ClearScore GroupAgentic Credit Broking Protocol

AI-mediated journeys need evidence, not assumptions.

ACBP is designed so that every journey produces a complete, structured, replayable case record. Regulated moments are explicit and logged. Trust is earned through demonstrated compliance — not assumed.

A structured case record

Every AI-mediated journey produces a structured, replayable case record. The broker maintains this record and is responsible for it.

User Agent identity

The identity of the AI platform or assistant that mediated the journey — so the source of the interaction is always known.

Structured data provided

The financial information the user shared with the AI, passed to the broker as structured data — not free-form text.

Broker actions and resolutions

Every compliance gate issued by the broker, and the explicit user response to each one. Consent, disclosure, declaration, instruction.

Conversation transcripts where available

Where the AI platform provides them, the conversation transcript is included — giving context for the structured events.

Replayable journey record

The full case record can be replayed against the regulated moments. Disputed or complained-about journeys can be inspected step by step.

Why replayability matters

The full case record can be replayed against the regulated moments. Disputed or complained-about journeys can be inspected step by step.

Not sure what a term means? See the glossary →

Inspection, not inference

In a traditional broker journey, the broker owns the surface and the audit trail is entirely first-party. In an AI-mediated journey, the broker receives structured events from the AI — and maintains its own case record from those events.

If a journey is later disputed or complained about, the ACBP record lets you inspect what happened at each regulated gate: what was presented, what was resolved, and in what order. You are not relying on a chat transcript or an AI's account of what happened. You are inspecting a structured record of the broker's own actions.

Inspection, not inference

Safe by default

Not all AI platforms are equal. ACBP is designed to be safe by default. Unknown or unproven platforms start with limited participation. Efficiency is earned through demonstrated compliance.

← Safe by defaultEfficient when earned →
Low / unknown

Introducer only

The AI gathers initial intent and data, then hands off entirely to the broker's own surface. The user completes the regulated journey on the broker's platform. Safe by default.

Medium

Data + redirect for gates

The AI handles data provision and offer presentation. At regulated moments — consent, disclosure — the user is redirected to the broker's own surface. Best of both worlds.

High

Full journey

The AI handles all operations including regulated gates. Applies to broker-controlled AI assistants and platforms with established compliance accountability.

ACBP includes a framework for compliance benchmarking — how AI platforms handle regulated moments is tested against standardised protocol scenarios. This is in development; details will be published on GitHub as the framework matures.

Portable by design

  • The protocol separates what must happen — structured communication, defined roles, enforced gates — from what the rules are, which varies by market.
  • The broker owns the regulatory layer. Whatever framework applies in their market, the broker remains responsible for adhering to it. The protocol governs the handshake. Not the content of the regulations themselves.
  • A conforming AI assistant is ACBP-compliant across markets — without needing to understand any individual regulatory framework.

Read the specification

View ACB Protocol

The full detail — operations, schemas, trust framework, and benchmarking — is on GitHub.

Sign up for updates

Stay informed as the protocol develops.