○ The public list
The 16 checks
Every Blindspot audit starts here, in this order, before a single finding gets written. A finding built on broken data is worse than no finding.
The whole list is public, no email wall. Each check has a pass condition. Score yourself as you go.
Section 1
Coverage. is the data all there
The gap scan
What it catches: days of missing data masquerading as zero-spend days.
a day with zero rows on a paid channel, sitting between normal days, is a hole in the export, not a day off. And platforms disagree on what "a day" is: Meta reports in UTC, Google in account time, Shopify in store time. Misaligned, every day-by-day comparison is off by up to a full day and nobody notices.
Run it yourself
pull daily spend by channel for the last 60 days. Count the rows. Then open each platform's timezone setting and write the three down next to each other.
Pass: no gap days, one timezone story.
The weird day ledger
What it catches: one outlier day quietly bending a whole month's comparison.
find the days far outside your trailing average, and name each one: promo, holiday, outage, or unexplained. Then decide whether it belongs in the comparison. A single spike inside a short window distorts the read more than the effect you're measuring.
Run it yourself
chart daily orders and daily spend for 90 days. Circle anything that jumps out. If you can't name a circle, that's its own finding.
Pass: every outlier named and either excluded or accepted on purpose.
The order census
What it catches: wholesale, marketplace, seeding, and test orders polluting every per-order number you have.
pull the raw order feed and read the actual channel and tag values, all of them. Stores that think of themselves as "web plus subscription" turn out to be carrying Faire, TikTok Shop, gift registries, influencer seeding, and somebody's test orders. Every order goes in exactly one bucket, the buckets have to sum to the total, and an order you can't classify means stop and ask, not "probably DTC."
Run it yourself
in Shopify, group 60 days of orders by sales channel and source name. If a value shows up that you can't explain in one sentence, pull three of those orders and look.
Pass: every order classified, buckets sum to total.
The CAC denominator
What it catches: a blended CAC that reads better than reality because a channel is missing from it.
"blended CAC" has to say which channels are in it. Sum spend across everything you actually run, compare it against the spend inside your CAC math. A declared-active channel contributing zero rows means your blended CAC is a partial CAC wearing a blended costume.
Run it yourself
list every place you spend money on acquisition. Then open your CAC calc and check each one is in the denominator.
Pass: coverage above 95 percent, or the CAC is labeled with what it covers.
Section 2
Truth. is each number real
Zero or broken
What it catches: broken tracking wearing a zero as a disguise.
a zero is one of three things: a real zero, broken tracking, or a bad export, and they look identical in a spreadsheet. Cross-check every zero that matters against a second source. Meta showing purchases with zero purchase value doesn't mean customers stopped paying, it means the pixel value is broken.
Run it yourself
pick your three most important zeros this month. For each, find one independent source that should agree, and see if it does.
Pass: every material zero verified against a second source.
The plausibility test
What it catches: numbers that are present, populated, and nonsense.
divide every rate by its natural denominator and ask if the answer is physically possible for your category. We've watched a connector report a $62 CPM at 5.4% CTR account-wide because impressions pulled low, while spend and purchases tied out perfectly. The row looked fine. The number was impossible.
Run it yourself
take your reported CPM, CTR, CVR, and AOV. Gut-check each against what's possible in your category. Anything strange, verify one campaign directly in the platform's own UI.
Pass: every headline rate survives the gut check or has been verified at the source.
One verdict per metric
What it catches: audits that contradict themselves.
before using any metric, give it exactly one status: trusted, broken, suspect, or verified zero. Then honor the verdict. A report that says "your pixel is broken" on one page and sizes an opportunity off that same pixel two pages later is not a report, it's a coin flip.
Run it yourself
list the ten numbers you rely on weekly. Write one of the four statuses next to each. If you hesitate on any, that hesitation is the answer.
Pass: every number you steer by has a verdict you'd defend.
The break date
What it catches: tracking changes masquerading as performance changes.
a level shift in conversions on the same date across multiple platforms, with no matching move in backend orders, is a measurement event, not a demand event. Any trend that crosses an unacknowledged break date is fiction.
Run it yourself
chart daily conversions per platform for 90 days next to backend orders. Look for a stair-step that shows up in the platforms but not the orders. Note the date.
Pass: no unexplained same-day shifts, or every trend you quote stays on one side of the break.
