Case study · Licensed EMI · 2025
TreasuryGrid
B2B bank goes AI-native, without losing audit trail.
A licensed EMI serving B2B clients embedded AI across compliance, onboarding, and ops — every agent run logged, every prompt reviewed, every irreversible action gated.
At a glance
- 180Headcount
- 4 weeksEngagement
- 8Pioneers
- 3Compliance reviews
- 1Marketplace
The brief
AI rollout inside a regulated boundary, without weakening it.
Cohort
- C-level (CEO, CCO/MLRO)2
- Heads of Department4
- AI pioneers8
Drawn from KYB, AML, Settlements, B2B Support, and Engineering — the functions where compliance and customer experience meet daily.
What the pioneers shipped
Auditable agents — useful before they're loud.
KYB intake triage
Agent ingests company docs, drafts a risk score, and flags discrepancies for the onboarding queue.
AML alert investigation
Pulls case context, drafts the SAR narrative, links similar past cases — analyst still signs off.
Treasury reporting
Daily P&L commentary with variance flags, broken into client-segment views — circulated by 9am.
B2B portal support
Tier-1 questions on transactions, fees, and limits — answered with audit trail, escalation if uncertain.
Engineering knowledge agent
Internal RAG over engineering docs, runbooks, postmortems — cuts onboarding from weeks to days.
“Auditable agents, not just useful ones.”
Approach
Four weeks. Three sign-offs. One marketplace.
Discover
Mapped the compliance perimeter, audit trails, and model-data flows. Defined what could leave the boundary and what couldn't — with the MLRO and DPO in the room from day one.
Build
Pioneers built first cuts in their own functions. Compliance reviewed every prompt, every tool call, every data flow — not just the end product.
Audit
Three demos with MLRO and CCO sign-off. Anything not auditable was redesigned, not deployed. Anything irreversible kept a human in the loop.
Embed
Patterns shipped to an internal marketplace with governance docs, change-log, and an owner per skill. Compliance reviews stay attached for the lifetime of each agent.
Impact · Year one
Outcomes leadership signed off on.
Annual savings
€1.4M
Across the engagement's first twelve months.
FTE optimized
12
Full-time equivalents freed for higher-leverage work.
Business satisfaction
9.5 / 10
Post-engagement leadership review.
Why it didn't go off the rails
Compliance designed in, not bolted on.
No customer PII to external models
Local or private inference for sensitive flows. Egress controls baked in, not bolted on.
Auditable prompts and runs
Every agent run logged with input, output, tool calls, and timestamps — same shape regulator inspections expect.
Compliance review per skill, not per project
Each reusable agent has its own compliance owner and re-review cadence as it evolves.
Human-in-the-loop on irreversible actions
Sanctions hits, account freezes, and large transfers always pause for a named human.
Where to start
Book an org assessment.
Ninety minutes, no slides.
We'll spend an hour with your leadership team mapping where AI creates the most leverage in your operation — and what an overlay would look like in practice.