verified reviews from our clients on the same kind of work, published on Clutch.
-
Custom Software Development
Web-Based Coupon App Development
The work produced by the team was of exceptional quality.
Jon Thies
CTO, Model Rocket Feb 2022 – Mar 2023
-
Custom Software Development
Mobile & Web Platform Development & Design
Silver Scintilla was responsive, respectful, and enjoyable to work with, quickly addressing any questions or requests.
DeShawn Brown
CEO & Founder, Lithios Jun 2021 – Nov 2022
-
Custom Software Development
Web-Based HRM System Development
Silver Scintilla provided the best solution with scalable functionality, personalized features, and robust security.
Suheb Khan
Director, Collaborative Insight Technologies Jul 2020 – Jun 2022
Free discovery call
Have an idea? Let's turn it into AI-powered software.
Book a free discovery call with our experts. Share your idea and we will help you shape the scope, timeline and budget, under NDA.
- Free consultation
- NDA before we talk
- Transparent estimate