TL;DR: Approval thresholds are the conditions under which a buy-side agent pauses and seeks human authorisation before proceeding. Setting them correctly is a balance between two failure modes: thresholds too low produce agents that escalate constantly and deliver no efficiency gain; thresholds too high produce agents that make commitments the buyer would have wanted to review. Good threshold design starts from the buyer's actual decision boundaries, not from a guess at what might be risky.

An approval threshold is not a safety net. It is a governance instrument. Its purpose is to route a specific class of decision to a human who has the authority and context to make it. A threshold set at the wrong level does not make the agent safer; it either creates unnecessary friction for decisions the agent should be empowered to make, or it fails to catch decisions the buyer needed to review.

Designing thresholds well requires the buyer to think carefully about which decisions genuinely require human input and which are within the scope of what they have already authorised.

The two classes of threshold

There are two distinct classes of approval threshold, and they should be designed separately.

The first class is financial thresholds: conditions based on the monetary value of a proposed commitment. A financial threshold might be: "seek approval if a single deal commitment exceeds £10,000," or "seek approval if cumulative committed spend in this session exceeds 80% of the daily budget cap." Financial thresholds protect against the agent making commitments that exceed the buyer's intended exposure without review.

The second class is scope thresholds: conditions based on whether the proposed deal is within the buyer's defined operating scope. A scope threshold might be: "seek approval if the proposed publisher is not on the authorised list," or "seek approval if the proposed audience segment combination includes a demographic category not specified in the mandate." Scope thresholds protect against the agent operating in territory the buyer has not explicitly authorised, regardless of the financial value.

Both classes need to be present in a well-designed mandate. Financial thresholds without scope thresholds leave the agent free to make low-value commitments in any territory. Scope thresholds without financial thresholds leave the agent free to make unlimited commitments within the defined scope.

How to set financial thresholds

Financial thresholds should be set from the buyer's actual decision-making patterns, not from a round-number guess.

The relevant question is: at what deal value would this buyer normally want to review a proposal before committing? For a buyer running a £50,000 campaign, a single deal of £500 is probably within the agent's operating scope; a single deal of £8,000 probably warrants review. The threshold should reflect that boundary.

The threshold should also account for the campaign's pace. An agent running a campaign with a £5,000 daily budget cap should have a lower per-deal threshold than an agent running a campaign with a £50,000 daily budget cap. The absolute value of the threshold matters less than its relationship to the campaign's total exposure.

How to set scope thresholds

Scope thresholds should be derived from the inventory and audience parameters already defined in the mandate. If the mandate specifies an authorised publisher list, the scope threshold for publisher scope is straightforward: any publisher not on the list requires approval.

The more nuanced scope thresholds involve combinations of parameters that are each within the defined scope individually but constitute a new situation in combination. For example: a publisher on the authorised list proposing a format that is not on the authorised list, or an inventory package that combines an authorised contextual category with a new placement type. Combination scope thresholds are harder to specify in advance, and some agents handle them through an escalation rule that fires on any parameter combination not explicitly covered in the mandate.

What happens at an approval threshold

The escalation experience at an approval threshold should be designed as carefully as the threshold itself. When the agent escalates, the human approver needs to receive, in a single notification: what deal the agent was considering, what threshold it triggered, what the agent's recommendation is (proceed or decline), and what the buyer needs to decide. An escalation that requires the approver to log into a platform to retrieve this information will not be acted on quickly.

The decision made at an approval threshold should be logged, with the approver's identity and timestamp. This log is part of the campaign's governance record: if the deal is later reviewed, the reviewer should be able to see not only what the agent proposed but what the human approver decided and why.