Sample readout

What an Audit readout looks like.

This is the shape of the document you pay for at readout. Every finding is graded settled, likely, or needs a test, and carries a dollar figure wherever one is defensible. It is written to be forwarded upstairs as-is.

Anonymized sample · The company, its conversion events, and every figure below are fictional · The structure is what every Audit readout follows

Prepared forAcme Pro, B2B software, subscription
Period audited90 days, Meta, Google, LinkedIn
Spend in window$540,000

The verdict

Your platforms report 745 Paid Subscriptions over the window. Your billing system recorded 412. The gap is not fraud and not a broken pixel. It is three separate counting problems stacked on top of each other, two of which you can fix in a week. The cost per Paid Subscription you have been planning on is $725. The real one is $1,311.

Reported cost per Paid Subscription
$725
$540,000 over 745 platform-reported subscriptions
Actual cost per Paid Subscription
$1,311
$540,000 over 412 subscriptions in billing
3 settled 2 likely 2 need a test

Findings

01

Meta and Google both claim the same 187 Paid Subscriptions.

Settled

What we foundMeta reports 388 Paid Subscriptions and Google reports 296. Matching click identifiers against billing shows 187 subscriptions that appear in both, each counted twice in the number your dashboards add up. Both platforms are behaving as designed. Neither knows the other exists.

Why it is settledEvery one of the 187 is a row in your billing table with both a Meta click and a Google click inside the attribution window. There is no alternative reading.

Worth: 187 of 745 reported subscriptions, 25% of the total, do not exist as separate customers.
02

The Pro Trial Started pixel fires on the pricing page, not the trial form.

Settled

What we foundMeta recorded 1,940 Pro Trial Started events. Your product database recorded 1,213 trial form submissions in the same window. The pixel is placed on pricing-page load, so a reload or a second visit counts as a new trial. Meta has been optimizing three campaigns toward pricing-page traffic and calling it trials.

Why it is settledWe reproduced it: loading the pricing page twice fired the event twice with no form submitted. The 60% overstatement matches the reload rate in your analytics.

Worth: $41,000 of the last 90 days of Meta spend sat in the two ad sets where the pixel-to-form ratio was worst, at 2.4 to 1.
03

LinkedIn is graded on a 30-day window while Meta is graded on 7 days.

Settled

What we foundLinkedIn reports 61 Paid Subscriptions on its default 30-day click and 7-day view window. Meta reports on 7-day click and 1-day view. Put LinkedIn on the same 7-day click basis and its count drops to 37. The two channels have never been compared on the same terms.

Why it is settledThe window settings are visible in each account, and re-running LinkedIn's export on 7-day click reproduces the 37 exactly.

Worth: LinkedIn's cost per Paid Subscription is $2,162 on a like-for-like basis, not the $1,311 in your channel report.
04

Google Brand is taking credit for demand that Meta created.

Likely

What we foundOf the subscriptions Google attributes to Brand search, 71% had a Meta click in the prior 14 days. Brand search is the last click on a path that Meta started. Brand spend was $52,000 in the window.

Why it is likely, not settledWe can see the Meta click. We cannot see whether the person would have searched your name anyway. The pattern is strong, the direction is clear, and the exact share defended is not observable from the data you have.

Worth: an estimated $15,000 to $21,000 of the 90-day Brand budget is paying for demand created elsewhere. Treat the range, not a point.
05

Seat Invited predicts a paid subscription three times better than Pro Trial Started.

Likely

What we foundTrials convert to Paid Subscription at 11%. Trials where a second seat is invited in the first seven days convert at 34%. Seat Invited is a cleaner optimization event than Pro Trial Started, and it cannot be inflated by a page reload.

Why it is likely, not settledNinety days is one cohort. The invite-first trials skew to larger companies, which may convert better for reasons that have nothing to do with the invite. Four more weeks on one campaign would settle it.

Worth: no figure yet. The dollar value depends on whether Meta can find these people at an acceptable cost, which is what the four weeks measures.
06

Whether Meta prospecting is incremental at all.

Needs a test

What we found63% of Meta-attributed subscriptions had an organic visit to your site in the 30 days before the Meta click. Meta may be reaching people who were already coming. Or it may be the thing that brought them back. Attribution data cannot distinguish the two.

The test that settles itA geo-holdout: 20% of regions, matched on baseline subscription rate, receive no Meta prospecting for eight weeks. The read is Paid Subscriptions per region-week in exposed versus held-out regions. Spend required in exposed regions: about $60,000. Priced separately under Experiment Design.

Worth: $270,000 per quarter of Meta spend depends on the answer. That is why it gets a test and not a guess.
07

9% of browser events have no matching server event.

Needs a test

What we foundYour Conversions API sends server-side events alongside the pixel, and the two are meant to deduplicate on a shared event ID. 9% of browser events arrive with no server twin. Either the server is dropping them, or they are genuinely browser-only and the count is right.

The test that settles itTwo weeks of stamping every event with a diagnostic parameter on both sides, then matching the logs. Low cost, no spend change, and it decides whether your Meta count is 9% high or exactly right.

Worth: up to 35 of the 388 Meta-reported subscriptions, if the drop is real.

What to do Monday

  1. Move the Pro Trial Started pixel to the form submit. One change, one day, ends finding 02.settled
  2. Make billing the only source of truth for Paid Subscription. Send it to every platform through server-side events and stop counting the platforms' own versions. Ends finding 01.settled
  3. Report LinkedIn on 7-day click whenever it sits next to Meta. Ends finding 03.settled
  4. Run Seat Invited as the optimization event on one Meta campaign for four weeks. Settles finding 05.likely
  5. Decide on the geo-holdout. It is the only way to answer finding 06, and it is the question your board is actually asking.needs a test
Settled

Observed directly in your data and reproduced. We would sign it.

Likely

The evidence points one way, but a plausible alternative explanation survives. We say which one.

Needs a test

Cannot be settled from the data you have. The readout names the test, its cost, and how long it takes.

Your numbers, graded like this.

Three weeks. Read-only access. You pay at readout, and only if the report addresses every criterion we agreed on up front.