How Long Does It Take a Client to Say Yes?

One provider wants to use AI on one piece of important client work.

The request enters legal. Security asks for the tool list. Privacy asks where the data goes. Procurement sends a questionnaire. The client team schedules a meeting to explain what the questionnaire could not capture. New questions appear. Old answers get copied into a different format.

Three weeks later, nobody can say how many hours the decision consumed.

The result may be yes, no, or yes with conditions. The process leaves almost no reusable evidence about how the decision happened.

We should measure it.

The study asks one bounded question

How much time and repeated diligence do a buyer and provider spend reaching one bounded AI-use decision, and what changes when the decision is supported by structured evidence and a reusable assurance assertion?

This is the AI Permission Cycle Time research method.

It is a proposed measurement system, not a result.

This is a proposed research method and participant call. No results have been collected, and participation is not certification, endorsement, or evidence of performance.

One decision is the unit

The unit of analysis is deliberately narrow:

→ one provider

→ one buyer

→ one bounded AI use

→ one documented permission event

An enterprise AI policy is too broad. A model review is too narrow. A provider-level average hides the object that buyer and provider teams actually negotiate.

The event begins when the buyer receives a complete request containing the minimum required information. It ends when the accountable buyer records one result:

• approved;
• approved with conditions;
• rejected;
• withdrawn; or
• unresolved at the study cutoff.

The method records the time before a request becomes complete as a separate preparation measure. That prevents a provider from looking fast merely by starting the clock after weeks of undocumented work.

Seven measures replace one vague complaint

Teams often say AI review “takes forever.” That is not a metric.

The study records:

1. Permission cycle time. Calendar time from complete request receipt to documented decision.
2. Buyer diligence hours. Human time spent reviewing, clarifying, meeting, escalating, and documenting the decision.
3. Provider diligence hours. Human time spent preparing evidence, answering questions, and revising the request.
4. Clarification burden. Material follow-up requests before the decision.
5. Evidence reuse rate. Accepted evidence items reused from an earlier permission event divided by accepted evidence items presented.
6. Decision result. Approved, conditional, rejected, withdrawn, or unresolved.
7. Control preservation. Whether the decision retained client-specific restrictions, escalation rights, and accountable ownership.

Fast is not automatically good.

A five-minute approval that skips the buyer's restrictions is a control failure, not a productivity win. More time may reflect necessary scrutiny. Less time may reflect better evidence, lower risk, weaker review, or a narrower use.

The method records decision quality beside speed.

Baseline and structured-evidence cases

Where a participant has comparable earlier decisions, the study records up to three baseline events. It then records the same fields for the study event supported by a structured evidence pack.

The evidence pack may include:

• the legal entity and service boundary;
• the exact AI use and purpose;
• tools, suppliers, and data classes;
• client restrictions;
• accountable owners;
• current control evidence;
• task-specific evaluation;
• known exceptions;
• material-change rules; and
• a concise buyer decision record.

The study does not assume the pack will make the decision faster. It may save time, move work earlier, add necessary control work, expose a reason to reject the use, or make no measurable difference.

Every one of those is a useful result.

What enters the clock

Included work:

• request preparation after the study start;
• evidence collection and redaction;
• questionnaires and written follow-up;
• buyer and provider review time;
• meetings tied to the permission event;
• legal, privacy, security, procurement, and accountable-owner review;
• remediation required for the decision; and
• final decision documentation.

Excluded work:

• product development unrelated to the permission request;
• ordinary sales activity;
• broad AI programs not tied to the named use;
• unrelated contract negotiations;
• research administration; and
• time recorded without a defensible event link.

Every exclusion receives a reason. Convenient omissions do not disappear from the limitations.

Buyer authority stays intact

The research cannot become a sales tool wearing a lab coat.

The buyer keeps:

• the right to request more evidence;
• the right to apply client-specific restrictions;
• the right to reject the use;
• the right to pause after a material change;
• the right to remove its data before publication, subject to the consent agreement; and
• the final decision authority.

The provider cannot condition participation on a positive result. Petrichor cannot condition fees, publication, or future work on approval.

Participation requires a real decision

Buyer participants must own or materially influence a real provider AI-use decision and agree to record the decision path.

Provider participants must bring a real, assessable service boundary, disclose known material conditions, and accept that the study may reveal gaps or produce an unfavorable result.

The first pilot target is:

• eight to twelve providers;
• at least three buyer risk owners;
• at least ten independently double-scored cases; and
• at least three live buyer permission decisions.

That gate can support a pilot report. It cannot support a broad market benchmark.

The report will include unfavorable results

A pilot report will state:

• who sponsored and funded the work;
• how participants were recruited;
• the sample and its limits;
• which events were live and which were simulated;
• medians and ranges;
• missing data and attrition;
• buyer and provider labor hours;
• clarification and evidence-reuse measures;
• decision results and restrictions;
• null, mixed, adverse, and favorable findings;
• method changes; and
• limits on causal inference and broader use.

No client or provider will be named without written permission. Small cells will be suppressed or combined when a person or organization could be identified. Privileged material, client content, security details, personal data, and unrelated evidence stay outside the public record.

Control failures stop the study

The pilot pauses or stops when:

• participants treat the evidence pack as guaranteed approval;
• the buyer's decision authority is weakened;
• the boundary cannot be stated truthfully;
• the provider demands a favorable result;
• protected information cannot be handled lawfully;
• the measures change after results are known;
• live and simulated decisions cannot be separated; or
• publication would create a serious reidentification risk.

Method integrity outranks a flattering chart.

Contribute one real decision

Petrichor is seeking buyer risk owners, professional-service providers, assessors, standards specialists, insurers, and research-method reviewers.

Choose the relevant research intake path or email rimmler@petrichorgrowth.com.

Bring one bounded decision. Not a testimonial. Not a success story.

The evidence gets to decide what this becomes.

Frequently asked

Have you measured AI Permission Cycle Time yet?

No. This page describes the proposed method and participant criteria. It contains no outcome claim.

Is participation a certification or endorsement?

No. Participation is research activity only. It does not grant a badge, preference, approval, or future eligibility.

Will a faster decision count as success?

Only when buyer controls and decision quality remain intact. The study records speed, labor, restrictions, evidence, result, and control preservation together.

Can simulated decisions enter the pilot?

Yes, if they are labeled and analyzed separately. At least three live buyer permission decisions remain part of the minimum pilot gate.

Will unfavorable findings be published?

Yes, within the consent, privacy, privilege, and reidentification rules. Null, mixed, and adverse findings are part of the research commitment.