The platform everything we run sits on.
Opnua is the platform Apridata runs its customers’ back offices on: it holds the commercial documents between a business and its counterparties, posts what fits under configured rules, and sends what doesn’t to an AI agent whose recommendations post only after a named person has approved them.
Rules post what fits; your people approve what the AI recommends. Ours run it, from our cloud or inside your network; where your team would rather run it day to day, we stay behind them. It sits alongside your ERP, or holds your documents when you have no system for them, and either way the ledger stays yours.
Every business has a system of record. Almost none has a system of resolution.
A system of resolution is the layer that sorts what doesn’t fit the system of record, before it is recorded and after, so the ledger only ever takes what fits tolerance.
The ERP accepts a transaction once it fits within the tolerances you have set: an invoice matched to an order and a receipt, a payment matched to the invoices it settles. Everything outside that has to be sorted before the ledger will take it, and again afterwards, when a recorded invoice is disputed. In most mid-market businesses that work has no system. It has a person, an inbox and a spreadsheet.
The system of record is the ERP. It accepts what fits: deterministic, auditable, and mostly fine. It is your system of record, and it stays yours.
The system of resolution sorts what doesn’t fit, before it’s recorded, and after. Today, in most businesses, that system is a person, an inbox and a spreadsheet. The Opnua platform replaces it with one: rules, an AI agent, and a log of every step.
The resolution tax is the avoidable cost of investigating and clearing exceptions across systems, documents and people; it is why back-office headcount rises with revenue. Skilled people’s hours spent copying, checking and chasing, plus the cash sitting uncollected while they do.
Rules record. AI resolves. People approve.
Rules post what fits, within limits you set. The AI works what doesn’t and recommends. Nothing it proposes posts until a named person approves it. Every step is logged, by whoever or whatever took it: rule, agent or person.
- 09:02 rule Invoice 4471 matched PO 8812 and GRN 3310 · posted
- 09:02 rule Remittance €18,240.00 matched 6 invoices · posted
- 09:04 agent Invoice 4478 short by €212.40 vs PO 8819 · investigating
- 09:05 agent Read delivery note 3318: 2 cartons short-shipped
- 09:05 agent Emailed supplier for credit note · awaiting reply
- 09:41 agent Credit note CN-1194 received · recommend: match and post
- 09:44 person M. O’Brien approved · posted with credit note
The judgement calls rules can’t make.
Most processes are largely rule-based with a stubborn remainder of judgement. The rules take the first part. The remainder is where workflow tools and traditional RPA give up, and where the agent works.
Works the exceptions.
Mismatched invoices, unmatched payments, ambiguous orders, late confirmations. It reasons from your past resolutions, the ERP and the correspondence, and recommends a resolution with the evidence attached.
Handles the correspondence.
Routine requests go out on their own, in your name and to wording you sign off: copy invoices, credit-note requests, missing confirmations, payment reminders on your schedule. Anything that disputes, concedes or changes what was agreed is drafted and waits for a named person. The agent learns from their edits.
Knows when it isn’t sure.
Every recommendation carries a confidence score. Below the threshold it says so, takes no action, and routes the case to a person with everything known so far.
Remembers what people decided.
When a person resolves a case, the pattern is kept. Similar cases are recommended sooner next time, and patterns that prove reliable become rules.
Every document between you and your counterparties.
Documents.
Orders, confirmations, purchase orders, deliveries, invoices, credits, statements and remittances.
Open positions.
Open items, open orders, open purchase orders, and quantities on order and received.
Cases.
Everything that didn’t fit, with a log of every step taken on it.
Your rules and tolerances.
The ones you set, versioned, so you can see which rule applied on any date.
What it has learned.
About each counterparty: how they invoice, how they pay, and how their exceptions were resolved before.
Either way, we don’t replace your ledger.
The ERP holds the orders and purchase orders.
Opnua reads them, resolves what doesn’t fit, and writes back what’s approved.
When you have no ERP module for orders and purchasing.
Opnua holds them, and posts invoices, credits and cash to your accounts package.
The systems you already run.
- SAP Business One Deepest
- Sage Deepest
- Exchequer Deepest
- NetSuite Production
- Business Central Production
Also integratedSAP S/4HANA · Oracle · Intact · QuickBooks · Xero · MS Dynamics GP · custom-built ERPs · UI-driven agents where no API exists
- Mail and documentsMicrosoft 365, Graph, Outlook, SharePoint, OCR, and structured e-invoicing through a certified Peppol provider.
