← Blog

Operations · · 4 min read · Updated

Your AI program has a review queue, even if you call it approval

Measure waiting time, reviewer effort, and rework in the queue between an AI-assisted draft and a published marketing asset.

Jonathan Haas

Calling a step “approval” does not make the queue disappear.

When AI-assisted content enters review, it waits for a person with the right context, time, and authority. The item may be technically complete and still unavailable to the campaign. It may also wait because nobody knows which reviewer owns the decision.

A review queue has a cost even when no one has touched the draft.

That cost is waiting time, reviewer capacity, coordination, and the chance that a campaign window closes before the asset is usable.

Name the states between draft and publication#

Use plain states that describe what can happen next:

State Meaning Next owner
Drafted The requested format exists Content owner
Awaiting review A named approver must decide Approver
Returned A specific change is required Content owner
Evidence needed A claim or source needs support Brief owner or subject expert
Approved The content passed the stated checks Channel owner
Published The asset reached its intended audience Campaign owner
Expired The decision window closed Brief owner

“In progress” hides too much. It can mean a person is writing, a reviewer is late, a source is missing, or the campaign no longer matters. Each case needs a different action.

Calculate the queue before it becomes a staffing argument#

Suppose 72 drafts enter a monthly review process. Each draft receives three review passes, and each pass takes an average of 11 minutes:

72 drafts × 3 passes × 11 minutes = 2,376 reviewer minutes

That is 39.6 hours before counting comments, meetings, and rework by the content owner. A provider invoice can be only $180 while the review queue consumes a full work week.

The first useful measures are simple:

Measure Formula
Time to first review First review timestamp − submitted timestamp
Review minutes Minutes recorded by each reviewer
Return rate Returned items ÷ reviewed items
Repeat pass rate Items with two or more review passes ÷ reviewed items
Queue age Current time − submitted timestamp
Expired rate Expired items ÷ submitted items

Do not use approval time as a single quality score. Fast approval can mean a clear process, or it can mean people are clicking through a queue they no longer trust.

Give the approver a decision, not a document dump#

An approval request should answer the questions a reviewer actually has:

  • What is this asset for?
  • Which audience and channel will receive it?
  • What changed since the last attempt?
  • Which claims need attention?
  • What source set supports those claims?
  • What happens if the reviewer returns it?
  • How long is the decision useful?

Attach the evidence and the change summary. Do not ask a reviewer to discover the risk by comparing four files with similar names.

Approval design covers the deeper interaction pattern: show the decision, its scope, the evidence, and the consequence. The same principle applies to marketing content. A reviewer should be able to accept, return, or request evidence without starting a separate coordination thread.

Set different service levels for different risk#

Every item does not need the same wait time. Define the path before the queue grows:

Asset type Review path Target response
Internal draft Content owner Same business day
Product claim Product marketing One business day
Customer-facing promise Product and legal Two business days
Regulated or regional claim Named specialist Hold until evidence is attached
Expired campaign Brief owner Close or rewrite the brief

The target is not a promise that people must meet at any cost. It is a signal that tells the process when to escalate, change the owner, or stop spending on the item.

Make a stale queue expensive to ignore#

For each item older than its service level, show:

  1. the campaign and owner;
  2. the missing decision;
  3. the time already spent;
  4. the amount of direct spend attached to the item;
  5. the next valid choices: review, return, expire, or rebrief.

This turns a vague backlog into work someone can resolve. It also prevents the team from producing more drafts while accepted work waits for the same reviewer.

The review-cost guide shows why human time belongs in the content denominator. The waste guide shows why a draft that never leaves the queue is not a finished asset.

AI can reduce the time needed to create a first draft. It cannot decide who has authority, which claims are acceptable, or whether a campaign still matters. The queue is where those decisions become visible. Measure it like a real production process, and the marketing budget gets a place to improve.