Use case · Finance

AI for accounts payable and reconciliation, with every entry approved.

An autonomous finance system reads invoices, matches them to payments and purchase orders, flags duplicates and anomalies, and prepares the close continuously, while your team approves what is posted.

All use cases

The problem

Why this work does not scale.

Month-end is slow because the work is collected rather than continuous. Invoices arrive by email, in portals and on paper. Payments appear in bank feeds with references that do not quite match. Someone has to line them up, chase the gaps and explain the differences, usually in the last days before the close.

The work is repetitive and rule-driven, which makes it well suited to automation, but it also carries financial and audit consequences. The right design does the matching and preparation continuously and keeps a person approving what reaches the ledger.

How it works

From first signal to finished work.

  1. _01_

    Collect

    Gathers invoices from the inbox, portals and document stores in whatever format they arrive.

  2. _02_

    Read

    Extracts vendor, amounts, dates, tax and line items, and links each field to its source.

  3. _03_

    Match

    Matches invoices to purchase orders, receipts and bank payments and explains any difference.

  4. _04_

    Flag

    Highlights duplicates, unusual amounts, new vendors and missing documents for review.

  5. _05_

    Prepare

    Proposes ledger entries and the close pack for approval, with lineage to the source.

Authority

You decide how far it goes.

The same function can run at different levels. Most systems begin at Level 2, where people approve each action.

  1. L1 AssistSummarizes and recommends matches. The team enters them.
  2. L2 ApprovePrepares every entry and match. A person approves each one.
  3. L3 LimitsPosts matched entries under a value and vendor limit. Everything else waits for approval.
  4. L4 RunRuns reconciliation end to end for defined accounts. Finance reviews exceptions and audits samples.

Payments stay under human control

Reading, matching and flagging are low risk and high volume, so they move fast. Releasing money is different. In most deployments payment release remains at Level 2 for a long time, or permanently, and moves to Level 3 only for narrow, low-value categories you define. Authority is set per action type, not per department. See Autonomy is a dial, not a switch.

Built for Thai finance teams

The system reads Thai and English invoices, including scans and photos, and is configured around your own chart of accounts and approval chain, including the tax documents your business handles. It works inside your accounting software and inbox rather than asking you to replace them. For a worked example, read A day in an autonomous finance function.

Governance

Guardrails built in from the first week.

Guardrail

Entries approved by people

Nothing posts to the ledger without the approval level you set, and every entry carries its evidence.

Guardrail

Source lineage

Every figure links back to the document and page it came from, so reviewers and auditors can verify it in seconds.

Guardrail

Complete audit trail

Each read, match, flag and approval is recorded, attributable and reviewable.

Days to closeStraight-through matchingDuplicate payments caughtHours of reconciliation

Questions

Accounts payable and reconciliation, answered.

Does it replace our accounting software?

No. It works inside the accounting, ERP and banking tools you already run, under scoped permissions.

How does it handle invoices it cannot read?

It flags them with the reason and routes them to a person rather than guessing, and early deployments run with every entry approved.

Can it find duplicate payments?

Yes. Duplicate detection across vendors, amounts and references is one of the first things the system does, and each finding comes with its evidence.

How is accuracy checked before go-live?

We replay the system against your historical cases, compare it with what your team actually did and study every difference before it works on live data.

Begin

Start with this one function.

Tell Primues where work slows down. It will outline the opportunity and connect you with an architect from our team.

See pricing and engagement