Form cover
Page 1 of 1

Help Test BAT With a Real Workflow

BAT is a decision-time accountability methodology for AI-supported work. It looks at whether reliance on an AI output can be defended: what the AI output is used for, who relies on it, what review or control happens, what evidence is kept, and what could be hard to explain later.
This form collects real workflow examples to test and improve BAT. You do not need to share confidential details; anonymised or generalised descriptions are welcome.

Name

You can leave this blank. Workflow examples can be reviewed anonymously.

Email

Optional - leave your email if you would like to receive a BAT challenge summary or if you are happy for me to ask follow-up questions.

Organisation

Sector

Sector

What is your role in relation to this workflow?

Select the option that best describes your perspective. If more than one applies, choose the one closest to your involvement in the AI-supported workflow.
What is your role in relation to this workflow?

How long have you been using AI in this workflow?

How long have you been using AI in this workflow?

Workflow Name

Optional - a short label for the workflow, such as “complaint triage,” “draft customer response,” or “case prioritisation.”

Workflow description

Example workflow description

Staff use an AI tool to summarise incoming customer complaint emails and suggest a draft response category.

A case handler reviews the summary and category before deciding whether the complaint should be resolved at first contact, escalated to a specialist team, or logged for formal investigation.

The AI output is not sent directly to the customer, but it influences how the case is routed. The complaint email and final routing decision are retained, but the AI-generated summary and any changes made by the handler are not always saved.

Who uses the AI-supported output, and what do they use it for?

Name the role or team, not the person. For example: “case handler uses the AI summary to decide routing,” or “manager uses the generated report to prioritise follow-up.

What happens next after the AI output is produced?

Describe the next decision, action, communication, routing step, prioritisation, or review. For example: “a case handler reviews the summary and decides whether to escalate the case.”

Who or what could be affected by reliance on the AI output?

Include people, groups, cases, services, queues, communications, decisions, resources, or operational priorities that could be affected.

Optional examples:

A customer, patient, employee, applicant, complaint case, waiting list, service queue, investigation, public communication, or internal resource allocation.

What evidence, data, or source material does the AI use?

Describe the information the AI output is based on. For example: customer emails, case notes, policy documents, call transcripts, medical records, application forms, previous decisions, database records, or uploaded files.

Is that source material retained and available later?

Is that source material retained and available later?

What review or control happens before anyone relies on the AI output?

This might be human review, sampling, exception review, automated checks, approval thresholds, peer review, audit, or no review. Describe what actually happens in practice.

Who is accountable for the final reliance decision?

Name the role, team, or function responsible for deciding whether the AI output is used or acted on. If ownership is unclear or shared, say so.

Is that ownership clear in practice?

Is that ownership clear in practice?

Who has authority to approve the action or decision that follows?

Name the role, team, or function with approval authority. This may be the same as the person using the AI output, or it may be a manager, specialist, panel, system owner, or delegated approver.

Is the authorised approver the same person or team that relies on the AI output?

Is the authorised approver the same person or team that relies on the AI output?

What records are kept about the AI output and how it was used?

Include any records of the AI output, source material, review, edits, rationale, approvals, escalations, final decision, or communication sent.

What records are kept about the AI output and how it was used?

Has anyone already relied on the AI output?

“Relied on” means the output has been used to support a decision, action, communication, prioritisation, routing step, recommendation, or other operational outcome.

Has anyone already relied on the AI output?

Are there any known concerns about this workflow

Select any concerns that apply, even if they are only occasional or uncertain.

Are there any known concerns about this workflow

Who has authority to approve the action or decision that follows?

Name the role, team, or function with approval authority. This may be the same person/team that uses the AI output, or it may be a manager, specialist, panel, system owner, or delegated approver. If unclear, say so.

May I use this workflow as a future BAT case study?

Case studies may be anonymised and used to test, explain, or improve the BAT methodology. No names of individuals or organisations will be used without explicit permission.

May I use this workflow as a future BAT case study?
This form is run by Accountable AI as part of the BAT decision-time accountability research programme.
If you have questions about the form or how responses may be used, contact kim@accountableai.uk