Janis Tirtey
/
[ Vertical ] FINTECH BAFIN-REGULATED

Three hours per investigator, daily. Anthropic build on EU inference, BaFin-compliant.

Hawk and Sumsub deliver tools, not decision layers. DORA register, MaRisk AT-9, and EU AI Act Annex III maintained three times manually. Three fixed-price builds on top of your existing stack; MLRO stays in the loop, code in your Git, EU inference documented.

[ Use-Case ] FRAUD-COPILOT

Three hours back per investigator

8,000 EUR35,000 EUR3,500 EUR/month240k EUR/year

Each analyst can reasonably review 30 to 50 alerts per hour. Alert fatigue is draining fraud teams.

Bottomline, 2025
Before

30 alerts/h, context-pulling dominates

0 alerts/h

Investigator pulls context from 6 tools; SAR narrative manual; alert volume grows 30 to 40 Pct/year.

Investigator time60-80 Pct
Alert volume+30-40 Pct/y
SAR narrativemanual
Context pull12-30 min
After

50 alerts/h, MLRO decides, signs

0 alerts/h

Context pull in 12 seconds, SAR draft AI-generated, MLRO retains sign-off.

Context pull12 sec
SAR draftAI-generated
MLROsign-off only
240k EUR/yearfreed up
Fixed-price audit
8,000 EUR
5 days investigator shadowing
Pilot
35,000 EUR
6 weeks
Care-Pack
3,500 EUR/month
optional
ROI
240k EUR/year
investigator equivalent
[ Use-Case ] KYC-FALSE-POSITIVE

Halve KYC hits without letting risk through

6,500 EUR28,000 EUR2,800 EUR/month360k EUR/year

Stale KYC data means risk profiles do not evolve as customers' behaviour changes.

ShadowDragon, 2025
Before

6 FTE compliance plus 18 Pct drop-off

0 FTE

Manual hit handling ties up 6 FTE; 18 Pct onboarding drop-off through friction.

Compliance FTE6
Drop-off18 Pct
Hit volumemanually triaged
Auto-close0 Pct
After

3 FTE, 9 Pct drop-off

0 FTE

Halved hit handling; auto-close for clear cases; MaRisk AT 4.2 compliant decision logic.

Compliance FTE3
Drop-off9 Pct
Auto-closeapprox 50 Pct
MLROsign-off only
Fixed-price audit
6,500 EUR
4 days
Pilot
28,000 EUR
5 weeks
Care-Pack
2,800 EUR/month
optional
ROI
360k EUR/year
FTE equivalent + drop-off
[ Use-Case ] DORA-VENDOR-ENGINE

One vendor register, three regulators happy

12,000 EUR55,000 EUR4,500 EUR/monthreduced to 2 weeks

93.5 Pct of financial entities had at least one data quality error in their Register of Information submissions.

SBS Software, 2025
Before

BaFin special-audit prep 80k to 150k EUR plus 6 months mgmt

0 months

Three registers maintained manually in parallel; 93.5 Pct data quality error rate; every refile binds management.

Refile time6 months
Audit cost80-150k EUR
Data errors93.5 Pct
Mgmt efforthigh
After

2 weeks refile-ready

0 weeks

Single-source register feeds all three regulators; audit trail documented; mgmt effort minimized.

Refile time2 weeks
Audit costavoided
Data errorsvalidated
RegulatorsBaFin + EBA + AI Act
Fixed-price audit
12,000 EUR
8 days
Pilot
55,000 EUR
8 weeks
Care-Pack
4,500 EUR/month
optional
ROI
reduced to 2 weeks
refile cycle

Why not Capco, Senacor, or Hawk AI?

Capco / Senacor

Tier-1 1,800 to 3,500 EUR/day T&M, pyramid of junior + senior.

Fixed price without pyramid, solo senior builds it himself.

Hawk AI / Sumsub

Tools with annual license, no implementation.

Decision layer on top of existing stack, no replacement.

Bucket-A consulting (zeb, BCG X)

Big-4 day rate range, strategy decks instead of production code.

Implementation in your Git, MLRO retains sign-off.

Fixed prices per build

Fraud-Copilot
Audit8,000 EUR
Pilot35,000 EUR (6 weeks)
Care-Pack3,500 EUR/month
KYC false positive
Audit6,500 EUR
Pilot28,000 EUR (5 weeks)
Care-Pack2,800 EUR/month
DORA vendor engine
Audit12,000 EUR
Pilot55,000 EUR (8 weeks)
Care-Pack4,500 EUR/month
FAQ

Frequent questions on the AI workshop for BaFin-regulated FinTechs

Answers on BaFin compliance, MaRisk AT 4.2, KYC auto-close, Hawk and Sumsub integration, and MLRO liability.

Q1
Is Anthropic Claude BaFin-compliant?
Yes via AWS Bedrock EU plus AVV. Sub-processor list documented. AT 4.2 fulfilled by traceable decision logic plus MLRO sign-off.
Q2
What happens during a BaFin special audit?
Care-Pack includes audit-trail reports plus sub-processor updates. On request a 5-day sprint for special-audit preparation.
Q3
Auto-close of KYC false positives allowed?
Yes under MaRisk AT 4.2 plus §25h KWG with documented decision logic.
Q4
How does this integrate with our Hawk or Sumsub stack?
Decision layer on top of the existing API. No replacement, no migration.
Q5
Who bears liability for MLRO sign-off?
MLRO stays in the loop. AI generates the draft, signature stays human (legal requirement MaRisk AT 4.2).
Customer Story
Anonym, Approval auf Anfrage
BaFin-regulated payment services provider, approx. 70 employees, MLRO plus 8 investigators

240k EUR/year investigator equivalent

BaFin-regulated payments player with Hawk and Sumsub stack and 8 investigators in compliance. Alert volume plus 30 Pct/year, 60 to 80 Pct of investigator time is context-pulling. 6-week build: Anthropic Claude EU with RAG over KYC plus transaction plus sanctions, SAR narrative draft auto-generated, MLRO decides plus signs. Result: 30 to 50 alerts/h, 4-analyst equivalent gained, approx. 240k EUR/year.

Diagnosis

30 min diagnosis plus DORA quick-check

If alerts are piling up, the DORA register has data errors, or the next BaFin special audit looms, send a slot. Fixed-price audit after diagnosis.

30 min diagnosis call
(c) 2026 Janis Tirtey Consulting. Kraków, Poland.GDPR-compliant, EU data residency, DPA per Art. 28.