01
Signal grammar.
Behavioural patterns are represented at signal level, so detection can operate across exchanges rather than isolated messages.
Hidden pattern intelligence
Signal-level detection of harmful patterns in digital conversations.
Iris is a deterministic signal-processing engine with confidence scoring. Built for institutions that need auditable answers, not generative guesses.
Get in touch- 8
- Patent families
- 43
- Verticals
- 17
- Source jurisdictions
- 1,000+
- Reference authorities
Coercion, exploitation and fraud unfold over time. Iris detects the cumulative pattern.
The gap
Why current tools miss the pattern.
Conversation analytics, transaction monitoring and case review tools share one structural blind spot: they look at events one at a time. Harmful behaviour often appears only when exchanges are read as a pattern across messages, transactions, channels and time.
Iris detects cumulative patterns that single-event analytics cannot see.
The engine
A signal-level architecture for auditable detection.
Iris is deterministic, confidence-scored and not generative AI. We do not generate content. We do not classify people. We process signals and surface patterns that humans then act on.
02
Scoring layer.
Each output carries calibrated confidence. The same input produces the same answer, so results are reproducible.
03
Evidence trail.
Every detection can be traced back to the source signals and the regulatory obligation it helps evidence.
The output
Six behavioural dimensions.
Every detection maps to one or more behavioural dimensions, creating a consistent language for patterns across financial services, safety, abuse and exploitation contexts.
01
Control intensity
Monitoring, permission-seeking, decision dominance, information gatekeeping.
02
Manipulation sophistication
Reality distortion, narrative construction, emotional weaponisation.
03
Exploitation severity
Financial drain, resource depletion, identity theft, data harvesting.
04
Escalation trajectory
Changepoint detection, cycling patterns, severity rate-of-change.
05
Network manipulation
Isolation, ally recruitment, support-network disruption.
06
Evasion sophistication
Coded language, channel-switching, evidence destruction.
Iris ASM
The authorities selection process.
Iris Authorities Selection Methodology is the criteria-based process by which Iris identifies, assesses and classifies every reference authority that could bear on a behavioural-signal detection layer.
Builds the universe.
Candidate authorities are identified per vertical, top-down from primary regulators with enforcement authority, then broadened by onward citation.
Classifies the authority.
Each authority is classified by the same deterministic rule: KEEP, HOLD or REJECT.
Records the criterion.
Every classification is recorded against a named criterion, making the output reproducible by a third party.
| Outcome | Meaning | Effect |
|---|---|---|
| KEEP | Relevant to the behavioural-signal detection layer. | Informs claim language, pilot scope, go-to-market framing or regulatory engagement. |
| HOLD | Not currently active or out of scope for now. | Carries an explicit revisit trigger; reviewed at the next ASM run. |
| REJECT | Fails one of seven REJECT criteria. | Rationalised against the named criterion; substitute authority noted where applicable. |
- 17
- Source jurisdictions
- 11
- Deployment-scope jurisdictions
- 70+
- Regulatory instruments
- 7
- REJECT criteria
Regulation
Built to evidence named obligations.
Iris positions pattern detection against the regimes institutions already need to evidence, without disclosing implementation detail.
| Regime | What Iris evidences | Anchor |
|---|---|---|
| FCA Consumer Duty | Detection of vulnerability outcomes including the four FG21/1 drivers; auditable evidence of the firm acting in good faith. | PS22/9, FG21/1 |
| FCA financial crime | Pattern detection across communications and transactions; APP-fraud safe-account narratives; investment scams. | FCG, JMLSG, PSR APP rules |
| Online Safety Act | Cumulative-pattern detection in user-to-user services; risk-assessment evidence at the platform level. | OSA Part 3, Ofcom codes |
| Data protection | Lawful-by-design architecture; controller-retained data; auditable purpose limitation. | UK GDPR, DUA Act 2026 |
| EU AI Act | Determinism, traceability, post-market monitoring; evidence appropriate to the risk classification of the deployment. | EU AI Act, Annex III |
| Modern slavery | Pattern detection in recruitment, debt bondage, and document control. | Modern Slavery Act 2015 s.54 |
| Online safety equivalents | US, EU, and Australia online-safety duties addressed under the same engine. | KOSA (US), DSA (EU), eSafety (AU) |
Coverage
All 43 verticals across two books.
The catalogue is public at typology level: V-number, vertical name and a short hook. It shows breadth without exposing protectable mechanisms.
