Compliance Data Services
Menu

Regulatory data services for established compliance teams at banks and regulated firms worldwide

Your regulatory taxonomy was built for people.
R2 makes it work for AI.

R2 evaluates the regulatory data your AI relies on, and checks its decisions against those your experts have approved.

Start with an AI readiness evaluation and get your score

Your taxonomy needs to be updated for AI

Taxonomy

What your GRC platform records

  • Topic
  • Citation
  • Description
  • Business lines
  • Products
  • Risk rating
  • Owner
  • Linked policies and controls
  • and more

Not consistently captured

  • Who must act
  • What they must do
  • Conditions
  • Scope note
  • Verified links to each obligation

Regulatory requirements

At the level AI needs

Granularity

  • One obligation per record
  • Cited to the paragraph
  • Quoted provision text

Metadata

  • Jurisdiction
  • Regulator
  • Source
  • Citation
  • Link to the source text

You don’t need another platform. The one you have needs what’s missing, and R2 adds it. An AI readiness evaluation measures whether your taxonomy gives AI the context it needs.

01 — What you get

R2 delivers the context and the evaluations that keep AI predictable as regulations change.

The evaluation scores whether your taxonomy gives AI enough context to decide applicability accurately and reliably, and recommends a modification for each issue it finds.

After the evaluation, R2 offers:

  • Regulatory sources. New, amended and repealed obligations from the sources you select, at the level AI needs, each with a decision on whether it applies to your firm and a justification for your experts to approve.
  • Taxonomy updates. As obligations are added, amended or repealed, and as your business adds products or changes existing ones, R2 updates the affected requirements on your behalf or alongside your team, so AI keeps the context it needs.
  • Ongoing evaluations. A set of applicability decisions your experts have approved, run against AI each time the taxonomy, the regulations, the model or its prompts change, so you have evidence, each time, of whether AI still decides as your experts would. When regulations or your business change, your experts review and reapprove the affected decisions before the set runs again.

How the evaluation works

Your team adds judgment.
R2 adds scale.

Your experts keep responsibility for interpretation and approval.

02 — Case study, part 1: The problem with your data

A taxonomy built for people misleads AI.

Your taxonomy was designed for expert review: your team brings the judgment and context that make it work. AI has only what’s written down, and broad labels don’t reliably tell it the scope you intend. These are the errors R2 tests for.

What wrong links look like

A consumer credit rule was filed under Governance.

Obligation

An MCD article 3(1)(b) creditor must comply with CONC 6.7 (post contract: business practices).

UKFCA HandbookCONC 1.2.8(2)

Linked by Topic, Jurisdiction and Description

Existing taxonomy

AI justification

The obligation requires compliance with CONC 6.7, which is a specific rule within the FCA Handbook. This topic covers ongoing regulatory obligations under the FCA Handbook for regulated firms.

Without R2’s extended taxonomy

What went wrong

All three fields fit loosely: a broad Topic (Governance), the right Jurisdiction (UK) and a Description that covers any FCA rule. None of them ruled out a consumer credit rule on post-contract practices.

What the data didn’t say

What Governance excludes. There’s no scope note, so the Description matches any FCA rule.

In each case, the taxonomy led the AI to a wrong answer.

R2 extended the client’s taxonomy, making the scope and conditions of its requirement topics explicit.

How would your taxonomy hold up? Get your AI readiness score

03 — Case study, part 2: Using AI to build a Regulatory Library

Why AI struggles to read regulatory structure

Your team can point a model at the law and have a library by Friday. It will look complete, and it won’t be. Each source document has its own quirks: hidden text, rules laid out as tables, labels that change meaning from one rule to the next. They trip up people as well as AI. Getting them right takes a process, not just code.

Extracted structure

  • 2021, c. 25
  • s. 1101Corrupted citation

In R2’s library

  • 2021, c. 25
  • s. 110
Source
Quebec › Privacy act P-39.1 › p. 9
Structural issue
The page carries a second, invisible copy of the citation text.
What breaks
The citation is corrupted: s. 110 becomes s. 1101.

When AI misreads the structure, everything downstream breaks.

R2’s library records the structure correctly, so each provision and obligation keeps its place and its citation.

04 — Case study, part 3: Screening at scale

Without proper context, AI scales the wrong answers.

R2 evaluated the complete corpus against the client’s criteria, using its existing taxonomy together with the extended one, and decided which publications and obligations applied. Every decision is defensible: it has a justification, and every exclusion has a recorded reason the client’s team can check.

Without the extended taxonomy and properly extracted obligations as context, both foundation and open-source models generated errors.

Canadian regulatory publications fetched and evaluated
33,000+
To build a new regulatory library, the client had to find which publications mattered to its business. R2 evaluated each one against the client’s criteria: jurisdiction, product, category, and whether it is guidance or a requirement.
Jurisdictions: federal, plus all 13 provinces and territories
14
Met the client’s criteria and went to expert review
Under 100
R2 then evaluated every obligation in them. Every result has a justification.

Start with an AI readiness evaluation

Does your regulatory taxonomy give AI enough to decide?

A taxonomy isn’t measured by how much it holds. It’s measured by whether AI decides as your experts would. Get your AI readiness score. Before you rely on AI for applicability, R2 tests whether your existing taxonomy gives it the context to decide, and scores it.

How the evaluation works

Your evaluation includes

  • Structure test
  • Applicability test
  • AI Readiness Score

    AI Readiness Score from R2 Compliance Data Services: 38.5

    Scores can be reported for the whole taxonomy or by source, jurisdiction or business line, using the same tests.
  • Suggested modifications

Request an evaluation

Tell us what you would like assessed. We’ll reply to discuss scope, timing and next steps. Not sure yet? See how the evaluation works first.

Fixed price covers a subset of the FCA Handbook or the US Code. Variable price covers all other regulatory sources.

If you have one from an event or an offer.

We’ll only use these details to reply to you. Privacy Policy (opens in a new tab)