AI invoice reconciliation: how agents are closing the loop with live bank data

TL;DR: AI invoice reconciliation uses specialised agents to match bank transactions to open invoices automatically, applying amount, reference, counterparty, date, and company-specific rules the way a trained finance analyst would. As of March 2026, Cortena is live with this workflow across 1,000+ European banks via a Yapily partnership, processing over 5,000 invoices per month at close to 90% accuracy. When a transaction doesn't match cleanly, the agents don't stall: they flag the exception with full reasoning attached and trigger the next action automatically, whether that's requesting approval, escalating, writing back to the ledger, or notifying the right team member. Workflow Builder, launched the same month, lets finance teams configure their own approval chains, escalation paths, and conditional logic without writing any code, so automation matches how the team already operates. Pre-accounting is live on DATEV, Exact Online, and Twinfield, covering the majority of European SME accounting stacks, with spend tools integration and receivables automation next on the roadmap. This is the execution layer finance operations have been missing.

Why invoice reconciliation is the quiet bottleneck

Ask any controller where the month-end hours go, and reconciliation shows up near the top of the list. Matching invoices to bank transactions is rule-heavy, exception-prone, and, despite being one of the most repeatable jobs in finance, still mostly manual at most companies. Accounting software shows you what's in the ledger. The banking portal shows you what's in the account. The gap between the two is closed by a human with a spreadsheet and a sharp eye.

Closing that gap automatically, at accuracy high enough to trust, is where agent-based finance automation earns its keep.

Live bank data across 1,000+ European banks

Cortena is now connected to over 1,000 banks across Europe through a partnership with Yapily. That connectivity means Cortena's agents pull live bank data directly, across every account and every entity a company holds, without CSV exports, screen scraping, or overnight batch jobs.

Live data matters for three reasons:

How agent-based reconciliation closes the loop

Cortena's reconciliation workflow runs as a chain of specialised agents, each doing one job and handing off:

The result: reconciliation that doesn't stop at "this doesn't match" but moves the work forward. When the rule breaks, most tools stall. Cortena triggers the next action.

Real numbers: 5,000+ invoices a month at close to 90% accuracy

As of March 2026, Cortena is processing over 5,000 invoices per month across live clients and pilots, with accuracy close to 90%. Pre-accounting is live on DATEV, Exact Online, and Twinfield, covering the majority of European SME accounting stacks.

Accuracy numbers without context are meaningless, so here's the context that matters: the 10% that isn't auto-resolved is not lost, it's surfaced as a traceable exception with the agent's reasoning attached, so a human can resolve it in seconds instead of reconstructing the decision from scratch. That's the difference between automation that helps and automation that creates more work.

Workflow Builder: automation that matches your process, not the other way around

Most finance tools treat every company the same. Your approval chains are different. Your exception rules are different. Your escalation paths depend on amounts, entities, counterparties, and policies that are unique to your business. Forcing a generic template is where automation stops being useful.

Workflow Builder, live this month, lets finance teams define exactly how their processes run:

When the rule breaks, the workflow doesn't stop. It triggers the next action. That's the execution layer finance operations have been missing.

What this unlocks next

Live bank data plus deterministic agents plus configurable workflows equals an execution layer that doesn't exist anywhere else in finance tooling. From here, coverage expands fast: spend tools integration and receivables automation are next. The underlying architecture is the same: automate upstream, handle exceptions reliably, and let clean execution become the source of downstream clarity.

For finance teams evaluating automation today, the benchmark is shifting. It's no longer "can the tool show me what happened?" It's "can the tool handle what happens next?"

FAQ

How does AI invoice reconciliation work?

AI invoice reconciliation uses specialised agents to ingest bank transactions and open invoices, match them using amount, reference, counterparty, and contextual rules, flag exceptions, and automatically trigger the next action: approval, escalation, or ledger update. The agents leave a traceable audit log for every decision.

Which banks does Cortena support?

Cortena connects to over 1,000 banks across Europe through a partnership with Yapily, covering every account and entity a company holds. Live bank data is pulled directly, no CSV exports or overnight batches.

What accuracy does Cortena achieve?

As of March 2026, Cortena processes over 5,000 invoices per month at close to 90% accuracy. Exceptions that don't auto-resolve are surfaced with the agent's reasoning attached, so humans can resolve them in seconds.

What is the Workflow Builder?

Workflow Builder lets finance teams define their own approval chains, escalation paths, reviewer assignments, and conditional logic, no code. It's the configuration layer that makes agent-based automation match how your team actually operates.

Where can I try Cortena?

Cortena is onboarding teams in Germany and the Netherlands, especially those using DATEV, Exact Online, or Twinfield. Try it at app.cortena.ai or visit cortena.ai.