Financial services book - 17 verticals
| Code | Vertical | Typology hook |
|---|---|---|
| V1 | Retail banking vulnerability | Behavioural drift across products and time |
| V2 | Motor finance | Coached affordability, ghost-broker referral, post-delivery uncontactability |
| V3 | BNPL and consumer lending | Cross-provider density, rapid drawdown, resale-marketplace correlation |
| V4 | Mortgage and property | Email-thread takeover at conveyancing, late bank-detail change |
| V5 | Pension and wealth | Cold-contact breach, non-standard underlying, illiquid lock-in |
| V6 | Regulated advice | Provider-concentration in adviser book, induced churn |
| V7 | General insurance | Staged claim, exaggerated claim, ghost broker |
| V8 | Savings and investment vulnerability | Behavioural drift in long-held positions, induced switching |
| V9 | APP fraud | Safe-account / impersonation pattern across channels |
| V10 | Open banking and economic abuse | Coercive consent reuse, account-aggregation surveillance |
| V11 | Crypto and DeFi | Pig-butchering, crypto investment build-and-extract |
| V12 | Cross-regulatory convergence | Single actor pattern visible across regulator silos |
| V13 | Market manipulation and securities fraud | Coordinated trading, narrative reversal across forums |
| V14 | Romance fraud (FS overlay) | Targeting, rapport, crisis, money request, escalation |
| V15 | Investment fraud | Clone-firm, boiler-room, recovery-room build-and-extract |
| V16 | Insurance fraud | Ghost broker, staged claim, exaggerated claim |
| V17 | Pension scams | Tracing pretext, transfer-out, illiquid lock-in |
Domain book - 26 verticals across safety, abuse, and exploitation
| Code | Vertical | Typology hook |
|---|---|---|
| V18 | Housing and cuckooing | Vulnerable tenancy taken over by exploiters |
| V19 | Identity fraud and synthetic identity | Credentials reused under variant data across applications |
| V20 | Authorised but manipulated transactions | Customer-authorised under coercion or false pretence |
| V21 | Trade credit and invoice fraud | Supplier-impersonation late in payment cycle |
| V22 | SME and commercial vulnerability | Director-level coercion in business banking |
| V23 | Charity and faith sector | Coerced donation, induced gift, leadership coercion |
| V24 | County lines and child criminal exploitation | Recruitment, grooming-into-crime, debt-bondage |
| V25 | Gambling harm | Velocity, chasing-losses, cross-operator rotation |
| V26 | Substance abuse and exploitation | Coercive supply, financial dependency, isolation |
| V27 | Radicalisation and extremism | Recruitment funnel from open to closed channels |
| V28 | Peer bullying (online) | Pile-on dynamics, sustained targeting, exclusion patterns |
| V29 | Workplace harassment | Sustained pattern recognised against published instruments |
| V30 | Modern slavery and human trafficking | Movement, debt-bondage, document-control patterns |
| V31 | Sextortion (adult) | Build, capture, threaten, extract |
| V32 | Stalking | Sustained surveillance, network manipulation, escalation |
| V33 | Domestic abuse | Cumulative pattern across communications and money |
| V34 | Coercive control | Monitoring, permission-seeking, isolation, decision dominance |
| V35 | Elder financial abuse | Power-of-attorney misuse, beneficiary-pressure pattern |
| V36 | Image-based sexual abuse | Capture, threat, distribution, coordination |
| V37 | Tech-facilitated abuse | Device, account, and platform pattern across surfaces |
| V38 | Child sexual abuse and grooming | Access, contact, normalisation, isolation, escalation |
| V39 | Honour-based abuse, forced marriage, FGM | Family-network coercion, travel-and-isolation pattern |
| V40 | Harassment and hate crime | Sustained pattern recognised against named protected characteristics |
| V41 | Fraud and social engineering (cross-typology) | Common patterns across multiple vertical contexts |
| V42 | Financial and ransom extortion | Build, threat, payment-instruction, escalation |
| V43 | Radical group recruitment and cults | Recruitment funnel, isolation, financial control |
Where it fits
Institutional and frontline deployment.
Institutional deployment.
Banks, platforms, and regulated firms.
Iris integrates with organisations that need to detect coercive, fraudulent, exploitative or manipulative patterns at scale, with audit-grade evidence and full data control retained by the institution.
Frontline deployment.
Advocacy, oversight, and research teams.
Iris also supports organisations working on harm directly. The same engine, applied diagnostically, helps surface patterns in evidence that would otherwise be invisible to a single reviewer.
Patent and IP
Patent pending across multiple families.
Iris is patent pending across multiple families covering signal-processing approaches to behavioural pattern detection in digital conversations.
Founder
Caroline Wells - Founder.
Caroline brings 25 years' financial services experience to Iris, with deep expertise in conduct, risk, and consumer outcomes. She founded Iris after seeing how poorly current systems detect the patterns that matter most: behaviour that unfolds across conversations, not within them.
Contact
Get in touch.
Tell us who you are and what you are working on. We reply within five working days.
Confidential enquiries only - please do not share personal data of third parties in this form.
Thank you.
Your enquiry has been received. We reply within five working days.