Worked example · Synthetic data

Inside a Blueprint: the Northstar worked example

One diagnostic, shown in full. A quarter of Shopify orders reconstructed from gross sales down to contribution after advertising, every figure marked with the evidence behind it, two findings that met the threshold, five areas that produced nothing, and a net profit figure that was never stated because the data could not support it.

Request your Blueprint The method behind it

Northstar Outdoor is a synthetic rehearsal store · GBP, excluding tax

Synthetic example — not a customer result

Northstar Outdoor is a synthetic rehearsal merchant — fictional brand, synthetic data. Not a customer, and not a customer result. It exists so the method can be shown in full, on a complete dataset, without publishing a real merchant's economics. Every figure on this page is transcribed from one generated Blueprint render dated 2026-09-23 and is reproduced unchanged — including the figures that are unflattering and the ones the report refused to state.

1 · The store

What was analysed

The inputs, stated before any conclusion, so the scope of everything below is fixed.

Store: Northstar Outdoor — synthetic rehearsal store
Analysis period: 1 April – 30 June 2026 (Q2 2026)
Compared against: 2025-12-31 – 2026-03-31
Currency basis: GBP, excluding tax
Orders in period: 5,812
Data sources: Shopify orders, products and refunds; one connected advertising account

An ecommerce diagnostic is only as good as the boundary it declares. Autopsify reads a Shopify store on read-only access and whatever advertising accounts are connected to it. It does not read the merchant's bank, payment processor, warehouse or accounting ledger — which is why several lines further down this page end in Unknown rather than in a plausible number.

2 · The question

Revenue nearly doubled. Was the business better off?

Sales moved +87.7%. Contribution moved +49.2%. On a revenue chart this quarter looks like an unambiguous success: net sales excluding tax went from £626,724 to £1,176,057, and orders rose +79.7%.

That is the exact situation a reporting tool handles badly. Both figures are true, both are up, and neither answers the question an operator actually has: did the economics of the business improve, or did it simply do more of something at a worse rate? Answering that requires reconstructing what happened to each pound of sales — not displaying the total.

The diagnostic was scoped to establish that and nothing more: what the evidence could support about the store's Shopify economics in the period, and what it could not.

3 · Economic reconstruction

From gross sales to where the evidence stops

Excluding tax throughout. Each line is the one above it, less a cost that could actually be evidenced.

Economic reconstruction for Northstar Outdoor, 1 April – 30 June 2026, in GBP excluding tax. Each row states a line of the economic bridge, its figure, and whether that figure was observed in source data, calculated from it, or is unknown.
Line	Figure	Evidence
Gross salesShopify order total. Includes tax and shipping; already net of discounts.Observed	£1,411,263	Observed
Less taxCollected for a government. Never the merchant's money.Observed	−£235,205	Observed
Net sales excluding taxIncludes shipping charged to customers; shipping cost is not available, so it is not deducted.Calculated	£1,176,057	Calculated
Less refundsReduced by the period's effective tax rate, because Shopify does not report tax per refund.Calculated	−£58,461	Calculated
Sales after refundsCalculated	£1,117,596	Calculated
Less cost of goodsCalculated	−£753,127	Calculated
Gross profitSales after refunds less cost of goods, before advertising and before every cost below.Calculated	£364,469	Calculated
Less advertisingObserved	−£118,309	Observed
Contribution after advertisingNOT profit. Excludes payment fees, fulfilment, duties and overhead.Calculated	£246,160	Calculated
Net profitNot calculable from the data Autopsify holds.Unknown	Not calculable	Unknown

Why the ladder stops at contribution

Contribution after advertising is £246,160. It is not profit, and the report says so on the line itself. Three costs sit between it and net profit — payment fees, fulfilment, overhead — and none of them is available from Shopify or from the advertising platforms.

The tempting move is to assume them: a typical payment fee percentage, a category average for fulfilment, a guess at overhead. Autopsify does not, because a net profit figure built on three assumptions is not a measurement of this store — it is a measurement of the assumptions. Net profit is reported as not calculable, which is a true statement about the evidence rather than a false statement about the business. How the reconstruction is built.

