use case · make the call
Neo measures the sales each channel really adds.
An ad platform counts every sale its ads touched, including the ones that would have happened anyway. Neo designs a geo test on your store’s sales, states before launch what it can detect and what it costs, then reads it and gives the factor to apply to the platform’s figures.
4,000 advertisers and agencies already use Creaboost.
A test written before launch: what it can detect, what it costs, what you will do with the result.
Maison Lune’s founders no longer want to approve the Meta budget on Meta’s own ROAS. Camille asks Neo to design a test, and attaches sales by département.
Before we approve the spring Meta budget, the founders want to know what Meta really brings in. Can we measure it?
Camille Durand · Wed, Oct 21 · 10:20
Neo · 10:46
- Shopify · live · 18 months
- export · sales by département
- Meta · live · spend by region
- memory · economics of Oct 14, events
Meta France · planned · plan of Oct 21 by incrementality_testing · Jan 11 → Jan 29, observation until Feb 7 · Shopify, export, Meta
Switch Meta off in areas that make 20% of sales for 19 days: the test will put a number on what Meta brings in, for about $1,000 of margin.
Meta goes first: 74% of spend, and the widest gap with the store, a 3.9 ROAS with views for an acquisition MER of 1.8. The test compares randomly drawn areas, with and without Meta, over the same days and on the store’s sales. The expected lift is above 2.5 times the noise: the test can size the iROAS, and the spring budget gets decided on Feb 10.
- minimum detectable effect
- 15%
- 1.7 × 9% placebo noise
- expected lift
- 29%
- conservative iROAS 1.4
- holdout areas
- 20%
- of sales · 89 areas in the draw
- cost of the test
- ≈ $1,000
- of margin · $9,900 of spend held back
decision at stakeThe Meta budget for February to April, approved on Feb 10: keep it, raise it, or move it to what adds the most sales.
the plan, section by section
- why meta firstrank 1
The largest spend, with the most doubt.
Meta and Google claim 114% of the store’s sales. Brand search ranks second: a short test, in other areas, over the same period. Catalog is included in the Meta test; its own test comes later.
- 74% of spend
- ROAS 3.9 · 7-day click + 1-day view
- Google brand · rank 2
- method89 areas
France’s départements, Greater Paris as one block, drawn at random.
Splitting a metro area lets the holdout see the ads at work or on the commute. Neo sorts the areas by sales, forms blocks of similar size, draws at random within each block, and keeps the best-balanced draw over 18 months. Belgium and Switzerland stay out of the test.
- seed 4127
- placebo noise 9%
- bias 0.6%
- powerMDE 15%
With 20% of sales in holdout areas, the test can size the iROAS.
Expected lift = Meta spend × conservative iROAS ÷ store sales, over the last four weeks: $73,000 × 1.4 ÷ $356,000, or 29%. At 10%, the test would see an effect without sizing it.
holdout · detectable effect · sizable from · 19 + 9 days
- 10%
- 24% · 35%
- 20% · chosen
- 15% · 23%
- 30%
- 12% · 18%
- 50%
- 10% · 15%
- cost≈ $1,000
$9,900 of spend held back, about $1,000 of margin lost.
Priced at the 2.6 target iROAS, so the cost is never understated: about $26,000 of sales lost at a 42% margin, minus the spend saved. If Meta brings in less than the target, the test costs less.
- sales lost ≈ $26,000
- target iROAS 2.6
- break-even 2.4
Frozen during the test and the observation window: Meta’s budget and settings in the test areas, Google’s spend and targeting, no regional promotion, no stockout of a hero product. Sales will dip in the holdout areas: agree on it with the founders before launch.
to decide togetherShopify has no region for most French orders: the plan relies on your export by département. The 2.6 target iROAS, above the 2.4 break-even, is a proposal to set with you before launch.
if the result is…, written before launch
- 01
Keep Meta, then test a +20% step.
if the lift is real and the iROAS clears 2.6 across most of the 80% interval.
