AIState Compliance™
The whole compliance unit, on one living graph.
AIState Compliance reads every system you already run, resolves your customers once, turns your policy into executable process, and puts agencies to work across onboarding, screening, adverse media, reviews, investigations and remediation — with your people deciding exactly where your authority matrix says they must. Live in weeks, inside your boundary.
4 weeksfrom first connection to a working Monday
93 minutesfor a corporate onboarding that took days
71 of 412overnight alerts needed a person
Same dayquality challenge on every closure
One platform for every function in the unit.
Not a tool per queue. One resolved estate, your policy as process, and agencies that do the work — so what one function learns, every other already knows.
One estate, resolved once
Every system you run and the outside sources you rely on, read in place and resolved once — then reused by every function in the unit.
Your policy, executed
The DAG engine breaks your policy into steps; the process engine runs them with SLAs, gates and named transitions.
Agencies that do the work
Composed in the Knowledge & Agentic Studio, working at the autonomy you sign — with people at every decision your authority matrix reserves.
Built to be examined
Every claim cited to its source, every re-plan versioned, every policy change replayed on history, every case carrying its decision record.
Your policy fires. It just fires on fragments.
Financial crime hides in the gaps between your systems — and a policy running on one fragment can’t see the relationship between them.
- Twelve copies of the same customer, reconciled by people
- The same alias cleared again next week by another analyst
- Periodic reviews re-performed in full, months after the world changed
- Remediation files taking hours while the evidence sits elsewhere in the bank
- A spouse two levels up the ownership chain that nobody sees
- A counterparty in one case that is the parent in another — and the cases never meet
From your systems to a governed decision.
Five layers, one flow. Every system of record stays where it is, and stays authoritative.
Your sources
read in place
- Core banking and payments
- Onboarding and CRM
- Documents and case systems
- Company registers
- Sanctions, PEP, watchlists
- Adverse media
Knowledge Fabric
one resolved estate
- Discover and catalogue
- Resolve entities
- Derive the ontology
- Extract knowledge
- Govern every source
Studio and process engine
your policy, executable
- Knowledge & Agentic Studio
- DAG policy engine
- BPMN process engine
- Policy replayed on history
Agencies
do the work, show their working
- Monitoring
- Screening
- Investigation
- Corporate Structure
- Reach-out
- Quality
Your people
where authority sits
- Team lead
- Analysts
- MLRO
- Second line
Every compliance function, on the same estate.
Each capability below runs on the platform today — and each is shown working in the day further down.
Onboarding and due diligence
Retail and corporate onboarding, CDD and EDD — with everything the bank already knows about the customer arriving before the case is opened.
Screening
Sanctions, PEP, watchlists and adverse media on the resolved entity, with decisions carried forward when nothing has changed.
See screeningCorporate structure and UBO
Ownership and control walked source by source to the beneficial owner, across 400+ global sources.
Perpetual KYC
Risk recomputed whenever the graph changes; only material change is pushed, each with its lineage.
Investigations
An agency that plans its own path from your policy, re-plans as evidence arrives, and hands exceptions to the right person.
Customer outreach
Requests, chasing and responses handled by an agency that explains exactly what evidence would satisfy the requirement.
Remediation
Files closed on evidence the bank already holds — including evidence sitting on a related customer’s file.
See the agenciesQuality control
Every closure challenged the same day; recurring gaps become versioned policy changes a person approves.
MLRO and second line
Restricted routes, governed views, incidents raised against case state, and cases that inform each other.
Screening the customer, not the string.
Screening runs on the resolved entity, so the same namesake is never cleared twice and a decision travels with the customer across every channel.
- 147 sanctions, PEP and watchlist sources, plus your own internal lists
- Decisions carried forward when nothing about the customer has changed — and reopened the moment the evidence does
- Scheduled batch screening that notifies only on newly flagged customers
- Millisecond screening where the operation needs it
Three of yesterday’s clearances were reopened because new evidence contradicted them. Illustrative overnight run.
Illustrative feed: articles in many languages, matched to the resolved entity — namesakes set aside with their reason.
Adverse media from our own search engine.
Not a keyword alert service. Coverage is ours, and every article is matched to the resolved customer — so namesakes fall away and real exposure doesn’t.
