How we work

Agents that explain themselves

We run reconciliation operations using AI agents. Every action has a reasoning trace. Every decision is auditable. Every escalation includes the full context. You stay in control — we do the work.

See our services

Delivery model

From scoping call to production in 3–6 weeks. No multi-year contracts. No platform licenses. You pay for reconciliation outcomes — not software seats.

1

Scoping call

30 min

We map your systems, data sources, exception types, and close calendar. You tell us what breaks. We tell you what we can run.

2

Configuration

1–2 weeks

We connect to your ERP, configure matching rules, tolerance thresholds, escalation paths, and reporting cadence. No code changes on your side.

3

Shadow period

2–4 weeks

Agents run in parallel with your existing process. Every proposed resolution is reviewed by your team before action. We tune rules based on feedback.

4

Production

Ongoing

Agents run the reconciliation. You receive daily exception reports with proposed resolutions and full reasoning. Escalations route to your team.

What our agents are not allowed to do

Every AI vendor publishes what their agents can do. We publish what ours cannot. These are hard constraints, not guidelines.

The agent cannot write to your ledger.

It proposes entries. A named human posts them. Write access is never implicit — it is granted per-system, per-action, with an approval workflow attached.

The agent cannot close a period.

Every close pack is reviewed and signed off by your team. The agent prepares the reconciliation, surfaces exceptions, and drafts the journal — but the final post is yours.

The agent cannot resolve outside tolerance.

Anything above threshold escalates with documentation attached — it never guesses. Tolerances are configurable per reconciliation type and per client.

The agent can be stopped mid-run

by any authorised user, and stops in a recoverable state. Paused agents hold context and resume cleanly when restarted — no data loss, no orphaned transactions.

The agent cannot take an untraceable action.

Every decision carries its reasoning and source documents. You can audit why it matched, why it escalated, and what data it read — nothing is a black box.

The agent cannot fail silently.

Every run reports, including runs that resolve nothing. Real-time dashboards show activity, exception rates, and resolution accuracy. Anomalies trigger alerts before they compound.

What the agent actually does

A reconciliation agent is not a chatbot. It does not generate text or answer questions. It executes a structured investigation workflow:

  1. 1Ingest — Pull data from your ERP, custodian, bank, or counterparty. Normalise formats, currencies, and identifiers.
  2. 2Match — Apply matching rules (exact, fuzzy, one-to-many, many-to-many). Flag items that don't match within tolerance.
  3. 3Investigate — For each break, pull supporting detail. Trace the transaction through the system. Identify root cause. Classify the exception type.
  4. 4Resolve — Propose a resolution with full reasoning. Route for approval if above threshold. Auto-resolve and log if within tolerance.
  5. 5Report — Deliver a reconciliation summary with matched items, resolved exceptions, open items, and ageing. Every item links back to its reasoning trace.

Audit and compliance

Reconciliation is a control function. The output needs to survive audit scrutiny — internal and external. Here's how we ensure it does.

Immutable audit log

Every action, decision, and data access is logged with timestamp, agent ID, and reasoning. Logs cannot be modified or deleted.

Quarterly review packs

We deliver structured review packs showing agent performance, exception patterns, and control effectiveness — ready for internal audit or external review.

SOC 2 Type II

Our infrastructure and processes are independently audited against SOC 2 Trust Service Criteria. Reports available under NDA.

Data residency

Processing and storage in your chosen region. We support UAE, EU, US, and Singapore data residency. No data leaves your designated geography.

End-to-end encryption

All data encrypted at rest (AES-256) and in transit (TLS 1.3). Customer-managed encryption keys supported for regulated environments.

SSO & access control

SAML 2.0, OIDC, and Active Directory integration. Role-based permissions, enforced MFA, and session policies across your organisation.

What we reconcile

Family office reconciliation

Positions, cash, corporate actions, and NAV across custodians and accounting systems.

Invoice reconciliation

Three-way match, price variances, quantity mismatches, and GL coding errors.

Accounts payable reconciliation

AP subledger to GL, vendor statements, accruals, and month-end close.

Ledger reconciliation

Subledger-to-GL, intercompany eliminations, timing differences, and cut-off errors.

See it in action

Tell us what you reconcile, what breaks, and what systems you use. We'll show you what the first cycle looks like — with full reasoning traces on every resolution.