The pre-launch compliance check that catches costly mistakes

In July 2013 the UK Home Office sent two billboard vans around six London boroughs for a week. The panels read "In the UK illegally? GO HOME OR FACE ARREST", with the line "106 arrests last week in your area". The ASA later upheld a complaint about that number: the arrest data covered a wider area than the boroughs the vans actually toured. The complaints about offensiveness were not upheld. So the element that failed was the smallest and most checkable thing on the panel, and it failed because nobody had traced it back to the dataset it came from before the vans rolled.

That is the work here: the order the checks run in, what sits in the evidence file, who signs, and the failure modes that walk straight through a completed checklist. Which regulator can do what to whom is the foundations lesson's territory, and it is assumed from this point on.

The gate, in sequence

Sequence matters, because each check can invalidate the ones after it. Sending artwork to a fair-treatment reviewer before the central claim has been traced wastes the reviewer, then wastes them again on the rewrite.

1. Claim tracing

One row per factual assertion. Each row carries the claim as written in the creative, the source (a line in a named dataset, a published figure with its publication date, a named official who will stand behind it), the person who checked that the source says what the claim says, and the date the source expires. The substantiation test itself belongs to the consumer protection lesson; what the gate adds is mechanical discipline about scope.

Public-body statistics rarely break by being false. They break by being true of something else: national data presented as local, a cumulative total presented as an annual rate, a modelled projection presented as a count, last year's figure presented in the present tense. The Home Office number was real. It was real about the wrong geography.

2. Audience, targeting and channel

Vulnerability screening produces a set of exclusions and adaptations, and that work is covered by its own lesson. The gate's job is narrower and duller: confirm that those exclusions exist in the media plan, in the platform targeting file, and in the insertion order the agency actually placed. An exclusion that lives only in a slide deck has no effect on who sees the ad on Tuesday.

3. Rights and consent

Model releases with their expiry dates, guardian consent for anyone under 18, licence terms per channel and per territory, music rights, and permission for any real case study to be identifiable. The recurring failure is reuse: a release signed for a 2022 campaign, an image relicensed for web but not for out-of-home, a case study whose subject has since asked to be removed and told a caseworker rather than the comms team.

4. Sign-off record

Names, dates, and a specific artefact. Not "v3 final" but a timestamped PDF or a file hash, so that the thing approved can be compared with the thing that shipped. Record who is allowed to override a red flag and what an override looks like in writing, because verbal overrides are the ones nobody remembers the same way six months later.

What the evidence file has to hold

The ASA can open an investigation on a single complaint and will ask the advertiser for evidence on a short deadline. If assembling that evidence takes three weeks of archaeology across four inboxes, the campaign is already off air by the time the answer arrives.

Keep in one place, per campaign: the claim-tracing sheet, the approved artefacts, consent and licence documents, the media plan with exclusions, the targeting parameters as placed, and the sign-off log. Retention should outlast the campaign by years, not months, because complaints about public-sector advertising often arrive after a policy changes and someone goes looking for what was promised.

The sanction is usually not the expensive part. The ASA cannot fine. It orders the ad withdrawn and publishes the ruling, which stays searchable forever. What hurts a public body is the write-off: production plus unspent booked media, plus the officials' time spent explaining the write-off. That figure is the one read out at a committee hearing.

The failure modes that survive a completed checklist

• The approved asset is not the shipped asset. The 6-second cut drops the qualifier that made the claim accurate. The resize crops the disclosure. The email subject line gets written after sign-off, by someone who was not in the review.
• The ad clears and the destination does not. Landing pageLanding pageA standalone web page built for a single campaign goal, designed to maximise conversions by removing distractions and focusing visitors on one action.View full definition → copy, the call-centre script, the FAQ, and the reply an officer types at 11pm from the campaign's social account are all part of the same communication in a regulator's eyes, and are usually reviewed by nobody.
• The message is tested against the code, not against the people it describes. The Australian Government's 2021 consent education campaign went live in April with a set of videos, including the widely mocked milkshake metaphor, and was pulled within days after educators and sexual assault services said the metaphors obscured what consent actually means. Nothing in it was misleading in the way the Home Office number was misleading. It cleared production, cost several million dollars by public reporting, and failed on a check that did not exist: sector review by the specialists whose subject it was.
• Translations approved by someone who cannot read them. Community-language versions carry the same legal exposure as the English original and typically get a fraction of the scrutiny.
• The gate runs once, the campaign runs twelve weeks. Statistics get restated, policy changes, a caretaker or pre-election period begins and neutrality rules tighten. Write down the triggers that force a re-check rather than assuming someone will notice.

The rapid lane, bounded

Emergency communications will not wait 48 hours, and pretending otherwise means the gate gets bypassed permanently rather than occasionally. Define the fast path in advance: two named people, one from comms and one from legal or policy, both reachable out of hours; a hard limit on what it can cover (safety-critical information, not recruitment or brand work); and retrospective full review within five working days, logged. An unbounded fast lane is just an absent gate.

Back to the vans

Run the sequence against the 2013 panels and the failure is visible before printing. Claim tracing asks which dataset "106 arrests last week in your area" comes from and what "your area" means in that dataset; the answer, once written in a cell next to the claim, does not match the six boroughs on the route. Audience and channel asks who is standing at that bus stop, which is a different question from who the message addresses. Rights and consent is clean. Sign-off asks who owns the number.

One completed row would have caught it. That is the whole economics of the gate: it converts unease into a written question that someone has to answer with a source.

Making the gate stick

• Build it into the campaign tool (Asana, Monday.com, a shared form, whatever is already in use) as a required field before publish unlocks. A gate that depends on remembering is a gate that fails under deadline.
• Set an SLASLAA formal commitment defining the service level a provider guarantees to a customer, with measurable targets and consequences if they are missed.View full definition → you can actually meet, 24 to 48 hours from final creative. Unrealistic SLAs are the main cause of route-arounds: teams reclassify paid as organic, or split a campaign into "assets" and "content" so half of it never enters the queue.
• Measure the share of live assets with a complete sign-off record. If it is not above 90 percent, the process is theatre and you are relying on luck.
• Keep a rolling log of decisions. Patterns show up (the same image type flagged four times, the same statistic queried by three reviewers) and belong in the creative brief, upstream of production.

🎬 [VIDEO: "How Charity Ads Get Regulated (and Sometimes Banned)" - youtube.com/@AdvertisingStandardsAuthority - the ASA's own channel explains real ruling case studies on charity and public sector advertising complaints]

Key Takeaways

• Run the checks in order: trace every claim to a dated source, confirm targeting exclusions are in the placed media plan, clear rights and consent, then record a named sign-off against a specific timestamped artefact.
• Scope is where public-sector statistics fail. National data shown as local, and cumulative totals shown as annual rates, are what regulators like the ASA actually rule against.
• Build the evidence file during the campaign, not after a complaint. Regulators work to short deadlines, and the real cost of a withdrawn campaign is the written-off media spend, not the sanction.
• Assume the checklist will be passed and the campaign will still fail: shipped assets that differ from approved ones, unreviewed landing pages and reply scripts, unread translations, and the Australian consent campaign's failure mode of clearing every legal test while the audience it described rejected the metaphor.
• Define a bounded rapid lane with two named approvers and retrospective review, or the whole gate gets bypassed the first time something is urgent.