Manufacturing Copilot pilot cohort now open. Join the pilot cohort
Contact
Nucleus One / Document intelligence for financial workflows

Documents in. Decisions out.

Nucleus One helps finance teams automate document-heavy workflows, cutting administrative cost while every decision stays governed and audit-ready.

UK-hosted, in production on AWS. Live demo available.
Product video
01/The problem

What slows finance teams down

From a growing manufacturer to a global group, the pressure is the same.

Manual data entry

Rekeying invoices drives up admin cost.

Extracted with confidence scores

Scattered documents

Files across systems stall approvals.

One sorted intake

High volumes

Backlogs tie up staff and slow decisions.

Batches split and sorted

Weak audit trails

Reviews cost more and carry more risk.

Every change logged

Errors and fraud

Duplicates and fraud cost money.

Flagged before approval

Buried contract risk

Obligations go unpriced and unmanaged.

Risky terms flagged
92%

Reduction in document reading and data entry time

Finance
48%

Error reduction versus manual data entry in case management systems

Finance
380%

Increase in the volume of document processing per FTE

Finance/Public sector
8,000

Hours saved in document and email processing versus RPA

Public sector
02/How it works

Capture. Check. Decide.

One platform from the inbox to the ledger. Document Intelligence and Contract Intelligence now run inside it.

Capture / Document intakeFrom any source

Every document in, sorted

Upload through the web, share by email, sync folders from SharePoint, OneDrive or your own storage, or connect by API. Nucleus One splits scanned batches and sorts every file by type, so nobody sorts by hand.

Web uploadEmail and portalsAPI connectOneDrive and SharePointSplits scanned batchesOCR with visual reasoning
Check / Extraction and reconciliationEvery value scored

Every value scored and matched

Each value arrives with a confidence score and is mapped to our ontology for finance. Where relevant, it is matched against purchase orders and ledger records. Uncertain values go to review instead of the ledger.

Confidence scoresOntology for financePurchase order matchingFraud analysis
Decide / Approval and auditEvery change logged

AI proposes, people decide

Your team confirms or corrects what is uncertain and approves what moves on. Approved data flows to your ERP or accounting system. Every correction is logged for audit, and the platform learns from it.

Approve or correctAudit logERP integrationBounded authority
RAG for Finance / Source-grounded answersBuilt in-house

Ask, and see the source

Context-aware answers, grounded in your own documents, with every source cited. Draft documents, summaries, tables and charts from the same sources. Our own RAG engine, built for financial documents and workflows, combines a knowledge graph with semantic search.

Answers with sourcesDrafts and summariesTables and chartsSemantic searchKnowledge graph
Contracts / Clauses and riskEvery clause read

Key terms flagged for review

Pull out clauses, dates, parties and obligations from contracts, leases and financing agreements. Compare each term against your standard positions and flag the risky ones. Plain summaries go to finance, procurement and legal teams.

Clause extractionRisk scoringStructured summariesDocument comparisonCompliance check
03/Built for financial workflows

Answers you can trace. Limits you set.

RAG for Finance

  • Built in-house for financial workflows
  • A knowledge graph combined with semantic search
  • Every answer grounded in your documents and traced to its source

Ontology for finance

  • Suppliers, accounts, documents and obligations defined once and linked
  • Every value validated and sourced

Fraud analysis

  • Duplicates and mismatches with purchase orders or ledger records
  • Changed bank details and unusual amounts, flagged before approval

Finance agents

  • Search, reason and act across your documents and systems
  • Always within the authority you set

Scales to the size of your business

  • Start with one team in the managed cloud
  • Grow to the whole enterprise on your own infrastructure
  • Forward deployed engineers adapt it to your documents and connect your systems at every step
  • Take only the modules you need now, and the full platform when you are ready

Risk assessment framework

  • Six questions: agent, action, asset, context, control and evidence
  • Sets each agent's limits when authority is granted
04/Use cases

Put it to work by team

Invoices and statements, reconciled

  • Extract data from invoices, receipts, purchase orders and statements, each value with a confidence score.
  • Compare extracted records with ledger, supplier or transaction data, and route discrepancies.
  • Post reviewed records to the accounting or core system.
Handles
InvoicesReceiptsBank statementsPurchase ordersFinancial statements
05/Supported documents

Ready for the documents your business runs on

Invoice and receipttransaction details, amounts, dates
Purchase orderprocurement, supplier, delivery terms
ID and passportidentity verification, KYC
Timesheet and payrollwork hours, pay, approvals
Contract and agreementclause extraction, risk scoring, key terms
Financial statementfinancial performance and position
Bank statementaccount transactions and balances
Custom document typesdefine your own fields for any layout
06/Control plane / Bounded authority

Capability need is not authority granted.

