Pillar 02 · Robotic operations

The standard case is machine work.

Onboarding, settlement, client records, ticketing, conversational support and desk supervision — automated where automation is safe, and routed to a named human with full context where it is not.

<60 s onboarding <3.2 s settlement 90% straight-through 6+ languages Append-only audit
08 · Robotic KYC

Verification completes before doubt forms.

Document validation, liveness detection and screening run inline at onboarding — raising rigour while removing the abandonment point.

Layer L08

For operators

  • Revenue. Onboarding abandonment is the cheapest conversion loss in a brokerage to eliminate.
  • Risk. Screening is applied uniformly on every file, including re-screening on triggering events.
  • Capital efficiency. Compliance headcount is spent on exceptions, not on document review queues.

For technologists

  • Capture. OCR plus government-ID authenticity checks, address proof, and 3D liveness with anti-spoof scoring.
  • Screening. Sanctions, AML and PEP sources queried inline; every match written as a case with an evidence bundle.
  • Decisioning. Categorisation and appropriateness captured at onboarding; rules configurable per jurisdiction.
  • Escalation. Any positive match routes to the reporting officer instantly with the full chain attached.
<60 s
Standard case
99.8%
Anti-spoof confidence
80%
Compliance staffing

Benchmarked against a 1–3 business day manual onboarding model · staffing reduction is a design target for a fully deployed pipeline

System view · onboarding pipelineFile #884920
Document OCRMatched ✓
ID authenticityValid ✓
Liveness · 3D meshPassed 99.8%
Address proofVerified ✓
Sanctions screenClear ✓
AML watchlistClear ✓
PEP screenClear ✓
AppropriatenessAssessed ✓
CategorisationRetail applied
Elapsed00:47
DecisionInstant approval
Drop-off target ≈ 0% · exception files escalate to the reporting officer with evidence attached
Per-jurisdiction rule setsRe-screening triggersCase management
Request KYC spec
09 · Automated Settlement

Withdrawal speed is the trust signature.

Deposits and withdrawals run one pipeline: rule checks, screening, rail selection, ledger posting with four-eyes, and instant escalation of outliers to a named officer.

Layer L07

For operators

  • Revenue. The first withdrawal decides retention; a settlement measured in seconds converts doubt into a second deposit.
  • Risk. Every hold is a logged rule with a named owner — never a discretionary retention tactic.
  • Capital efficiency. Back-office clerks are removed from the standard path entirely.

For technologists

  • Rail selection. Faster Payments, SEPA Instant, card refund-to-source and a treasury rail for eligible professional clients.
  • Controls. Free-margin and open-position checks, screening re-run, velocity and device-fingerprint analysis.
  • Ledger. Client-money posting with four-eyes: machine decision plus reconciled ledger entry.
  • Escalation. Outliers by size, velocity, jurisdiction or behaviour route with full context to a risk officer.
<3.2 s
Settlement target
90%
Straight-through
24/7
Coverage

Design targets · conventional benchmark is 1–5 business days · rail availability depends on the operator's banking and PSP counterparties

System view · payment pipeline#WD-8842 · $4,800
00.00s · requestPortal · destination verified
00.41s · rule checkFree margin ✓
01.12s · screeningClear · velocity 1/24h ✓
01.90s · ledgerReconciled · four-eyes
02.90s · railSettled ✓
Rail status
Faster PaymentsUp · 2.4s avg
SEPA InstantUp · 7.9s avg
Card refundUp · same day
Treasury railProfessional only
Human-guarded share routes with size, velocity, jurisdiction and behaviour context attached
Segregation-awareIdempotent APINamed-owner holds
Request payments spec
06 · Brokerage CRM

A client record wired to the book.

Identity, verification state, funding history, trading behaviour, risk tier, support history and revenue contribution in one profile — with an immutable action log.

Layer L04

For operators

  • Revenue. Segmentation on actual trading behaviour rather than on self-declared profile fields.
  • Risk. Suitability and categorisation state travels with the record, visible at every touchpoint.
  • Capital efficiency. One system of record removes the duplicate tooling and the reconciliation between them.

For technologists

  • Events. Every layer writes to the same event bus; the profile is a projection, not a manually maintained record.
  • Automation. Trigger-based campaigns across email, SMS and push with suppression and consent state.
  • Partners. Multi-tier introducee hierarchy with automated commission calculation and payout instruction.
  • Audit. Append-only action log with actor, timestamp, reason and the rule that permitted the action.
1
Unified profile
100%
Action audit trail
Partner tier depth

Architecture properties of the deployed module · not performance claims

System view · client record#77120
Client #77120 · professionalVerified
Risk tierB · recomputed 60s ago
KYC stateComplete · re-screen due 41d
Funding · 90d$184,000 in · $62,400 out
Settlement history11/11 straight-through
Support2 tickets · both auto-resolved
Revenue contribution$4,120 · 90d
IntroducerIB-2291 · tier 2
Active automations
Dormancy · 21darmed
Tier upgrade reviewarmed
Re-screeningscheduled
Record contents illustrative · profile is projected from the event bus, not maintained by hand
Event-drivenConsent-awarePartner payouts
Request CRM spec
15 · Automatic Ticketing

Classified before a human reads it.

Intent and sentiment classification, priority routing by rule, language-matched queues, and context that travels with the ticket so no agent starts cold.

