What is Approval workflow?

An approval workflow is the process a proposed change passes through before it takes effect.

The shape is consistent wherever it appears. Somebody or something proposes a change. The change is held rather than applied. A check happens. The change is applied, rejected or amended, and a record is written.

What varies is what gets gated, who or what performs the check, and whether anything is recorded afterwards. The last of those decides whether the workflow is useful or ceremonial.

The check can be exercised in two places, and both are legitimate.

At the moment. A person reviews the specific change and decides. This suits changes that are unusual, large, or that depend on context the process cannot encode — a campaign for a sensitive product, an unusual budget, a claim that needs legal sign-off.

In advance. A rule written earlier decides. A budget increase under a threshold applies automatically; one above it waits. A change to a bid target within a range goes through; outside it does not. A creative from an approved library runs; a new one waits.

The difference between these is when the judgement was exercised, not whether it was. A rule somebody wrote deliberately, reviewed periodically, and can point at is an exercise of authority as real as a click on an approve button — and frequently a better one, because it was made without time pressure and applies consistently.

What matters is that the rule was chosen rather than defaulted into, and that the record afterwards says which authority cleared the change. A history that shows a rule applied is auditable; one that shows only that something happened is not.

The test is cost and reversibility.

High cost, slow to reverse. These are the cases the workflow exists for: budget increases past a threshold, new campaigns, changes to bid strategy or target, structural changes, anything touching a regulated claim.

High cost, fast to reverse. A lighter check, or a rule with a tight threshold. A budget increase that can be undone in a minute needs less ceremony than one that commits a quarter.

Low cost, fast to reverse. Should not be gated at all. Pausing a fatigued ad, adding a negative keyword, adjusting a small budget, rotating creative. Gating these builds a queue that slows everything and teaches people to route around the process.

Anything with an external obligation — a regulated category, a contractual commitment, a partner's brand — regardless of its cost, because the consequence is not financial.

An approval that leaves no trace is a conversation that happened.

Three months later, when a target is producing odd results, the useful questions are what it was before, who changed it, when and why. An account without a record cannot answer any of them, and the investigation becomes archaeology.

A usable record holds four things: what was proposed, what it replaced, who or what cleared it, and when. That is enough to reconstruct a decision and to distinguish a deliberate change from a mistake, which is most of what an investigation needs.

It also changes behaviour before the fact. A change that will be recorded with a name against it is considered slightly more carefully, which is a modest effect and a real one.

Gating everything. The queue grows, the routine work waits behind the important work, and people find a path around it. A process that is bypassed is worse than none, because it creates a false belief that changes are being reviewed.

Approving without reading. A reviewer facing forty items will clear them. The remedy is fewer items rather than a stricter instruction.

No record. Covered above and the most common failure, because the approval step gets built and the logging does not.

Rules nobody revisits. A threshold set for an account at a tenth of its current scale now gates every ordinary change, or waves through amounts that used to be significant.

Unclear authority. Where nobody knows who can approve what, changes either stall or happen unreviewed, and which one occurs depends on who is available.

Write down what is routine, explicitly, as a list rather than a principle. Everything on it applies without review.

Set thresholds in proportion to the account, and revisit them when the scale changes.

Name the authority for each category of change, so nobody has to ask.

Log everything, including the changes that were auto-approved by rule, because those are the ones nobody remembers later.

Review the log periodically rather than only when something has gone wrong. Patterns in what is being changed, and by whom, are informative before they are urgent.

Keep the routine path fast. The credibility of the check on the important changes depends on the unimportant ones not being stuck behind it.

A workflow has to cope with proposals arriving from several directions, and the directions have different risks.

People working in the account. The traditional case, and the one most processes were designed around.

Automated rules the account set up. A rule that pauses an ad on a threshold, raises a budget on a signal, or rotates creative on a schedule is proposing changes continuously. These are usually not gated at all, which is defensible only if the rule itself was reviewed when it was written.

The platform's own optimisation. Bid strategies, budget distribution and placement allocation change what happens every hour and are not proposals in any reviewable sense. What can be governed is the instruction they follow and the limits around it.

Agents and integrations acting on the account's behalf. These sit closest to the first category in what they can change and closest to the second in how fast they change it, which is why the record of what cleared each change matters more here than anywhere else.

The practical implication is that a workflow covering only the first category governs a shrinking share of what actually happens to an account. The rules, the limits and the instructions given to automated systems are where most of the governance now has to live.

A useful workflow separates two questions that are easy to conflate.

Is this change permitted? A matter of policy, answerable in advance, and the right thing for a rule to decide.

Is this change wise? A matter of judgement, dependent on context, and the right thing for a person to consider.

Most accounts gate on the first and imagine they are gating on the second. Making the distinction explicit — writing down which changes are policy questions and which need judgement — is what stops a review process becoming a rubber stamp on a queue.