readout Feb 9 · step test in the spring - 02
Move budget toward the most incremental pockets.
if the lift is real and the iROAS stays under 2.6. Money moved toward what adds the most sales pays, even when every iROAS is low.
spring budget · Feb 10 - 03
Hold the budget, and rerun the test longer or with more areas.
if the interval is wide and crosses zero. Learning more before moving is a valid outcome.
new plan proposed Feb 10 - 04
Decide nothing on this number, and rerun the test.
if the estimate is negative, or too high to be plausible: it is imprecision.
new plan proposed Feb 10
Saved in Docs · Maison Lune · incrementality test · Meta France · planned · 10/21/2026
- Open the full plan
- Follow the test every Monday
how Neo gets there
Areas with and without ads, compared on your sales.
- data
The store’s sales, area by area.
Neo reads Shopify live, by region, over 9 to 24 months. When addresses carry no region, as is common in France, it reads the export of sales by area you upload. On Meta: spend by region, on 7-day click, 7-day click + 1-day view, and incremental attribution.
- skill · incrementality_testing
The noise in your sales first, then the test’s power.
Neo first measures the noise in your past sales: the minimum detectable effect is about 1.7 times that noise, and sizing the iROAS takes 2.5 times. Never under 14 days; 3 to 4 weeks for Meta conversion campaigns. A platform A/B test does not measure incrementality: each cell picks its own audience.
- memory · Maison Lune
Break-even and the calendar, read before the plan.
Gross margin in economics sets the break-even, 2.4; events keep Black Friday, Christmas and Valentine’s Day out of the window. The measured factor then goes into Meta’s metrics & funnel map, and the reviews apply it.
One test, from plan to readout, in a single deliverable.
The plan is saved in Docs. During the test, Neo adds a check every week; after the observation window, the readout and the factor.
Explore Docs →docs
The incrementality test.
Summary, decision and criteria, why this test first, design and power, cost, what stays frozen, setup steps for your team, log, readout, factors, limits and sources.
tasks
A check every Monday while the test runs.
It checks that the test holds, never the lift: spend in the holdout areas, changes, events, stockouts. The readout goes out on the first run after the observation window.
docs · readout
The factor to apply to the platform’s figures.
iROAS = platform ROAS × factor. With a 0.71 factor, a 3.5 ROAS gives an iROAS of about 2.5; reaching a 3.0 iROAS takes a ROAS of about 4.2.
skills
Then applied by Neo’s other methods.
The business review, the paid media audit and each platform’s pilot read the factor. It holds for the season and spend level tested: Neo offers to refresh it every quarter.
go further
More use cases, and the product pages behind them.
questions
What people ask us about incrementality.
Why not an A/B test inside Meta?
In a platform A/B test, each cell’s algorithm picks its own audience: the result mixes the creative with the people it was shown to. It is for iterating inside the platform. For incrementality, Neo designs a geo test, read on your store’s sales.
Isn’t Meta’s incremental attribution enough?
It is a platform reading: Meta’s model estimates which of the conversions it tracks its ads caused, with no control group you choose and no view of sales elsewhere. Neo shows it as context and computes the factor against it too, without ever using it as the factor.
Does Neo set the test up?
Neo reads, analyzes and recommends. It has no write access to your ad accounts: the plan gives, platform by platform, the campaigns in scope, the areas to exclude, the dates and what to check on day one. Your team sets it up; Neo checks every week that the test holds.
How long does a test run, and what does it cost?
Never under 14 days: 2 to 3 weeks for brand search or retargeting, 3 to 4 for Meta conversion campaigns, 4 to 6 for video on TikTok or YouTube. The cost is priced before launch: spend held back, sales and margin lost.
What if the store doesn’t know where orders come from?
Neo then reads an export of sales by region or area that you upload. When a geo test cannot be precise enough, it proposes a platform lift study: your team creates it, Neo reads the result.
Measure what your channels add.
Book a demo: we connect your accounts and your store, and Neo designs your first test in front of you, with what it can detect and what it costs.