Operations
Paid social agencies service standards: write metric cards
How to write paid social agencies service standards as metric cards: population, event, numerator, denominator, clock, owner, escalation and thresholds.
Paid social agencies service standards hold up only when each measure is written down the same way twice. A metric card does that work: it fixes one measure, what counts towards it, how it is evidenced, the clock it runs on and the owner who answers for it.
In a paid-social agency those cards cover work that runs on Meta (Facebook and Instagram), TikTok, LinkedIn, Pinterest and Snapchat: campaign set-up and approval, creative and copy approval against platform policy, ad-account access and roles, tag and pixel integrity, delivery pacing, and spend reconciled to platform invoices. The media measures they fix — cost per acquisition (CPA), cost per lead (CPL), return on ad spend (ROAS), click-through rate (CTR), cost per mille (CPM) and frequency — only compare when the card is written the same way twice.
What to take away
- A metric card is reproducible when two people run it and reach the same decision.
- Every card names population, event, numerator and denominator. It then fixes period, timezone, exclusions and evidence system.
- Add query version, clock, owner, threshold method and escalation route.
- Paid-social cards apply those same fields to creative approval, ad-account access, tag and pixel integrity, pacing and platform-spend reconciliation.
- Derive the first threshold from four to six weeks of your own unthresholded data.
- These are service levels you set, not agency benchmarks or recommended service levels.
Write a metric card first
A card is a blank specification, not a result. It fixes what counts, over what window, and what happens next.
Every card names its population, event, numerator and denominator. It records the period, timezone, exclusions and evidence system. Two people must reach the same decision from it, so the query version, clock, owner, threshold method and escalation route belong on the card too.
Metric card fields
- Population, event, numerator, denominator or unit
- Period, timezone, exclusions, evidence system
- Query version, clock, owner
- Threshold method, escalation
Analysis Function quality guidance asks for fitness for purpose, stated methods and disclosed limitations. Apply that discipline to a commercial measure.
Intake cards follow the target audience definitions you agreed, because a request that cannot name its decision role is not complete.
Before you act
- Write the card before you set any threshold.
- Name the population, event, numerator and denominator.
- Record the evidence system, query version, clock and owner.
- Derive the threshold from your own unthresholded data.
- Escalate recurring gaps to the named owner.
Example: a creative approval completeness card
The card below is illustrative. Change the population and timezone to match your operation.
- Populationad creatives submitted for approval during the stated period and timezone.
- Eventthe first approval decision on a creative.
- Numeratoreligible creatives holding every mandatory element, including client sign-off, claim evidence, a platform-policy check and tracking parameters.
- Denominatoreligible creatives, less those withdrawn before review.
- Evidenceversioned creative register and a saved query.
- Clocksubmission to recorded decision, paused only for a named client dependency.
- Ownercreative delivery lead.
- Escalationrecurring gaps go to the client owner.
For illustration only, a team that cleared 91 per cent in three of the last four weeks would set 90 per cent as its first threshold.
The other cards at a glance
Beyond the worked example, an agency run usually carries cards for creative and copy approval completeness, policy and compliance checks, ad-account access and roles, tag and pixel integrity, live-render defect rate, delivery pacing, correction elapsed time, and media-spend reconciliation to platform invoices. Each names the same fields, then a clock, an owner and an escalation route.
Other cards at a glance
| Card | Clock | Owner | Escalation |
|---|---|---|---|
| Approval-evidence completeness | Presentation to pass, fail or withdrawal | Release authority | Delayed decisions to the client owner |
| Live-render defect rate | Activation to end of observation window | Live monitor | Material mismatch to the release authority |
| Correction elapsed time | Ticket registration to accepted containment | Incident lead | Breaches to client and specialist |
| Access-removal completion | Recorded role change to verified removal | Security administrator | Exceptions to security and client owners |
| Billing reconciliation difference | All records received to finance sign-off | Finance | Unexplained differences before further media authority |
The glance table carries the clock, owner and escalation for each card. Population, event, numerator, denominator, period, exclusions and evidence sit on the cards themselves, in the versioned register the team works from.
One refusal still stops release in the approval card, so claim evidence must be held before distribution under the CAP Code's substantiation rule, rule 3.7, enforced by the Advertising Standards Authority. Access due points sit beside the contract clauses that fix ownership, because roles in a platform account such as Meta Business Manager or TikTok Ads Manager are an asset question.
Correction elapsed time records elapsed minutes or hours from registration to accepted containment. Rejected tests and duplicates stay on the trail. Report the distribution and unresolved cases.
Deriving a first threshold
- Run the card for four to six weeks with no threshold attached.
- Record the weekly value and note what limited it, such as reviewer cover.
- Set the first threshold at a level the team cleared in three of the last four weeks.
- Hold that threshold for one quarter before raising it.
- Raise it only after two consecutive periods above it and added capacity.
A threshold is a capacity statement you negotiated, not a hope. Record the dependency behind the number so the next reviewer can check it.
Govern thresholds and exceptions
Approve thresholds before live work and version every definition. A change to platform, contract, workflow or evidence system triggers review. GOV.UK's service monitoring guide recommends defined checks, alerts and issue records. Treat that as an operational analogy, not an agency rule.
A dashboard colour never waives an independent gate, so an exception needs a named owner, a trigger and a response time.
Do not lift a threshold from published market figures. Their populations and reporting periods rarely match your own, and that mismatch is what makes them poor benchmarks. Trade bodies such as IAB UK, the IPA, ISBA and the Advertising Association publish useful market context, but none of it sets the service level your own cards define.
Common questions
What must every metric card specify?
Each card names its population, event, numerator and denominator. It also fixes the period, timezone and exclusions. The evidence system, query version, clock, owner and escalation route complete it.
How do I set the first threshold?
Run the card unthresholded for four to six weeks. Set the first threshold at a level the team already cleared in three of the last four weeks, hold it for a quarter, then raise it only with added capacity.
When should an exception be escalated?
Escalate when the trigger fires: recurring intake gaps, a delayed decision, a render mismatch, an access breach or an unexplained billing difference. Each card names the owner and the response time.
Do these cards count as official benchmarks?
No. They are service levels you define and govern for your own operation. Analysis Function guidance informs the method, not the numbers.
