DataFab / Industries
Where it is proven
Six regulated estates, examined.
The beachhead is deliberately the estates where the data cannot move, the work is regulated and repetitive, and every decision must survive an inspector, a court or an oversight body. If it holds there, it holds anywhere.
01 · banking & financial services
Financial crime & compliance
The unit sees fragments. DataFab unifies every source in place, derives the schema and ontology, and assembles a dynamic knowledge graph of customers, counterparties and money flow. Governed agencies then run over it — from detection to defensible narrative.
Because the graph already knows what an entity is, rules express intent rather than brittle joins. “Funds in from a high-risk counterparty, out within 24 hours” is written against meaning. Change a threshold, replay history, see the effect — before it reaches production.
Connect everything, move nothing
Core banking, payments, CRM, screening, trade finance, documents, warehouses and streams — read in place, resolved on the way back.
Twelve copies, one customer
Ownership chains, beneficial owners and counterparties become traversable structure rather than rows to reconcile by hand.
Agencies over the graph
Monitoring, screening, triage, investigation, narrative and quality — each at the autonomy its risk allows.
When the regulator asks why
The answer is the artefact, not a reconstruction: every statement cited, every disposition logged, every decision replayable under the rules in force at the time.
02 · defence & intelligence
Every source. Every warning. Nothing leaves.
The hardest failures in this mission are not failures of collection. The signal is usually already there, in fragments, across disciplines, discounted at a single desk. DataFab draws every source you already hold into one governed picture, tests it continuously against the threats you are watching for, and forces what matters into a warning that cannot quietly disappear. All of it inside your enclave.
Nothing leaves
The picture is real; the data movement is zero. Every source read in place and resolved inside your enclave — the one all-source capability you can actually run on classified ground.
The warning that cannot be buried
Weak signals are scored continuously across the estate and accumulate into an institutional warning that resurfaces on new evidence. It cannot die inside one analyst’s judgement, a silo, or a prevailing assumption.
A virtual team, commanded by dialogue
One analyst directs a governed team of specialists that does the enormous, complex work — in conversation, not queries. The reach of a unit, from a single seat.
The floor — assumed, native, present on day one
Institutional de-biasing
An analyst defines a future threat hypothesis. The system decomposes it into observable indicators, tests the agency’s entire internal and external estate against it continuously, accumulates corroborating evidence across disciplines, preserves the contradictions, maintains the hypothesis over time, and escalates it into an institutional warning.
A warning should not disappear because one person does not believe it.
The system does not remove human judgement — it removes the ability for important evidence to disappear inside itContradictions preserved
Evidence for and against is kept, so the assessment stays honest rather than one-sided.
Confidence over time
The score is not fixed at one assessment. It moves as evidence arrives, up or down.
Institutional escalation
At the threshold it becomes an agency-level object demanding acknowledgement, escalation, or an explicit decision not to act.
Resurfaces
New signal reopens it, and a more senior person must disposition it again. It cannot quietly die.
Horizon scanning
Most tools react to known threats. A researcher defines a new scenario, risk or typology and reverse-engineers it into the smaller, observable indicators that would signal it — and the system then watches for them across every source, measuring every new piece of data and every new ontology concept against every live scenario.
Each indicator is answered with policy before a byte is collected: which source holds it, its credibility, the authorisation required to task it, its format and frequency, and whether the data is already within reach. New scenarios, sources, concepts and typologies fold back into the semantic layer — the system extends its own reach.
Map everyone linked to vessel MV Orwell across the last 90 days.
DataFabResolving across signals, port registries and open sources — 47 entities, 6 relationships. 3 uncertain matches flagged for your review. The imagery pull needs your authorisation before I task it.
YouApprove the imagery. Prioritise the financial links.
DataFabImagery tasked and logged to you. Two front companies share a beneficial owner at confidence 0.91. Building the network now, and I will watch for new movement while you read. illustrative dialogue · every agent inherits your clearance and acts only on what you may see
Air-gapped
Full functionality inside an isolated network; agent-based and outbound-only where a boundary is crossed at all.
On-premises
Inside your data centre, reading in place to the systems you already run.
