Not a tool. Not a hire. A new kind of analyst.
Until now there were two ways to get work done. This is the third — which is why capacity is no longer tied to headcount.
Excel, OMS, RPA
Deterministic and exact — but no judgment, and capacity is fixed by licences. Rules break the moment a case needs interpretation.
Analyst / ops team
Judgment and context — but capacity is bound to headcount, it doesn't scale, and consistency is hard to audit.
Powered by agentic AI
Judgment on an exact, auditable core — managed like a person, scales like software. Onboarded in days, then cloned across every book.
Build once. Configure for any process.
We didn't build six tools — we built one governed engine and configured it six times. The engine is fixed and proven; a new process changes only the configuration layered around it, never the core.
The engine
The same core powers every process — never re-engineered for a new one.
- Deterministic core — the exact work: calculations, validations, lookups
- Agentic reasoning — classification, diagnosis and judgment over the evidence
- Human-in-the-loop — the system proposes; your expert signs off
- Audit & evaluation — every run logged, every finding gate-scored
A pack
Everything specific to one process — supplied as configuration, on the unchanged engine.
- Taxonomy — the data it reads, and how each feed is recognised
- Rule library — the checks, and how each finding is judged
- Escalation owners — who each finding is routed to
- Pipeline — the deterministic + agent stages, declared not coded
- Gate-scored test set — proof it catches what it should
Client Onboarding & KYC went live as a pack: a watchlist + applicants taxonomy, a seven-rule library (sanctions, PEP, adverse-media, CDD completeness, beneficial ownership…), Compliance / MLRO and Onboarding Operations as escalation owners, a screen → adjudicate → report pipeline, and a gate-scored dataset — every planted issue caught, and a look-alike sanctions hit correctly cleared as a false positive. The core engine was not touched.
Judgment from AI, certainty from code, control from people
AI on its own isn't auditable; code on its own can't reason. Together — with oversight — they make a digital analyst you can actually trust.
Deterministic core
The exact, repeatable work — calculations, validations, lookups, thresholds — stays in code, so it is always correct and fully auditable.
Agentic reasoning
Everything that needs interpretation — classification, diagnosis, decisions, free text — is handled by AI agents that reason over the evidence.
Human in the loop
People review, approve and escalate. The system proposes; the expert decides. Nothing material clears without sign-off.
The same five-step pattern, whatever the process
Every pack runs this same pattern on the same engine — only the configuration differs. That's why a new process is days of setup, not a build.
Ingest & validate
Pull data from source systems and confirm it is complete and well-formed.
Detect & analyse
Scan for anomalies, breaches and gaps — exact checks plus judgment.
Diagnose & decide
Group findings, reason to root cause, and propose the resolution.
Act & report
Produce the fix, the audit trail and the report — for human sign-off.
Monitor & learn
Track issues until resolved; carry context forward across the cycle.
Six processes live — each a configuration, not a rebuild
Every one of these was added to the same engine as a new pack, with zero changes to the core — and each ships with a gate-scored test set. Run any of them end-to-end.
Cash Reconciliation
Book ledger vs custodian — exact match, then agentic triage, investigation and synthesis of every break.
Performance & GIPS Validation
The month-end equity-mandate close, day by day, aged to a clean factsheet gate.
NAV Oversight & Pricing
Validate prices and recompute each fund's NAV vs the struck NAV, then diagnose every break.
Mandate / Guideline Compliance
Test each portfolio's exposures against its IMA limits — then judge each breach: active violation, passive drift, or a restricted holding.
Fee & Expense Validation
Recompute management & performance fees from the contract and validate what was booked — then judge each gap: client overcharge, manager leakage, or a legitimate waiver.
Client Onboarding & KYC Review
Screen applicants & beneficial owners against sanctions / PEP lists, score risk and resolve ownership — then adjudicate each alert: true match, homonym false positive, or needs EDD.