Section 3
Meaning. do your metrics measure what they claim
What Google is calling a conversion
What it catches: calls, leads, and page views inflating the conversions column.
never trust a search platform's conversions column without decomposing it by action. We've seen a phone call action literally categorized as a purchase. Recompute cost per purchase on purchase actions only, and compare it to the blended number. The gap is often the entire story of the account.
Run it yourself
in Google Ads, segment conversions by conversion action. Read the list. Ask which of these you'd pay for at your target CPA.
Pass: you know your true cost per purchase and it's within shouting distance of the blended number.
The purchase share
What it catches: efficiency metrics quietly built on non-purchases across every platform.
across all platforms, what share of reported conversions are actual purchases? When the non-purchase share gets material, every CPA and ROAS built on the blended count is measuring engagement with extra steps.
Run it yourself
for each platform, split reported conversions into purchases and everything else. Write the percentage down.
Pass: you can state the purchase share per platform from memory, and your targets account for it.
The subscription tell
What it catches: subscription revenue hiding in plain sight, and take rates that are impossible to compute.
the subscription boolean in your export lies. We've seen it read false on every row of stores running thousands of recurring orders. Walk sales channel, then order source, then tags, until recurring orders actually separate from the rest. And remember: the first order of every subscription checks out through the normal web path, so a channel split alone can never tell you your take rate.
Run it yourself
count orders your platform tags as subscription, then count orders from your subscription app's own dashboard. If they disagree, your export is lying to you.
Pass: recurring orders identified by a method you can name, and the two counts agree.
The comp strip
What it catches: seeding, gifts, and test orders reading as a margin problem.
strip zero-dollar seeding orders, hundred-percent comps, and internal tests before reading AOV, discount rate, or margin. A comp-heavy account looks like it has a discounting problem right up until the comps come out, and then the paying economics often land on plan.
Run it yourself
filter to orders under a dollar and orders with 100 percent discounts. Look at what they are. Recompute AOV and discount rate without them.
Pass: you know your paying-customer AOV and it's the one you plan against.
Section 4
Reading. can you compare anything safely
The promo overlay
What it catches: sale-window economics presented as your steady state.
build the promo calendar for the window before reading anything: discount code usage, email sends, ad copy. A month dominated by a sale produces margin and AOV readings that are true for the month and wrong about your business.
Run it yourself
write down every promo that ran in the window, with dates. Lay it over your daily revenue chart. Now re-read your month.
Pass: every margin or AOV claim you make states whether a promo was running.
The visibility floor
What it catches: claiming something is missing when it's actually just invisible.
before saying a flow is missing or underperforming, establish what you can see: which flows exist, whether revenue attribution is present, how deep history goes. A flow you can't observe is not a flow that doesn't exist.
Run it yourself
open your email platform and list live flows next to what your reporting shows. Any flow in one list but not the other is your finding.
Pass: your reporting sees every live flow, or you know exactly which ones it can't.
The split test
What it catches: comparisons with data on only one side.
confirm which splits your data actually supports: brand vs non-brand, new vs returning, by channel. A comparison with an empty side isn't a comparison, it's a guess with a denominator.
Run it yourself
try to pull last month's revenue split new vs returning, and search spend split brand vs non-brand. Note which ones you actually can.
Pass: you know your available splits and don't quote the unavailable ones.
The mix trap
What it catches: the artifact that fools sophisticated teams.
blended CPA can rise 26 percent while every single channel's CPA improved, purely because spend shifted toward the structurally more expensive channel. Before reading any blended number's move, decompose it. If mix explains the move, the "problem" is an accounting artifact, and cutting spend would cut the channels that just got better.
Run it yourself
next time a blended metric moves, compute the same metric per channel for both periods before reacting. If the segments disagree with the blend, believe the segments.
Pass: no blended move gets acted on before its decomposition.
0 of 16
your reporting is making decisions for you. Get the audit before you make another budget call.
The checks tell you whether your data can be trusted. What they don't tell you is what's broken inside it. For that, we scan every account against a library of 85 named failure patterns we've built from opening accounts like yours: things like the silent pixel value break, where your pixel fires on every purchase but reports zero dollars, so Meta optimizes toward nothing.
That library is the audit. Give us read-only access and we run all of it: the 16 checks, the 85 patterns, the sizing. Back comes the Mini Audit: the single biggest recoverable dollar we found, four sized lenses, and the total opportunity. Free, 3 to 5 days, then we're done.