Control planeBounded authorityAgent Risk PassportRisk assessment framework
How the control plane decides An AI agent proposes posting an invoice to the ledger. The control plane tests the action against the authority set by the organisation and returns one of four outcomes: permit within limits, require approval, recommend only, or block. Every decision is recorded in the Agent Risk Passport. Proposal and data Authority and actions Evidence recorded AI agent Proposes an action Post INV-20422 to ledger Control plane Tests each action againstyour authority settings Permit within limits Require approval Recommend only Block Ledger systemEnforcement point Finance approverApproves or rejects StoppedOutside the authority set Proceeds Needs approval If approved Advice only Refused Every decision RECORD Permitted within limits Agent Risk Passport

Authority by action

  • Recommend only, require approval or proceed within limits
  • Set separately for each action

Limits outside the model

  • Role-based access, validation and tool restrictions
  • A prompt cannot widen them

A record an auditor can follow

  • Every approval and agent action recorded
  • Each decision traced and reviewable

Revalidated on change

  • A change to a model, tool or permission triggers revalidation
  • Approval keeps pace with the system
07/Deployment and security

Built to deploy on your terms

A complete platform, adopted rather than built. Our forward deployed engineers provision your environment, adapt it to your document set and connect your systems.

Default

UK-hosted

In production on AWS in the UK, managed by Data Nucleus.

Option

Your private cloud

Deployed into a cloud you control.

Option

On-premises

Runs on your own infrastructure.

Option

Fully offline

Air-gapped, with no outbound connection.

Configured to your documents

Custom document types and fields. Choose your AI engine, bring your own model, or use Data Nucleus models fine-tuned on your data.

One secure sign-in

Single sign-on and role-based access as standard.

Audit-ready by default

Every output, approval and agent action is recorded, with usage monitoring across agents and users.

Aligned with UK GDPR, EU AI Act and the NCSC's Cloud Security Principles

Works with your assistants

Ask from Microsoft Copilot, Claude or ChatGPT

Connect Nucleus One by API, so questions and source-grounded answers happen in the assistant your teams already use.

Microsoft CopilotClaudeChatGPT
Read about trust and security
08/Questions

What finance teams ask first

Which documents does Nucleus One handle?

Invoices and receipts, purchase orders, bank and financial statements, timesheets and payroll, contracts and agreements, and identity documents. You can also define custom document types with your own fields.

What happens when Nucleus One is not sure?

Every value carries a confidence score. Values below your threshold go to the review queue instead of the ledger, where your team confirms or corrects them. Every change is logged for audit, and the platform learns from each correction.

Can it act without a person?

Only within the authority you set. Each action is set to recommend only, require approval or proceed within limits, and consequential actions go to a person. The Data Nucleus control plane checks every proposed action against the authority in the agent's Agent Risk Passport.

Where does it run?

UK-hosted, in production on AWS and managed by Data Nucleus. It can also run in a cloud you control, on your own infrastructure or fully offline (air-gapped). Choose your AI engine, or bring your own model through its API. Answers are drawn from your own documents alone.

Can we ask questions from Microsoft Copilot, Claude or ChatGPT?

Yes. Nucleus One connects to Microsoft Copilot, Claude or ChatGPT by API, so your teams can ask questions where they already work. Answers still come from your own documents, with every source cited.

Do we have to replace our systems or AI assistants?

No. Nucleus One works alongside your ERP, accounting and document systems, and alongside Microsoft Copilot, ChatGPT or Claude. Your existing tools stay in place, and approved data flows into the systems you already run. Nucleus One adds the finance layer: capture and extraction with confidence scores, checks against your purchase orders and ledger, approvals and an audit record.

Who sets it up?

Our forward deployed engineers, on every deployment. They provision your environment, adapt Nucleus One to your documents and connect it to the systems you already run.

Can we use only part of Nucleus One?

Yes. Individual modules can be provided on their own if you do not need the whole platform, and you can add the rest later.

How does a trial work?

Scoped trials run on representative documents, with a baseline agreed first, so you can measure the difference on your own work. Terms and conditions apply.

Try it on your documents.

Scoped trials run on representative documents, with a baseline agreed first. A scoped trial takes about two weeks and needs one person from your team for the baseline and the review. Terms and conditions apply.