Software for the paperwork nobody sees.
Dantum is a UK software company. Its platform automates end-to-end workflows for finance, operations and maintenance teams — the processes where the work arrives as email, the numbers matter, and a missed message costs real money.
The work Dantum automates is high-stakes and almost entirely manual: reconciling payments line by line, tracking records through states, checking expiry dates against a register. It is repetitive, it is exacting, and it hides in an inbox. Dantum exists to do that work automatically and to make the exceptions impossible to miss.
What the platform does
A processing pipeline that runs on a connected mailbox. It classifies each message against a fixed label set, extracts fields into a strict schema, executes the customer's business process in deterministic Python, and routes anything below a confidence threshold to a person. The sample output on the home page shows what that four-stage sequence produces.
- The model reads. The code decides. Classification and extraction use a language model; the arithmetic, the tolerance checks and the state transitions are ordinary tested code. That separation is the whole product.
- Every decision carries a score. Confidence and rationale are written to an event log alongside the outcome, so a decision can still be explained months later.
- Failure is loud. There is no code path where a broken run reports success. An unreadable dependency halts processing rather than proceeding on partial data.
- Humans resolve, the system remembers. Review-queue resolutions are persisted, so the same edge case doesn't return.
Current status
The platform is deployed and processing live mail for customers on Google Cloud in the UK region. We are onboarding a small number of additional organisations while we harden the product. If that could be you, email us.
Roadmap
Where the platform is heading. One product, configured per customer, extended in this order.
- Connected business mailboxes
- Classification and extraction with confidence scoring
- Deterministic reconciliation and state tracking
- Review queue, exception alerts and audit log
- Self-serve configuration of labels, schemas and rules in the client console
- Client reporting on throughput, exceptions and time to action
- Ready-made templates for common back-office processes
- More document types and system integrations
The company
Company no. 17378775 ↗
Company page ↗
Who's behind it
I built Dantum to close the reliability gap in AI — so that automated systems never hallucinate a number that someone is going to act on. As a former Big Four auditor I spent years on the manual daily work that should have been automated years ago, and watched good people spend their attention on it instead of on the problems that needed them. My goal is to give that time back.
How to reach us
Sales and enquiries: sales@dantum.ai. Data protection and support: help@dantum.ai. We reply to everything.