What This Is About
MiFID II requires every financial advisor to demonstrate, for every recommendation, on what basis it was reached. When AI tools are used in the advisory process — whether for extracting priority nodes from the initial conversation, risk assessment, or scenario simulation — a documentation obligation arises that goes beyond the classic suitability assessment.
Three questions that can come up in any BaFin audit:
- What share of the recommendation came from the AI tool, what share from the advisor
- What data and assumptions underpinned the AI output
- Were alternatives evaluated and rejected with documented reasoning
Advisors who cannot answer these three questions within minutes have a compliance gap.
What MiFID II Actually Requires
The suitability assessment under Art. 25(6) MiFID II requires the advisor to provide a written explanation of why a specific financial product is suitable for the client. When AI tools are involved, this implicitly expands to three requirements:
- Provenance of the recommendation basis: from which client data, plus which external data, plus which AI model was the recommendation generated
- Authorship clarity: which recommendation came from the advisor autonomously, which from the tool, which from the combination
- Alternative documentation: which options were evaluated by the tool or advisor and rejected with what reasoning
The ESMA Guidelines on Suitability Assessment 2018 raise the bar further: for systematic tool usage, tool outputs must be archived in a revision-proof manner.
The root cause of this documentation challenge is what we call Ghost Ownership — the attributability gap that arises when AI shapes a recommendation without leaving a clean audit trail of who decided what.
What Must Be Documented in Practice
Per client advisory session, at minimum these elements:
- Timestamp and version of the AI tool used
- Input data sent to the tool (anonymised or with consent)
- Output data including confidence values where available
- Advisor override wherever the advisor deviated from tool output, with reasoning
- Rejected alternatives with reasoning for why they were not recommended
- Client consent for AI use in the advisory process
In practice this is rarely done systematically. Excel notes are not sufficient. Word memos are not revision-proof. Standard CRM tools have no provenance logic.
How Steerable Solves This
Steerable is Decision Governance Infrastructure for CFPs. Per advisory session, a Decision Packet is automatically generated:
- Provenance layer with all sources, timestamps, and tool versions
- Authorship classification per recommendation element (advisor autonomous, tool output, advisor override with reasoning)
- Rejected alternatives automatically documented with advisor reasoning
- Audit-ready export for dispute cases, client-anonymised on request
The documentation is created during the advisory process, not after. The advisor provides the suitability reasoning, the system builds the audit-proof dossier.
For the complete architectural specification behind this approach, see the SR7D Framework whitepaper.
What Steerable Is Not
Not a compliance tool in the strict sense that checks whether a recommendation is MiFID II-compliant. Professional responsibility stays with the advisor. Steerable is infrastructure that ensures the documentation exists when it is needed.
Not a robo-advisor and not a recommendation engine. Steerable makes no advisory decisions. The human decides, the system makes the decision audit-proof.
For Ambitious CFPs
Steerable makes the CFP a live coach for strategic family decisions. A co-pilot that externalises mental models, cascades What-If impulses in simulation, and documents the decision trail in audit-proof form.
Those who want to advise not just audit-proof but also strategically will find more at steerable.org/whitepaper.
In this series
- Ghost Ownership — The attributability gap named. When AI shapes the decision, who actually decided?
- 7 Objections That Topple Every §34h Advisory — Structural documentation fix for DACH practitioners (PDF, DE).
- Reliable AI Doesn’t Save the Advisor — Why better AI makes the documentation gap worse, not better.
- This essay — AI Documentation under MiFID II. Compliance requirements in the German regulatory context.
Sources
References are verified against current ESMA and BaFin publications on updates.
| Source | Date | Schema |
|---|---|---|
| MiFID II Art. 25(6) | EU Law 2014/65/EU | Regulation |
| ESMA Guidelines on Suitability Assessment | May 2018 | ESMA35-43-1163 |
BaFin Interpretation FAQ MiFID II | updated 2024 | BaFin Publication |
| Steerable Whitepaper Decision Packet v0 | 2026-03-03 | Decision Packet Schema |