- 100K+ news sources and gazettes in 110+ languages
- Matched to the entity, using everything the estate knows: directors, addresses, ownership, identifiers
- Every hit cited to its article and graded, with namesakes recorded and their reasons kept
- Continuous, feeding perpetual KYC and event-driven review
Client lifecycle management
The whole client lifecycle.
One client, resolved once.
AIState Compliance runs the client lifecycle end to end — from the first screen of a prospect to the record of an exit — on one resolved estate. It does not become another system of record for the client. It resolves the client across the systems that already are, and every stage runs on that.
Already running a CLM? Keep it.
AIState resolves it as one more authoritative source, fills in what it cannot see — the rest of the estate, Market Data, adverse media and the institution's own case history — and writes back wherever your policy says it should.
Client lifecycle management without another system of record.
CLM as a new system of record
AIState Compliance
Mapped to what the lifecycle answers to.
Indicative mapping for UK and EU regimes. Your policy defines the obligations; AIState executes the policy you load, per jurisdiction, and records which version decided each case. Leaving the platform is planned for too — under PRA SS2/21 and DORA, the systems of record never moved.
Named deployments, the architecture and assurance dossier and the evidence pack are shared under NDA.
Book a sessionAgencies do the work. People decide what policy reserves.
Each agency has one job and works at the autonomy level you set for the process, one of five, changeable at any time. Your operating policy is what they obey — they don’t invent policy of their own.
Monitoring
Watches the graph continuously, recomputes risk when it changes, and pushes only what is material.
Screening
Screens the resolved entity, not the string, and carries a decision forward when nothing underneath has changed.
Triage
Separates signal from noise and says why.
Investigation
Builds the case from the graph, plans its own path from your policy, re-plans as facts change, and hands exceptions to the right person.
Corporate Structure
Walks ownership and control, source by source, until it reaches the beneficial owner.
Narrative
Drafts every pack, request and notice from your templates, with every claim linked to its source.
Reach-out
Owns the conversation with the customer, and explains exactly what evidence would satisfy the requirement.
Quality
Challenges every other agency’s work the same day, and proposes policy changes rather than quietly learning.
Your policy, executed — not re-keyed into a workflow tool.
The DAG engine consumes your policy, whatever its form, and breaks it into steps and dependencies. The process engine runs them — and re-plans when the evidence changes.
§4.1 No risk rating may exist before the prohibition check. §4.2 Screen the entity and every connected party. §5.3 Verify ownership to the beneficial owner from primary evidence. §6.1 A PEP relationship requires second-line approval before routing. §7.2 Approval by a named person; the reason is recorded.
- Policy to process: steps, dependencies and precedence derived from the text — no rating before the prohibition check
- BPMN, standard modeler and engine, with SLA timers that escalate
- Re-planning within policy: every version of the plan kept and replayable
- Human gates and two-person approvals; every status change a named transition with its reason
- Replayed on history before any policy change goes live
- Parallel branches and conditions, deterministic where the policy demands it
The Knowledge & Agentic Studio.
Where your experts turn what they know into agencies and chains the bank owns — in words, not code.
high?
two people
Planned from one sentence — then traced and metered on every run.
- Agencies are configuration: purpose, context, tools, output, effects and model — published without a deployment
- Plan with AI: describe the outcome in a sentence; the Studio drafts the agency or the whole chain
- Chains with logic: tool steps, conditions, parallel branches and human gates — agentic or propose-only, step by step
- Traced and metered: every run’s model calls, tools, timings and cost on the record
- Every write governed: applied within policy, proposed where you reserve the decision
- Automation: policies, playbooks, schedules and system agencies on events
See it run: one Monday in the unit.
An illustrative bank and illustrative data, run on the real platform from half past eight to five. Pick a moment to see the screen, what happened, what the Fabric pushed unasked — and who decided.
08:30
The day starts itself
Overnight the Monitoring agency recomputed risk across all 4,012 customers: 37 registry events, 412 screening alerts, 9,300 transactions. Ruth, the team lead, doesn’t arrive to a list of four thousand cases — she gets the handful that changed materially, each with its lineage and an owner.
A person decidesRuth assigns the morning’s work.
08:40
One customer, not twelve records
Records from across the bank — the registry, CRM, trade finance, the screening vendor, core banking, payments — under half a dozen spellings of the name. Eleven resolve into one customer, every match with its source. One the Fabric won’t merge — similar is not the same.
A source system sees its record. The graph sees the customer.
A person decidesA person decides the record the Fabric held back.