Sovereign cloud
Your accredited environment, your keys, your jurisdiction.
Tactical edge
Deployable forward and disconnected — the picture travels, and the guarantees travel with it.
03 · insurance
Claim handling, finished.
Extraction stops at the start. Tagging a claim for a small per-case fee is worth exactly that, because extraction alone cannot finish a written claim — and a manually-settled case costs a large multiple of the extraction fee, fully loaded.
The value is not the extraction slice. It is the fallout pool. Every claim moved from manual to straight-through removes a manual case, and the right measure of any solution is one number: how far it shrinks the fallout rate.
| Scenario | Fallout falls to | Straight-through rises to | Manual workload removed |
|---|---|---|---|
| Conservative | 45% | 55% | −31% |
| Base case | 35% | 65% | −46% |
| Ambitious | 25% | 75% | −62% |
Illustrative model output from a worked example, against a 65% starting fallout rate. Not a quote, and not a measured result — real figures are set per engagement against your case mix and volumes.
In place, nothing copied out
The claims decisioning core is adapted, not replaced. Data stays in the policy administration system, the intake channel and the document store.
Confidence per step
One confidence per decision, many per claim, rolled up on the case — not one number for the whole document.
Source per statement
A caseworker can verify any statement in seconds; an auditor can replay any decision end to end.
Missing information handled
Reach-out is drafted, the response re-enters through intake, and the case re-evaluates automatically — no manual re-keying.
04 · legal & professional services
The firm as the platform.
A firm’s knowledge is spread across matter management, the document system, the practice-management estate, the finance ledger and the heads of the people who did the work. None of it is reachable as one question, and none of it survives the person leaving.
DataFab resolves matters, clients, counterparties and documents into one governed graph over the systems the firm already runs — and then lets the firm encode its own expertise once, own it, and deploy it across every client engagement. Be the platform, not the reseller of somebody else’s.
Intake to renewal
Parties, terms and dates extracted; obligations tracked and alerted; clause-level review kept governed.
The implication nobody wrote down
A directorship in a filing plus a shareholding in a register plus a matter in the document system — together they put the firm on both sides. That conclusion exists only between the records.
A wall around knowledge
A document wall asks whether you may open something. A knowledge wall also asks whether you may reconstruct what it would have told you — governed at entity and relationship types, with inference and aggregation bounded by the same policy. See ontology governance →
Precedent that finds you
A change in a client’s world, resolved against what the firm already knows, delivered as a brief rather than a search result.
DataFab does not train on client data. Client work product, matter content and privileged material are never used to train models — the value comes from resolving what the firm already holds, in place, under the firm’s own controls.
05 · consumer redress
Remediation at population scale.
A redress programme is a data problem wearing a legal deadline. The affected population has to be identified across systems that disagree, the calculation has to be deterministic and reproducible, the outreach has to happen at volume, and the whole thing has to be defensible on the day it is delivered.
Cohorting
Affected populations identified across the estate, with the basis for inclusion recorded per customer.
Calculation
Redress computed deterministically — fixed steps, reproducible, no model discretion in the flow.
Outreach
AutoFab runs contact at volume, with the record of every attempt and every response attached to the case.
Packaging
Audit-ready output produced as the work happens, rather than reconstructed against a deadline.
06 · government & public bodies
A record that survives scrutiny.
The hardest question in public-sector casework is not what the system found. It is whether you can defend how you found it. Every judgement carries its evidence, its authority and its lineage, so the finished assessment and the record behind it are the same artefact.
Tiered approval
Sensitive requests clear at higher authority. Access is denied without it — and the denial is logged too.
Deconfliction
Requests are checked against live and historical casework, so one team never blindly steps on another.
Overlap detection
When two records share an entity or a location, the evidence is presented side by side and a person decides: merge, or keep separate.
Oversight-ready
Who saw what, who decided, on what evidence, and when — recorded as it happened rather than reconstructed after the fact.
Next step
Bring the workflow that has to hold up.
The first engagement is one urgent use case and a handful of critical sources, deployed where the data already lives — cloud, private cloud, on-premises or air-gapped.