- BankingOpen Banking, BACS and SEPA feeds, payment gateways.
- WarehouseWMS, barcode scanners, goods receipts and stock movements.
- LogisticsCarrier portals, proof-of-delivery capture, route systems.
- ReportingSQL warehouses, and the Excel models you close with today.
- Where no API existsUI-driven agents that work the screens your people use today.
How this differs
| Approach | What it covers | What still lands on a person | What changes in your ERP |
|---|---|---|---|
| AP automation | Invoice intake and matching inside accounts payable; some tools now handle exceptions inside AP too. | The exception that crosses AP, AR, procurement and the bank: the invoice that doesn’t match the order, the payment that doesn’t tie out, the dispute in someone’s inbox. | A connector to the ERP. The ledger stays. |
| Your ERP vendor’s AI | What is already inside the ERP. | The work between systems: the supplier’s PDF, the email thread, the remittance that doesn’t match the bank. | Depends on your version and modules. |
| Replacing the ERP | A new system of record and whatever its modules do. | The same exceptions, in a new system, after the programme ends. | Everything. A programme measured in months. |
| Opnua, with Apridata | The commercial-document loop: orders, purchase orders, deliveries, invoices, statements and payments, before they are recorded and after. Run by us, or by your team with us behind them. | Approval. A named person approves what the AI recommends before it posts. | Nothing in its configuration. Approved entries are written back through its API or its user interface. |
Rules post. The AI recommends. People approve.
The AI never posts to your ledger. Configured rules post eligible transactions; the AI investigates the exception and recommends; anything it proposes is written back only after a named person approves it. Every action is logged, by whoever or whatever took it.
Approval on every recommendation
The agent proposes a resolution with its reasoning and evidence. A named person accepts, changes or rejects it. The same holds for any message that disputes, concedes or changes what was agreed; routine requests go out on their own.
A log an auditor can follow
Rule, agent or person: every action, in order, with what it looked at and why.
Deterministic where it can be
If it can be a rule, it is a rule. The AI is confined to the judgment work, and it says so when it isn’t sure.
Built for product businesses
Three-way matching, proof of delivery, short shipments, rebates and returns: the exceptions manufacturers and distributors actually have.
Nothing intelligent posts to your ledger. Rules post. The AI recommends. People approve.
Where it runs, and who runs it.
Three separate questions, with the same answer to the first every time.
| Question | By default | The alternative |
|---|---|---|
| Who approves? | Your people. | Your people, always, in every mode. |
| Who runs it day to day? | Us: monitoring, tuning rules, onboarding new suppliers and customers, handling what the agent escalates. | Your team, with us behind them, where you host it and want control. |
| Where does it run? | Our cloud. | Your own infrastructure, behind your firewall, with local models where cloud AI isn’t allowed. |
In practice that makes three modes: in our cloud, run by us; on your infrastructure, run by us; or on your infrastructure, run by your team, supported by us. Same architecture in every mode, with local models where your data, your regulator or your customer says cloud AI isn’t allowed.
Behind your firewall
Hosting where the data lives is a separate choice from who runs it day to day.
Open-weight models running on Apple Silicon for lighter workloads, or on on-premises GPU servers for heavier ones, air-gapped where required. Same platform, same log; no document leaves your network.
- Healthcare providers under HSE, NHS or equivalent data-handling rules.
- Financial services firms whose client data can’t touch a third-party model.
- Regulated and defence-adjacent manufacturers.
- Any business whose customers contractually prohibit cloud AI, or whose jurisdiction requires data to stay in-country.
“Nobody needs artificial intelligence to post a clean invoice. A rule does that, and a rule can be audited. They need it for the four hundred that don’t match. Every finance director I meet is being told to put AI into their systems. My advice is to keep it out of the ledger and put it on the work around the ledger.”
Straight answers.
Does anything write to our ERP without a person approving it?
Rules post what fits, deterministically, the way your matching rules do today. Anything the AI agent resolves is recommended first and posted only after a named person approves it. Every action is logged.
Which ERPs does it work with?
SAP Business One, Sage, Exchequer, Microsoft Dynamics, NetSuite and Oracle among others, on-premises or cloud. The Opnua platform reads the ERP and writes back only approved resolutions. The ledger stays yours, and we don’t reconfigure it.
What if we’re acquired, or want to leave?
Everything is logged, every document is exportable, the stack is standard and the ledger was always yours. We build for the day someone else’s advisers look at it.
See what Opnua would hold for you
Thirty minutes, your documents, no slides. We’ll show you what would record itself and what would become a case.