All AI offers

AI agents for finance, procurement and HR that wait for your approval

The agent reads each request, checks your ERP, mailbox and documents, and drafts the reply or the transaction. A named person approves the steps that matter, then it acts and logs each one.

The problem

Skilled people spend their day moving information between email, the ERP and spreadsheets. They match invoices, answer status questions and chase approvals. The work repeats, but each case differs a little. Basic automation breaks on the unusual cases.

The agent prepares each case, and a named person signs it off

The agent reads each request and checks it against your ERP, the mailbox and your documents. It applies your rules and drafts the posting, the reply or the query. Each connector allows only what the agent’s role needs. It can read an order, but it cannot post or pay. When a case does not fit the rules, the agent hands it to a person with its reasons.

Systems it touches

  • Your ERP
  • Outlook and shared mailboxes
  • SharePoint and document stores
  • Ticket and request systems

Example

An invoice that is 3% over the purchase order

A made-up case that shows the mechanism. It describes no client.

  1. A supplier emails an invoice to the accounts payable mailbox. It is 3% over the purchase order.
  2. The agent reads the PDF, finds the purchase order in your ERP and compares them line by line.
  3. It finds the gap. On one line, the unit price is higher than the agreed price.
  4. It drafts a query to the supplier and a note for the finance lead, with both documents attached.
  5. The finance lead sees the query and the note side by side and approves the query.
  6. The agent sends it from the shared mailbox and logs each step. The invoice stays unposted.

It works in the systems you already have

The agent reaches your ERP and Microsoft 365 through connectors with set permissions. We test candidate models on your own past cases and choose the one that handles them best. For sensitive data, that model can run on your own servers or our private GPUs.

Where your data stays

Permissions limit what it can touch

Where a person approves

  • Before anything leaves the organisation
  • Before any posting or payment in the ERP
  • When the agent is unsure, it hands the case to a person and explains why
  • Reading a purchase order and posting a credit note are separate permissions. Each connector grants only the actions the role needs.
  • The agent starts by asking for approval on every step. A step runs without approval only after you have seen enough evidence to allow it.
  • Every run is recorded: the input, each step, the draft and who approved it.

Everything we build is handed to your team

Who owns the code, and how handover works
  • A working agentCode in your repository, running in your accounts.
  • Test cases built from your real workKept in your repository and run on every change.
  • Operating guidesWritten for the people who supervise the agent.

How we would start

Discovery, then a trial agent

We choose one task with a clear owner and real volume, map it, and agree where a person approves. The trial agent then does that task in live use, with a person approving its actions.

How we work together

Questions

Which tasks suit an operations agent?

Tasks with steady volume, a clear owner and a mix of rules and judgement: invoice queries, purchase requests, order status, employee requests.

Does the agent replace the people doing the work?

It does the collecting and drafting. The people who know the work approve it, correct it and handle the unusual cases.

Can it post to our ERP directly?

Only after approval. The agent drafts the posting or the payment, and a named person approves it. A step runs without approval only after you have seen enough evidence to allow it.

Send us the request that takes your team longest

Include the systems it touches and who signs it off. We will map how an agent would prepare it and where a person approves.

Start a project