This January, Droit will release its first product update featuring generative AI. The Decision Decoder integrates with the Droit Logic Viewer, pairing the Viewer’s visualization of a decision tree with an AI-generated decision summary. The summary decodes the reasons for the decision, explaining in plain language whether or not a rule condition was met and how the ultimate trade decision was reached.
In the vast majority of cases, the generated summary should be sufficient for understanding the basis for a trade decision. Compliance professionals tasked with a decision audit will now be able to move faster and with greater confidence, supported by explanations that clearly connect the requirements of a rule to the transaction under review.
Introducing the Decision Decoder
The Decision Decoder is available exclusively through Droit’s cloud-hosted Logic Viewer. A decision summary is automatically generated whenever an Adept decision is uploaded, covering why the product, parties, and method of transaction brought the trade into or out of scope for the specified regulatory requirement.
Below is an excerpt from a generated summary that describes why the product of the transaction – a EUR-denominated single-name CDS – is not reportable under Article 21 of EU MiFIR:
The product is not subject to the MiFIR post-trade publication requirement because, although it is a MiFID II credit derivative and is centrally cleared and denominated in a covered currency, it does not fall into any of the specific derivative categories listed in Article 8a(2) of MiFIR that trigger the publication duty under Article 21.
– Article 8a(2)(b) limits the single‑name CDS category to contracts that both are centrally cleared and reference a global systemically important bank. Although this trade is centrally cleared and in a covered currency (EUR), it does not reference a G‑SIB. Therefore, it does not fall within Article 8a(2)(b).
– Article 8a(2)(a) covers centrally cleared derivatives that are subject to the EMIR clearing obligation. This transaction is not subject to the EMIR clearing obligation, so it does not meet the (a) criterion either.
– Because the trade is not within any of the specific derivative categories listed in Article 8a(2), it is not one of the “OTC derivatives as referred to in Article 8a(2)” that Article 21 of MiFIR makes subject to the post‑trade publication requirement. Consequently, no MiFIR APA publication obligation arises for this transaction.
The summary provides all the necessary context for a comprehensive understanding of the Adept decision. The applicable articles of MiFIR are cited and explained. Relevant trade data is referenced. A detailed analysis unpacks why the eligibility conditions for a single-name CDS, as set out in paragraphs (a) and (b) of MiFIR Article 8a, were not fully satisfied.
This level of detail is possible only through leveraging the dense, rigorously maintained knowledge assets available on the Logic Viewer. For every trade decision, the Logic Viewer provides a traced decision tree, with each branch annotated against the relevant paragraph of the applicable rule. These and other informational elements, including the trade record itself and the eligibility decision made by the Adept decision engine, form the context-window used by the LLM when it generates its decision explanation.
This context both guides and constrains the LLM. The generated summary follows the facts of the traced decision while incorporating the nuance of the annotated regulatory text. Importantly, no external assets may be referenced by the LLM, ensuring that its generated summary remains grounded entirely in Droit’s expert-curated knowledge models. Because the summary is made available in the Logic Viewer, additional context is readily accessible to the user when needed. Relevant content includes the decision tree and links to rule citations, which are viewable in the Digital Library.
Suffice it to say, the trade decision itself continues to be determined using the rules-based, formalized logic of the Adept decision engine. That is, generative AI is performing explication, not evaluation. As we discussed in a blog post last year, the Adept decision engine will always rely on defined knowledge models that deterministically produce the same outputs for the same set of inputs. The role of Decision Decoder, by contrast, is to enhance interactivity, accessibility, and explainability of evaluated transactions.
Looking Forward
From its founding, Droit has pioneered an approach to regulatory decision-making that prioritizes traceability and explainability. The launch of the Decision Decoder builds on this commitment to transparency, both reducing the cognitive load of understanding complex regulatory requirements and enabling our users to more efficiently engage with our knowledge models.
This is just one of several planned initiatives to augment interactivity with the Adept decision engine through generative AI. We are excited to put the Decision Decoder into the hands of our users and hear their feedback, as we continue to leverage new tools to simplify the hard work of compliance and empower compliance professionals.
