BAT Rapid
AI Human Review Reality Check. A fixed-scope review of one AI-supported workflow to test whether human review is meaningful, realistic, and evidenced.
Evidence for AI-assisted decisions
AI-assisted decisions need an evidence trail.
I assess whether AI-assisted decisions and workflows remain reviewable, reconstructable, and defensible under scrutiny.
Using BAT (Backward Accountability Traceability).
The evidence gap
AI can accelerate work without leaving enough behind to defend it.
In regulated and accountability-sensitive environments, the question is not only whether an AI tool is useful. It is whether the surrounding workflow can explain what happened, who reviewed it, what evidence was relied on, and why the final decision was reasonable.
What I do
Defensibility assessments for live AI-assisted workflows.
BAT Review
Focused AI Operational Defensibility Assessment. A deeper single-workflow assessment with targeted evidence review and reconstruction testing.
AI Defensibility Review
Full structured BAT assessment of consequential AI-supported operations, for one consequential workflow or a small workflow cluster.
The lens
Reviewable. Reconstructable. Defensible.
BAT stands for Backward Accountability Traceability.
It is the assessment method used by Accountable AI to examine whether AI-assisted workflows remain reviewable, reconstructable, and defensible after the point where AI output begins to influence real operational outcomes.
The work focuses on whether an AI-assisted workflow can be understood later by someone who was not in the room: an auditor, regulator, complaints handler, legal reviewer, risk lead, or senior accountable owner.
Reviewable
Can a qualified person inspect the workflow and see where judgement was applied?
Reconstructable
Can the team recreate the material context, evidence, and handoffs behind a decision?
Defensible
Can the organisation explain why the decision process was reasonable and controlled?
Assessment logic
From AI use to operational evidence.
Exposure Map
Identify where AI output enters the workflow, who owns the decision, and which operational outcome may be influenced.
Human Review Conditions
Test whether review is realistic in the live process, not just required in policy or implied by a sign-off field.
Defence Trail
Examine whether the organisation can show the source evidence, AI contribution, human intervention, and decision rationale.
Priority Actions
Translate evidence gaps into practical controls, ownership, and time-bound remediation.
Best fit
For teams using AI where accountability matters.
Consequential workflows include operations where AI-supported outputs influence decisions, records, actions, recommendations, escalation pathways, or customer outcomes.
Operational teams embedding AI into casework, triage, review, monitoring, or advisory processes.
Risk, compliance, legal, audit, and governance leaders who need practical evidence of control.
Product and transformation teams moving from AI pilots into governed operational use.
The work draws on experience across regulated operational environments, including clinical research and NHS settings, where accountability, review integrity, and defensible decision-making already carry real operational consequences.
Example Assessment Outputs
A clear view of what would stand up, and what would not.
- AI operational exposure map showing touchpoints, decision owners, review points, and influenced outcomes.
- Evidence gap summary covering logging, source traceability, human review, uncertainty capture, and reconstruction risk.
- Defensibility assessment with priority actions, accountable owners, and suggested timeframes.
Example evidence map
AI touchpoint
Where AI output begins to influence the operational record.
Review evidence
What shows that human review was meaningful, timely, and recorded.
Reconstruction gap
What would be difficult to explain later under audit, complaint, or regulatory scrutiny.
Start with one workflow
Find out whether your AI-assisted process can be defended.
Share the workflow, the decision type, and the scrutiny it may need to withstand. I will help identify the evidence trail that should exist around it.
Discuss a workflow