Production AI Institute · Public recordRecord
30-day evidence preparation path

A practical path from AI interest to inspectable evidence.

This route no longer describes a 30-day launch plan. It now sequences the evidence work a delivery team can do before making public readiness, disclosure, or deployment claims.

Start with discoveryOpen assessment guideEvidence comms kit
Days 1-5

Collect the baseline facts.

Start with factual context before recommendations. The first week should establish what is deployed, who owns it, what data is in scope, and where public claims need supporting evidence.

Inventory

List AI systems and assistants already in use

Record the products, owners, user groups, data classes, and current business purpose for each AI-enabled workflow or tool.

Tenant

Run Microsoft 365 discovery where relevant

Use the retained discovery scripts to collect licence, Copilot, security, sharing, Purview, and automation-readiness evidence.

Disclosure

Identify public-answer gaps

Mark where an organisation cannot yet explain AI use, data use, human review, or escalation in plain English.

Days 6-12

Map the work before suggesting changes.

Workflow

Capture current-state workflows

Document handoffs, systems, approvals, exceptions, manual checks, and output review before any AI-assisted future state is proposed.

Stakeholders

Interview the people closest to the risk

Speak with system owners, IT, operations, compliance, and representative users so evidence is not based on assumptions.

Records

Request policies and incident context

Ask for AI acceptable-use rules, data classifications, incident logs, vendor records, approval paths, and existing audit findings.

Days 13-20

Compare controls and prepare the review evidence.

Controls

Run a control diff

Compare current and proposed AI-assisted workflows against PSF controls, then record where human review, logging, access, or escalation must change.

Value

Document assumptions without overclaiming

Tie value claims to baseline effort, confidence, control dependencies, and owner sign-off rather than presenting a public price or revenue promise.

Agents

Prepare connected-agent evidence

For agents that can call tools or trigger workflow actions, capture owner, action limits, telemetry, rollback, and incident response evidence.

Days 21-30

Turn the work into a public-ready record.

Readout

Write the readiness summary

Summarise what was inspected, which facts are supported, which gaps remain, and what should not yet be claimed publicly.

Comms

Explain evidence in plain language

Use the communication kit to request missing evidence, explain the review, and follow up without hype.

Record

Link the evidence to the strongest public surface

Route the final summary to a disclosure, benchmark, product record, incident note, or future graph record rather than a commercial landing page.

Next step

By day 30, the useful output is evidence.

The goal is a public-ready summary of what is known, what is supported, what remains uncertain, and which controls need attention before production AI expands.

Open proof previewAgent readiness benchmark