Layer L04

For operators

  • Revenue. Resolution speed correlates with retention more strongly than any incentive scheme.
  • Risk. Complaints are tagged into the dispute process automatically, with the clock started and evidence attached.
  • Capital efficiency. Priority is assigned by rule, so senior agents are not spent on password resets.

For technologists

  • Classification. Intent and sentiment on every inbound message, in every supported language.
  • Routing. P1 execution and client-money issues open the dealing desk in real time; P2/P3 queue against SLA.
  • Context. Risk tier, verification state, settlement history and relevant trade log attach automatically.
  • Measurement. Auto-resolution share, median SLA and satisfaction reported per team, per language, per module.
92%
Auto-resolved
0.4 s
Tier-1 resolution
99%
SLA met

Design targets for a fully trained deployment · actual auto-resolution share depends on the operator's product surface and documentation coverage

System view · ticket board24h window
NEW · 12
#T-5120 · settlement queryP2 · FA · auto-answered
#T-5121 · categorisationP2 · RU · → compliance
HUMAN · 4
#T-5117 · execution complaintP1 · EN · dispute clock started
#T-5119 · fee disputeP2 · AR · allocation log attached
RESOLVED · 1,846
By machine92%
Median0.4 s
By human8%
Median6 m
Satisfaction4.7 / 5
Board contents illustrative · priority rules and escalation thresholds are configured per operator
Dispute-clock awareLanguage-matched queuesContext attachment
Request ticketing spec
16 · AI Support

Sub-second answers, hard guard-rails.

A conversational layer trained on the operator's own documentation and product surface — resolving Tier-1 work in the client's language and escalating regulated topics by rule.

Layer L04

For operators

  • Revenue. Coverage in every timezone without a follow-the-sun staffing structure.
  • Risk. The assistant cannot give investment advice, override a risk decision or close a complaint — by rule, not by prompt.
  • Capital efficiency. Support seats are removed from the standard path; retained staff handle judgement work.

For technologists

  • Surfaces. Embeddable web widget, WhatsApp and Telegram, all backed by one intent engine.
  • Grounding. Retrieval over the operator's own documentation, policy set and account state — no free-form generation on regulated topics.
  • Actions. Executes password reset, settlement status, leverage request routing, allocation links and document upload.
  • Escalation. Hands over with transcript, intent, sentiment and account context attached.
6+
Languages live
<1 s
First response
0
Advice given

Guard-rail is an architectural constraint: regulated intents are routed, not answered · language coverage is configurable per deployment

System view · conversational layerMultilingual

How do I connect my account to manager #104?

✔ INTENT RESOLVED · 0.4 s · EN · allocation link attached

وضعیت برداشت من چیست؟

✔ INTENT RESOLVED · 0.3 s · FA · settlement #8842 cleared

Можно поднять плечо на золото?

✔ SMART ESCALATION · RU · categorisation review · context attached

هل يمكنني رفع مستند إثبات العنوان من الهاتف؟

✔ INTENT RESOLVED · 0.3 s · AR · onboarding module linked

Conversations illustrative · the assistant never provides investment advice and cannot close a complaint
Grounded retrievalRule-based escalationWidget · WhatsApp · Telegram
Request support spec
10 · Call Centre Monitoring

Support quality, measured not remembered.

Supervision layer for the support and dealing desk: transcript analytics, escalation queues with elapsed timers, agent scorecards and dispute-clock visibility.

Layer L04

For operators

  • Revenue. Coaching loops driven by scored interactions rather than by anecdote.
  • Risk. Complaint clocks and escalation ages are visible in real time, not reconstructed at month end.
  • Capital efficiency. One supervisor can oversee a desk that would otherwise need team leads per shift.

For technologists

  • Capture. Call and screen capture with retention policy configurable per jurisdiction.
  • Analytics. Transcript NLP for intent, sentiment and required-phrase coverage.
  • Queues. Escalation queue with routing target, elapsed timer and context bundle.
  • Reporting. Scorecards, QA sampling and trend reporting per agent, team and language.
24/7
Desk coverage
4m 12s
Median human SLA
100%
Interaction scoring

Design targets · capture and retention configuration is subject to the operator's jurisdictional obligations

System view · escalation queue3 analysts on shift
Size · #TX-4881902:14
12.0 lot · tier C · 4.1× plan → analyst 1
Velocity · #WD-885101:05
$48,000 · 3rd in 24h · new device → risk officer
Jurisdiction · #KYC-221000:38
Address change → higher-risk region → reporting officer
Closed · #TX-4880204:40
Cleared by desk · context archived
Median straight-through2.9 s
Auto-cleared · 24h90.4%
Escalations · 24h37
Audit trailComplete ✓
Queue items illustrative · four-eyes applies to every human override
Retention-configurableNamed-owner escalationQA sampling
Request monitoring spec
Integration matrix

What each module exposes.

Automation modules are only useful if they write to the same record and obey the same permission model as everything else.

ModuleLayerAPIWebhooksWhite-labelStandaloneSLA class
Robotic KYCL08RESTCritical
Automated settlementL07RESTWith ledgerCritical
CRML04REST · eventsHigh
Automatic ticketingL04RESTHigh
AI supportL04REST · WSHigh
Desk monitoringL04RESTInternalWith L04Standard

Standalone indicates deployment against a non-CoreVex stack · settlement requires an agreed ledger interface where the operator keeps its own client-money ledger

Pillar 02 · next step

Count your manual hours first.

Bring your current onboarding, settlement and support volumes. We will model which share is genuinely automatable in your jurisdiction and which must stay human — before any commercial conversation.

Book a call Architecture