Two distinctions this page never collapses

Gross profit is not contribution

Gross profit (£364,469) is sales after refunds less cost of goods, before advertising. Contribution after advertising (£246,160) is that figure less advertising spend. They differ by the £118,309 of advertising in the period, and they carry different margins — 32.6% and 22.0%. Reporting one as “profit” would overstate the other by a six-figure sum.

Two refund figures, both correct

Refunds as issued were £70,153. The bridge deducts £58,461 — the same refunds reduced by the period's effective tax rate, because Shopify does not report tax per refund and the tax portion was never the store's money. Both appear on this page. They are different measurements of the same event and must not be treated as one.

4 · The evidence model

Known, supported, unknown — applied to every figure

The classification travels with the number. A reader can see which conclusions rest on source data and which rest on a documented derivation.

Observed — read directly from source data

Gross sales, tax collected, order count and advertising spend were present as fields in the store's own records and in the connected advertising account. Reading them required no transformation and no assumption. Advertising spend of £118,309 is observed, not modelled.

Calculated — derived by a stated, deterministic method

Net sales excluding tax, sales after refunds, cost of goods, gross profit and contribution after advertising are arithmetic on observed figures, with the scope declared. Each sale was costed with the unit cost in force on its order date, and 100.0% of revenue in this window carried a recorded unit cost — which is the reason contribution after advertising was calculable at all.

Unknown — the evidence does not reach

Payment fees, fulfilment and overhead. Not small, not approximately nothing, not assumed from a benchmark: unestablished. Every figure that depended on them — net profit — is withheld.

Unknown is an evidence limit, not a value of zero

This is the rule that decides what this diagnostic would and would not say. An absent measurement and a measured zero look identical in a spreadsheet cell and mean opposite things. Had the three unavailable costs been silently read as zero, this report would have stated a net profit of £246,160 — a number that is wrong by whatever those costs actually are, presented with the same confidence as the figures that were measured.

Nor does Unknown mean a figure is unimportant, or a classification of Known mean good news. Advertising spend is Observed and it is the largest single deduction on the page. Net profit is Unknown and it is the figure the operator most wants. The evidence model in full.

5 · The measurements

What the period actually measured

Every figure against the same comparison period. No industry benchmarks appear anywhere in the report — the target is the store's own prior quarter.

Net sales excluding tax: £1,176,057; +87.7% vs comparison period; Calculated
Orders: 5,812; +79.7% vs comparison period; Observed
Average order value (excluding tax): £202; +4.4% vs comparison period; Calculated
Gross profit: £364,469; +62.2% vs comparison period; Calculated
Gross margin: 32.6%; −13.7% vs comparison period; Calculated
Contribution after advertising: £246,160; +49.2% vs comparison period; Calculated
Contribution margin: 22.0%; −20.6% vs comparison period; Calculated
Advertising spend: £118,309; +98.0% vs comparison period; Observed
Revenue with a recorded unit cost: 100.0%; +0.0% vs comparison period; Calculated

Scope-specific figures, kept separate

Contribution before advertising, costed products only: £360,731; +62.2% vs comparison period; Calculated
Contribution margin, costed products: 32.4%; −13.7% vs comparison period; Calculated
Refunds: £70,153; +82.5% vs comparison period; Observed; As issued. The bridge deducts £58,461 — the same refunds less the tax portion.

Decomposed across the products carrying a recorded unit cost in BOTH periods — 98.8% of this period's costed revenue. A margin computed on the products that carry a recorded cost is reported as exactly that, never as the margin of the whole catalogue. The costed-scope figures and the headline figures differ because their scopes differ — not because the arithmetic disagrees.

6 · Measurement coverage

How much of the business each figure could see

Coverage decides how strong a conclusion the evidence can carry, and the two coverage figures in this diagnostic point in opposite directions.

Revenue with a recorded unit cost

100.0%

Complete cost coverage, revenue-weighted. This is why the ladder reached gross profit and contribution at all. Earlier runs on this store sat below 97%, and because an unknown cost is never read as zero, contribution after advertising could not be stated in them.

