Methodology & data handling
How the memo reaches a verdict — and what it will never claim.
The memo pressure-tests one question: is your current evidence strong enough to justify a specific budget move? Here is exactly how that's judged, and exactly what data we need — and don't.
Every memo runs the same five-stage framework — Decision → Evidence → Distortions → Economics → Verdict. The sections below map to it: the seven risks are the Distortions stage; the eight-step workflow is how the whole framework is executed.
What the memo looks for
Seven ways a spend decision goes wrong.
Every memo checks your decision against these seven risks — the Distortions stage of the framework below. Each is a specific, nameable way the numbers can mislead a scale, cut, or reallocation call.
Attribution mismatch
“Meta says yes, but Shopify and GA4 say maybe not.” Platforms each claim the same orders, so the ROAS you'd act on is inflated.
Margin blindness
“ROAS looks okay, but contribution profit may not.” A move that wins on revenue can still lose on margin once COGS, fulfillment and discounts are in.
Time-window mismatch
“The comparison period is distorting the conclusion.” A promo, launch, or seasonal spike inside the window overstates the baseline you're scaling from.
Channel cannibalization
“One channel is taking credit for demand another created.” Reallocating on stacked per-channel numbers can defund the thing that's actually working — capturing existing demand is not the same as creating new demand.
Returning-customer distortion
“That channel looks efficient because it's harvesting repeat buyers.” Revenue from returning customers can flatter a channel's apparent ROAS and hide what new-customer acquisition actually costs.
Cross-device / window leakage
“The click and the purchase happened on different devices or days.” Journeys that cross devices or fall outside a platform's attribution window get miscredited, so the reported number moves the wrong way.
Evidence insufficiency
“You may be about to act on data that can't support the decision.” Sometimes the honest answer is that the numbers simply aren't strong enough yet — and saying so is the value.
Where the decision actually lives
We find the one slice that decides it — not everything.
Real spend decisions are made hierarchically: you look at blended numbers first, then drill in only where the economics diverge. A blended average can hide the truth, so the memo starts blended and then commits to the one slice that changes the call. Each maps to a report you already have.
Channel / product-group
Meta vs. Google vs. email — or Google Shopping product groups
The classic case: platforms disagree, or a freight-heavy category drags a blended number no one can act on.
Product / SKU
Shopify profit by product & variant
Two SKUs in the same category can behave completely differently on return rates, shipping cost, and ad responsiveness — the blended average hides it.
Discount / promo
Shopify sales by discount code
A sale window can drive volume that looks like healthy paid performance but is really just discounted revenue with thin contribution.
Customer cohort
Shopify cohort analysis by first-order channel & product
If acquisition context changes downstream lifetime value, a decision made on same-week ROAS can be structurally wrong.
If a blended read genuinely settles the decision, that's a valid, honest answer too — the point is a bounded read on the slice that matters, not a deep dive across your whole business.
When the decision rides on repeat purchases or subscriptions, the memo can go one level deeper on the cohort slice — reading breakeven against customer lifetime value (CAC:LTV), not just single-order ROAS. That's an optional add for those cases and needs a little cohort data; the default read stays screenshot-simple.
The workflow
Eight bounded steps, one decision, one slice.
Deliberately narrow. This is a second opinion on one move — not an audit, a rebuild, or ongoing analytics.
- 01
Name one live decision
A single scale, cut, reallocation, or agency-performance call — with the deadline and what happens if it's wrong.
- 02
Start blended, then find the deciding slice
Read the blended store/channel numbers first, then narrow to the one slice — channel, product/SKU, discount/promo, or cohort — where the economics diverge enough to change the call. One slice, not a full-business scan.
- 03
Collect bounded inputs
Shopify, the relevant ad platforms, CRM/lifecycle, and rough margin assumptions — for one decision window, at the chosen slice's grain. Screenshots or exports are enough.
- 04
Normalize the window and definitions
Line up the same dates and metric definitions across every system so the comparison is valid, not apples-to-oranges.
- 05
Reconcile top-line discrepancies
Put each system's claim next to store-recorded net revenue and quantify where — and by how much — their definitions disagree.
- 06
Overlay profit and margin
Translate revenue and ROAS into contribution profit, and find the margin-adjusted breakeven the decision actually rides on.
- 07
Assess confidence
Judge whether the evidence is strong enough: supported, unsupported, or unresolved — with the reasoning shown.
- 08
Deliver the memo
A written memo with the discrepancy table, scenarios, risks, limits, a recommended next action, and what would change the conclusion.
What it claims — and doesn't
Calibrated on purpose.
The value of an independent memo is that it doesn't over-claim. It will not tell you a tool is “broken,” invent a “true” ROAS, or manufacture confidence the data can't support.
It does
- Reconcile your reported numbers and show where they conflict.
- Overlay contribution margin on the decision.
- Mark every claim observed or inferred, with its limits.
- Give a verdict and a recommended next action.
It doesn't
- Replace MMM, MTA, or incrementality testing.
- Stand in for a finance close.
- Rebuild attribution or implement tracking.
- Guarantee a revenue outcome.
Your data
Bounded, redactable, and never required in full.
Trust has to come before you hand over sensitive numbers. So the data ask is staged — you start with the least, and only go deeper if you choose to.
No sensitive access. Just enough to begin.
- A description of the decision and the window.
- Screenshots or CSV exports of the numbers you already see.
- Redacted figures are fine — hide names, emails, order IDs, payment details.
- Rough margin assumptions (COGS, fulfillment, fees).
A deeper look — never required.
- Additional exports for a specific discrepancy.
- A shared report or dashboard view.
- A bounded additional export when the accepted checklist requires it.
- A short call to walk through the decision together.
No passwords or account access.
You never share credentials. Redacted screenshots or exports are the complete operating rule.
Only the decision window is reviewed.
We look at what the one decision needs, not your whole business or history.
Redaction is always welcome.
Customer names, emails, order IDs and payment details can be hidden before you send anything.
Your data isn't reused.
Nothing you share becomes a public example, benchmark, or piece of collateral without your explicit permission.
The accepted Decision Brief records the evidence-transfer method, deletion date, and permitted use. Customer evidence is not reused for training, benchmarks, testimonials, referrals, or public examples without separate consent. Read the full data terms.
See the method in a real deliverable.
Read a full sample memo end to end, then get one for your own decision.