08:45
He is not new
A new retail application — but the person isn’t new. Two older records resolve to him — a dormant savings account under “A. Balogun” and a 2023 card application that was declined — and they follow him into the new case, so nobody starts from nothing.
Pushed unaskedFound before Priya opened the case.
A person decidesPriya confirms; Tomasz approves.
09:15
Who is really behind the client?
A corporate onboarding the relationship manager needs by tomorrow. Nobody drew its workflow: the Investigation agency builds the plan from your policy and re-plans as evidence arrives — a lapsed licence, an offshore holding company, a director who is a PEP, then a certified share register showing the holding company is owned by her husband.
Pushed unaskedAt 09:16, before anyone opened the case, the Fabric linked the majority shareholder to another customer under a sanctions match: same address, same director.
A person decidesThe MLRO approves the PEP subject to enhanced due diligence — and refuses the relationship manager’s request for a provisional clear.
10:48
Ninety-three minutes, every step replayable
Three to five days becomes ninety-three minutes, end to end. Every re-plan is versioned on the case, and the case shows exactly what each chapter added and when.
A person decidesTomasz closes the case.
Provisional clear refused
A trade goes through
Case cleared on its merits
Incident raised to the second line
15:00
The incident nobody had to look for
One minute after the provisional clear was refused, a foreign-exchange trade went through for the client. The case was later cleared on its merits — but the restriction existed at 10:31, so the incident stands. Found while the operation is running, not at audit.
Pushed unaskedMonitoring watches transactions against case state and raised it unasked.
A person decidesIt goes straight to the second line.
15:30
The agency that checks the others
The Quality agency challenges every closure the same day, against the requirements your policy sets. In Aurelia a party has no screening from the last six months — a blocking requirement, so nothing can be cleared until it is met. Every other requirement shows how it was met: by the case’s own facts, or attested by a named person against the evidence that came back. When the same gap keeps recurring, the agency proposes a versioned policy change.
A person decidesDaniel classifies the finding; Ruth approves the change.
16:30
Cases that inform each other
The MLRO doesn’t look at a queue. He asks which cases are informing each other today. The book draws every case and the connections between them: Aurelia Fintech and Helios Trade, joined through a director the two companies share — a link no person drew, carried into both cases.
Pushed unaskedWhat you’re not permitted to see is absent, not dimmed — the same book, governed views.
A person decidesThe MLRO sees the whole picture because he’s on the restricted route.
17:00
The day as numbers
2,041 cases closed. 1,927 by agencies alone. 114 with a person — exactly where the authority matrix put them. 71 human decisions on 412 overnight alerts. Every closure challenged the same day, and every case carrying a six-element decision record a regulator can read without the analyst in the room.
A person decidesKarim, head of operations, reads the day — and adjusts the autonomy level of tomorrow’s processes.
Demonstration recorded on the live platform. The bank, its customers and every figure are illustrative.
Built to be examined.
When the regulator asks why a decision was made, the answer is on the case — not in somebody’s memory.
- Every match keeps its evidenceand every merge is reversible. Uncertain matches go to a person.
- Policy replayed before go-liveEvery policy change is replayed against your history first.
- Refusals are decisions toorecorded with their reasons, like approvals.
- Absent, not dimmedGoverned views: what someone may not see isn’t there at all.
- A six-element decision recordon every case, every claim linked to its source.
- Both scores keptWhen quality fails a case, the fail and the fix stay on the record.
Lower cost across the whole life of the platform.
Each stage against market-standard cost.
Bring it in
40–50% lowerWeeks, not quarters — read in place, with no migration programme.
Build and extract
75–85% lowerYour people and your partner compose it. No standing team of vendor engineers.
Run it
45–65% lowerDepending on volume — the agencies carry the work, people take the decisions.
Leave it
90% lowerSystems of record were never touched. Export the identity layer, policy, decisions and audit trail.
Illustrative. Run reduction varies with volume; every figure is validated against your own estate in the proof of value.
Inside your boundary
Deployed in your cloud, on your premises or in a sovereign environment. Raw records never leave.
Any model
Bring your own language model, under a data boundary you set for each provider.
Delivered your way
Directly, or through partners who run their own propositions and managed services on the platform.
Start with one process. Usually screening.
- Prove it on your own data, inside your boundary
- Connect your estate in place and resolve it
- Encode your policy and replay it on your history
- Run the agencies at the autonomy you sign — with us beside you