Attribution coverage

32%

Only 32% of revenue carried a recognisable marketing source. Below half, every acquisition figure is a floor rather than a measurement, so no channel-level conclusion was issued — in either direction.

Coverage is not performance. Low attribution coverage does not mean the advertising is working badly; it means the question cannot be answered yet. Treating a coverage gap as a negative result is the same error as treating an unknown cost as zero, in the other direction. How coverage is assessed.

7 · The findings

Two findings met the evidence threshold

Each states what it proves and, in the same card, what it does not.

Revenue is growing but contribution margin is shrinking

Contribution margin fell from 27.7% to 22.0% (5.7 points) while revenue grew +88%.

Classification: Scenario
Claim strength: Bounded Diagnosis
Severity: High
Confidence: High
Coverage: 100%
Owner: Founder

Illustrative counterfactual — not money lost

£67,214 (illustrative, this period)

Illustrative: net revenue excluding tax for the whole business in this period, held at the preceding period's margin of 27.7% instead of 22.0%. A counterfactual on a measured base — NOT money lost, not a forecast.

What the evidence supports

A smaller share of each pound of sales survived cost of goods in this period than in the preceding one, for the products whose costs are known.

What it does not prove

That contribution fell. It did not — the pounds grew; the share of each pound did not.

Signal to re-read: Contribution margin, monthly

Most revenue cannot be traced to a source

Only 32% of revenue can be attributed to a recognised marketing source.

Classification: Observed
Claim strength: Bounded Diagnosis
Severity: Medium
Confidence: High
Coverage: 32%
Owner: Paid acquisition

Impact not quantified

Attribution coverage is a measure of what can be SEEN, not of money moving. Attaching a figure would require assuming what the unattributed revenue would have shown, which is the very thing that cannot be established.

What it does not prove

That the unattributed revenue is organic — it is unknown, which is not the same. Or that advertising is underperforming: low coverage limits what can be concluded in either direction.

Signal to re-read: Attribution coverage, monthly

Synthetic example — not a customer result

The £67,214 attached to the first finding is an illustrative counterfactual on a measured base: this period's net revenue excluding tax held at the preceding period's margin instead of this period's. It is not money lost, not recoverable profit, not a saving and not a forecast. It is shown because an operator deciding where to spend attention needs the order of magnitude a margin movement represents — and it is qualified every single time it appears.

8 · Why the margin moved

One movement, three separable causes

A five-point margin fall looks like one problem in the total. It was three, and they need different responses.

Decomposition of the contribution margin movement into three drivers, with each driver's effect in percentage points, its share of the combined movement, and the strength of the evidence behind it.
Driver	Effect	Share	Evidence basis
Unit cost	−2.84 pts	54%	Established
Realised revenue per unit	−1.29 pts	25%	Supported
Product mix	−1.08 pts	21%	Established

Unit cost carries 54% of the combined movement. Cost per unit rose on products that kept their revenue share. This is the merchant's own recorded cost, dated, so the movement is measured rather than inferred.

A rising unit cost, a falling realised price and a shift in which products sell look identical in the headline margin and call for three different decisions — a supply conversation, a pricing decision, and a merchandising one. Separating them is the difference between an ecommerce diagnostic and a number on a dashboard.

Established against a driver means the measured evidence accounts for that component of the movement within the stated scope. It does not mean an experiment has shown that changing the variable will produce a future outcome. Autopsify never issues a causal claim: establishing cause requires a controlled experiment, and observational data cannot produce one.

9 · What produced nothing

Five of eight areas examined returned no material finding

A result in its own right. A diagnostic that finds problems everywhere is measuring nothing.

Checkout completionAnalysed; did not cross the materiality threshold this period.
Advertising efficiencyAnalysed; did not cross the materiality threshold this period.
RefundsAnalysed; did not cross the materiality threshold this period.
Product concentrationAnalysed; did not cross the materiality threshold this period.
Order volumeAnalysed; did not cross the materiality threshold this period.

