From four hours to five seconds: AI document extraction for a UK cloud accounting platform
A UK cloud accounting platform needed to turn the financial documents its customers send into usable data far faster. We deployed Caelum AI, the extraction engine we developed. It now processes around 100,000 documents a month at 99.5% accuracy, and average turnaround has dropped from four hours to about five seconds.
- documents processed each month
- ~100,000
- average turnaround
- 4 hours → ~5 seconds
- accuracy
- 99.5%
- AWS support cost removed
- £6,000 a year
The challenge
The platform's customers send in financial documents every day. Before anything could be booked, each one had to be read and turned into structured data, and on average that took four hours.
Documents also arrived by more than one route, including email and WhatsApp, so the platform needed one dependable way to take all of them in.
Behind the scenes, an ageing database was still running on AWS extended support, at a cost of £500 a month.
What we built
We worked with the platform's team to add document intelligence and to clear the technical debt that was in the way.
- AI extraction with Caelum. We deployed Caelum, the financial document-extraction engine we developed, to read incoming documents and return structured data. Caelum is also available on the AWS and Azure marketplaces.
- Ingestion by email and WhatsApp. Customers send documents the way they already do, and they are picked up automatically.
- Analytics and data-health modules. Clear views of the state of each customer's books, built on the extracted data.
- Legacy database migration. We moved the platform off its ageing database, ending the AWS extended-support charge.
The result
Around 100,000 documents a month now go through Caelum at 99.5% accuracy, and average turnaround has fallen from four hours to about five seconds.
Retiring the legacy database removed a £500 monthly AWS extended-support fee: £6,000 a year.
Technology
- Caelum AI
- AWS
- WhatsApp and email ingestion
Processing documents by hand? Let's look at your volumes.
A 30-minute discovery call: your documents, systems and volumes, and what could be automated. No obligation.
More case studies
An AI agent that runs quote-to-cash for a UK accountancy outsourcer
Minutes for a multi-step manual process
An AI agent runs quote-to-cash for a UK accountancy outsourcer, from new prospect to recurring invoice in QuickBooks and Xero.
Read the case study →Food & beverage (loyalty)My Professional Rewards: invoice validation from months to seconds
Months → seconds invoice validation
A loyalty scheme for Unilever Away From Home, rebuilt on AWS with Omne Agency: around 250,000 invoices a year.
Read the case study →Facts verified October 2026. All case studies