How the agent earns the right to move your cash.
This isn't a compliance product bolted onto a platform pitch. It's the mechanism behind one thing: the dunning agent that recovers your cash — sending outreach, matching remittances, posting credits — bounded by budgets your finance team sets, and reversible if it gets something wrong.
agentONE AGENT · LEDGER-BOUND
The ledger, and what keeps it honest.
Two things, not a platform: a ledger that records every action the agent takes, and a set of guardrails that decide what it's allowed to do before it does it.
Impact ledger
Append-only · two-person signed · the system of record.
Bounded autonomy
Autonomy budgets, circuit breakers, human review.
Control is the accelerator, not the brake.
Autonomy budgets
Every rule runs under a spend cap — in euros, actions, or both — that your finance lead can set, audit, and revoke. When the cap exhausts, the rule circuit-breaks to a human automatically. Blast radius is a dial, not a gamble.
Two-person attestation
Revenue-material actions require two signatures before posting to the ledger. Author + finance lead on bookings. Operator + compliance on reversals. Default on. Configurable — downward only, never up.
Reversibility windows
Every action carries a redo path. Inside the window, a single signed reversal unwinds the act, posts a credit note, and keeps both the original and reversal visible forever. Confidence comes from undo, not accuracy.
Policy-as-code
The dunning agent is governed by a signed, versioned policy file in your repo. PR-reviewable. Diff-able. Any change that widens blast radius requires two approvals from listed attestors. Your auditor sees the diff, not a slide deck.
EU AI Act, by construction
Kaiva is built around the EU AI Act (Reg. 2024/1689). Risk management, logging, and human oversight patterns are built into the product, so your technical file and conformity work sits on a compliant substrate — not a hand-rolled one.
Data residency, zero-trust
Customer data never leaves your chosen region. Model calls can be pinned to in-region inference. Every tool call is capability-scoped, short-lived, and logged. Shared-secret architecture, everywhere.
What this looks like in production.
One pattern, running today: dunning and cash application. Same policy file, same budget, same ledger — every time. Metrics shown are illustrative design targets — full attribution methodology is on the governance page.
Dunning, cash recovery
The agent reads the AR ledger hourly, ranks overdue invoices by risk-weighted recovery, and dispatches level 1→3 nudges through email, SMS, and human-review dialer. Recovery is attributed only when cash lands in the 7-day window, measured against a held-out control. Unmatched actions are visible, not hidden — they're the unit of continuous improvement.
From assessment to a proven receipt.
Our methodology is designed to prove ROI on one leak in four weeks, then extend coverage across entities and currencies the following quarter.
Impact mapping & readiness audit
We don't hunt for automation — we hunt for high-ROI friction. A real engineer works alongside your AR team. We assess data maturity, governance posture, and where the exceptions actually sit, and we scope the single leak where pilot ROI is defensible to the CFO.
Rapid validation pilot
The agent ships into a live environment, initially in suggestion mode to establish trust. We measure against predefined KPIs — your system of record, not ours — and only then do we grant execution privileges. Autonomy is granted once the KPIs hold.
Scale across entities
The validated pattern extends across legal entities, currencies, and aging buckets — same policy file, same budgets, same ledger. This is depth, not breadth: still one workflow, more of your AR running through it.
Let's look at your AR ledger.
Book a 45-minute working session with a real engineer. We'll map one candidate leak, draw the ledger shape, and tell you — on the call — whether Kaiva fits.
Book a cash leak review →