Each cleared area is listed with what was measured in it, not merely omitted. An area that was never examined and an area that was examined and found clean are different results, and a report that shows only its hits gives a reader no way to tell them apart.

And one observation, deliberately not actioned

Footwear Granite Lite 70 revenue has fallen

Footwear Granite Lite 70 revenue fell 59% against the preceding period.

Exposure £926 – £1,157 · low confidence · monitor, do not act

Real, measured, and not worth acting on this period. It is published so the operator can watch it move — a finding whose claim strength is only “observation” tells you a number changed, not why.

10 · The number that was not published

Three different numbers, deliberately not combined

Advertising spend sits inside the contribution calculation, so counting an advertising opportunity and a contribution opportunity separately counts the same pounds twice.

So no total appears. The alternative — a single headline “money at risk” figure — would be the most quotable number in the report and the least defensible, because it would double-count the same pounds across overlapping findings. Opportunity estimates are not additive and are published individually, never summed.

11 · Verification

How each finding was set up to be proved or disproved

Every finding carries a baseline, a window, a success condition and a failure condition, written before the change is made.

Finding 01 · Margin compression

Measure: Contribution margin, costed products, before advertising
Baseline: 37.5% (2025-12-31 to 2026-03-31)
Window: 91 days from the date the change takes effect
Success: Margin on the same costed scope recovers by at least half the 5.2-point fall — to 34.9% or better — WITHOUT net revenue excluding tax falling against the same comparison period.
Failure: Margin is flat or lower at the end of the window, or it recovers only because revenue fell — which is a smaller business, not a better one.

Finding 02 · Attribution coverage

Measure: Share of counted revenue matched to a marketing source
Baseline: 32% (2026-04-01 to 2026-06-30)
Window: 30 days from the date every paid link carries its parameters
Success: Coverage reaches 70% or above of counted revenue, and the orders that remain unmatched are identifiable by channel rather than blank.
Failure: Coverage stays below 70%, or rises only on one channel while others stay blank — which means the measurement is still a floor.

The two windows are different, and that is the point

Attribution coverage is re-read after 30 days because it responds immediately: a link that carries its parameters starts recording a source on the next order, and a broken one shows up within days. Contribution margin is re-read after 91 days because the movement being tested is a portfolio average over a quarter, and reading it weekly invites acting on noise.

Nothing here claims that every finding is proved within 30 days. The re-diagnosis cycle establishes whether the figure moved, on the same rules and the same scope as the baseline — including the answer that it did not. Some findings need a longer observation window than one cycle provides, and the report says which. Verification and re-diagnosis, and what the 30-day re-reading covers.

The decision rules are written in advance for the same reason: judging finding 01 on margin alone would score a shrinking business as a success. Margin up with revenue held is the result being tested. Margin up with revenue down is not, and must not be recorded as one.

12 · The deliverable

What an operator is left holding

A reconstruction they can audit

Eleven lines from gross sales to the point the evidence stops, each with its source, its arithmetic and its evidence class. A traceability appendix takes every material claim back to the field and the formula behind it. A number you cannot audit is a number you should not act on.

A ranked first action, not a list of twenty

Two actions for this week, each tied to the finding that justifies it, with the owner, the effort and the signal that would show it working. Where no finding justifies an action, none is invented.

The limits, in writing

What each finding does not prove. Which comparisons could not be made and why — purchase completion could not be compared across these two periods, because the comparison period begins before Shopify's 90-day abandoned-checkout retention. Which costs are missing, and therefore which conclusions are unavailable.

A baseline to be judged against

The measure, the scope, the window, the success condition and the failure condition for each action, recorded before the change — so the next reading is a verification rather than a fresh opinion.

Autopsify does not make the change. The actions are the merchant's, in the merchant's store, on read-only access. What the Blueprint supplies is the decision support: what is happening, how strongly the evidence supports it, what to do first, and how you will know whether it worked.

Run the same diagnostic on your store

£497 for one cycle. A short conversation first, then a readiness check on your data — if the evidence cannot support the analysis, we say so before you pay.

Request your Blueprint Create your account