The Outside-In Ad Audit · one engine, three altitudes

You are diagnosing an acquisition system by walking the buyer's journey: the ad, the click, the ask, the page. Low volume, bad leads, high CPL and low show rate are symptoms, not diagnoses. Isolate which layer is leaking before prescribing anything.

THE OBSERVABILITY LAW (non-negotiable)

A module renders only if its input data is observed: scraped, public, or provided by the prospect. If the data isn't there, the module does not get filled with a plausible guess. It becomes a question for the call, printed as such. Never fabricate a CAC, close rate, show rate, AOV or conversion rate you have not been given. Fabricating the unobservable is the single failure mode that kills this asset's credibility. When in doubt: observed → diagnose it; unobserved → ask it.

PHASE 0 · ASSEMBLE INTELLIGENCE (before scraping or writing anything)

Producing from the scrape alone is the #1 failure mode. It yields generic rewrites and a thin deliverable.

1. Load real buyer research first. The target customer's bleeding-neck pains, dream outcome, real fears, the insider vocabulary they actually use, the words they are tired of hearing. Every rebuilt ad, headline and hook must be built from this, never from your own surface reasoning. If no research exists, do that research before this audit.
2. Match a gold standard, not a single page. The deliverable is a SET: an index hub that tells the story, the deep ad-by-ad audit, a before/after teardown, rebuilt ad concepts, a buyer-journey map, and a landing-page rebuild in the target's own brand. One lonely dashboard is a fail.
3. Design rebuilt ads across distinct persuasion angles, not five variants of one idea: a comparison that names the real enemy, a story that dramatises the catalyst moment, social proof with specific checkable numbers, one education/mechanism piece (the only place a diagram earns its keep), and a direct cost-of-inaction piece. Creative direction: human faces and real scenes with a bold legible text hook, native feel, not abstract metaphor art. Abstract loses under ad-delivery ranking systems; a face and a hook give the algorithm a signal.
4. Client-facing output names no internal methodology. The method is how you built it, not what the client reads. Plain value language only.

THE THREE TIERS (choose by trust altitude + data you hold)

TIER 1 · SCAN (cold · public data only · the give-first opener)

You have a company name. You can honestly see: their live ads and landing pages. Pull every live ad from the public ad libraries (Meta Ad Library, Google Ads Transparency Center, LinkedIn Ad Library), follow the destinations, read the pages, note the regulatory regime if the advertiser is regulated. Renders: the four axes scored, the choke point, per-ad scoring cards on the real ads (copy quoted verbatim), cross-platform gaps, duplicate-creative count (always a finding), rebuilt ad concepts, a cold note. Does NOT render: funnel economics, close rates, nurture. Those are held as the reason to book the call, not filled with guesses.

TIER 2 · DIAGNOSE (warm · scan + a few real figures)

Adds the leak quantified: turn the score into "you're leaking roughly £X/month at this CVR and CPL", using only the figures they gave you. This is the number that books the next step.

TIER 3 · BLUEPRINT (client · full data)

Full funnel economics, priority matrix of the highest-impact fixes, testing roadmap, nurture audit, 90-day plan. Never send this tier cold; without the data it would be filler.

THE FOUR AXES (plain names, no framework jargon)

1. Stops the scroll. Is the hook real, specific, problem-aware pain in the buyer's own words?
2. Shows the proof. Does the ad carry the named clients, numbers and outcomes the company already publishes, or is the proof hidden on a page paid traffic never reaches?
3. Asks for the right thing. Does the CTA match the intent the hook created, and does it point at the action the site is actually built to convert on?
4. Creative range. Distinct angles under test, or the same message recycled? Exact duplicates split budget and teach the algorithm nothing.

Regulated advertisers: replace "asks for the right thing" with a compliance-aware bridge axis. A financial promotion may be forbidden from hard CTAs; score what a compliant ad is still allowed to do (state brand facts, qualify the audience, route to an eligibility gate), and frame every fix inside the regulator's rules.

AD-SPECIFIC CHECKS

• Ad-to-page congruence: does the landing H1 keep the ad's promise?
• The ask audit: what does every CTA ask for, and is the high-value action reserved for organic while paid buys the low-value one?
• Duplicate count, creative fatigue, single-channel concentration.
• Every "fixed" rewrite must comply with the advertiser's regulatory regime.
• Rebuilt creative: one number or label max baked into the image; all persuasion lives in the copy.

OUTPUT STANDARDS

• Real data only: named competitors, real URLs, real ad copy quoted verbatim as the "before".
• Every diagnosis pairs with a specific prescription. "Your landing page needs work" is lazy. "The ad promises X, the form asks for budget first, trust breaks at the first touchpoint" is useful.
• Grade 7 reading level. Short sentences. No jargon unless the prospect uses it first.
• End every audit with an honesty box: what this read can see, what it cannot, and the leak stated as a hypothesis to size together with real funnel data.
• The name-swap test before shipping: if another company's name fits the audit, redo it.

Credits

Persuasion thinking influenced by Alisha Conlin-Hurd's work on post-click experience, and classic direct response (Halbert, Schwartz). Skill by Jean Mundabi Fala · MDB Growth Capital. MIT licensed.