Can you stop it in time?

Intervene’s AGDA® assessment examines whether your organisation can act before harm from an AI-enabled system becomes irreversible.

Discuss your system Explore AGDA®

One named system. One defined pathway. One intervention scenario.

Governance is an input. Intervention capability is the outcome.

The problem

Responsibilities are documented. Can the people act?

Who can actually stop it?
Not who owns the risk. Who has the authority to intervene?
Will they know in time?
Does actionable information reach the right people before the intervention window closes?
What happens when they act?
Does intervention work under actual operating conditions?

The method

Assess the ability to intervene.

AGDA® assesses whether organisations can detect, escalate, decide and intervene when consequential systems require human intervention.

The intervention window.

Compare the timing of intervention In time Too late

Conceptual illustration. No assessment result, measured durations or scale.

Window closes

Detect
Escalate
Decide
Intervene

In time

Detection, escalation and a decision with authority lead to effective intervention before the window closes.

Window closes

Detect
Escalate
Decide
Intervene

Too late

The chain reaches intervention after the window closes. Detection alone does not establish the ability to act in time.

Select a stage.

01 Detect Can the organisation recognise when intervention is required?

Evidence examined
Signals, monitoring records and evidence of how the need to act is recognised.

Possible failure
A warning exists but does not become an actionable signal.

Consequence for intervention
The available intervention window is spent before the response begins.
02 Escalate Can the right people be informed in time?

Evidence examined
Escalation routes, contact arrangements and records of reaching decision authority.

Possible failure
The signal reaches someone who cannot authorise action.

Consequence for intervention
Reaching the person with authority may consume the remaining time.
03 Decide Can accountable decision-makers determine what action is necessary?

Evidence examined
Decision records, halt authority and arrangements for decisions under pressure.

Possible failure
A decision depends on a quorum or approval route that cannot operate in time.

Consequence for intervention
Detection and escalation cannot compensate for unavailable decision authority.
04 Intervene Can the organisation execute an effective intervention?

Evidence examined
Intervention procedures and records of exercises or action taking effect.

Possible failure
An authorised action does not take effect under the relevant operating conditions.

Consequence for intervention
The response must take effect before the consequence becomes irreversible.

The weakest stage constrains the chain. Public explanations only, not assessment criteria or findings.

The evidence

A verdict. Supported by evidence.

A full AGDA® Assessment returns a verdict, evidence register, SEDI findings and executive interpretation. The Review does not produce a verdict.

Constructed illustration

Detection works. Authority does not.

Can the auto-decline pathway stop within an assumed six-hour window? No client data; not an issued assessment or a verified outcome.

What prevents intervention in this example.

Assumed window about six hours, held as a scenario assumption. Each stage below is compared with that window on its own.

01 Detect ~12 min Within the window Drift identified above threshold.
02 Escalate 4 to 96 hours May consume or exceed the window To reach a person with authority to halt.
03 Decide ≥ 7 days Exceeds the window on its own To obtain the quorum that person needs. No out-of-cycle route.
04 Intervene Not assessed Preceding delay already exceeds the window The source assumes about six minutes to propagate a halt if instructed; no duration is assessed here, and decisions already made are reprocessed next day.

Decisive constraint in this example Authority, not detection. The route to a person with authority may itself exceed the window, and the quorum that person needs exceeds it on its own. Faster detection cannot change that. Naming this constraint does not show that the other stages are adequate.

Constructed illustration: no client result, issued assessment or verified outcome. Intervals are separate illustrative facts, not to scale and not added together. Their endpoints and any overlap are not established; no total response time is computed.

Inspect the scope, decision and limitations

System: CRED-MOD-04, a retail credit decisioning model. Automated approve, decline or refer decisions; the question concerns the auto-decline pathway.
Scenario: A macro shock coincides with model drift and creates a need to stop or redirect the affected pathway.

Decision the finding supports: Whether the auto-decline pathway can keep operating under its present authority arrangements, and what has to change before the organisation relies on it: who may authorise a halt, how that person is reached out of hours, and how the route is exercised before it is needed.
Assumptions and limits: One pathway and one scenario. Nothing here is a conclusion about any institution, other systems or legal conformity.

Issuance requires approval separate from the assessor's own judgement. Signed SEDI result records are not a standard deliverable. They may be supplied only under an approved output-access arrangement, where the engagement terms provide for them and the delivery configuration supports them. What verification establishes.

Cryptographic verification establishes the integrity and provenance of an assessment artefact within the scope of the verification process. It does not independently establish the truth of every underlying finding.

Where should we begin?

Two starting points.

Neither is a prerequisite for the other.

Choose by the intervention question and available evidence.

Route A · Preliminary
Intervention Readiness Review
Appropriate when

You need to establish the material intervention question for one system, or clarify its scope and evidence position.

Question and decision
Whether the selected system carries a material intervention question, and what you would need to assess, and to have access to, before relying on the answer.

Evidence examined
Review of the evidence you already hold.
What you receive
- Executive review: what to assess and why
- Evidence and access position
- Scoped recommendation
Boundary

Preliminary chain mapping only. The Review does not include the AGDA® verdict, the SEDI findings or signed SEDI result records.

Indicative fee

£20,000 to £35,000 + VAT
Explore the Review
Route B · Full assessment
AGDA® Assessment
Appropriate when

The system, pathway and scenario are defined, and relevant evidence access is available.

Question and decision
Whether the agreed pathway can continue operating under its present intervention arrangements, and what has to change before the organisation relies on it.

Evidence examined
Structured interviews, walkthroughs and review of relevant evidence.
What you receive
- Evidence register with provenance and gaps
- SEDI findings and chain analysis
- Verdict, confidence basis, limitations
- Executive interpretation and remediation priorities
Boundary

One system, one pathway, one scenario. Connected systems are scoped separately.

Indicative initial range

£75,000 to £95,000 + VAT
Explore the Assessment

A Review may recommend a full Assessment but does not commit you to one or determine its verdict.

Initial pricing remains subject to market validation. The proposal confirms scope, outputs, fee, terms and timetable. Contact us to discuss scope and availability. We confirm delivery arrangements and a start date before accepting an engagement. Scope, fees and how an engagement runs.