Platform · Proof
A smaller number that answers the question nobody else can
Competitors report total revenue from the group that got the message. That number is always prettier, and indefensible the moment someone asks what you would have earned by sending nothing. The difference is smaller — and it answers.
The limit we state first
“A holdout on every campaign” sounds rigorous. Statistically it isn’t.
detail
So the method follows the cohort size, not the appetite to report. The system always shows a confidence interval, and says plainly when it is still too early.
An unsettled result reads “not enough to conclude,” never a flattering number.
The rules in force
The method is chosen by the size of the treated group
treated
1,204
Sent the message
held back
212
Deliberately not sent
The held-back group never receives it. If they buy at the same rate anyway, the message caused nothing — and the console says so in those words.
| Treated group size | Method | What can be reported |
|---|---|---|
| 1,000 and above | True holdout, 10–20% withheld | Reportable per campaign |
| 300–1,000 | Holdout, but too small to conclude on any single send | Pooled every 90 days, never per campaign |
| Under 300 | Time-shifted, or matched against a behaviourally similar cohort | No holdout — it would return noise, not signal |
The 281-person champion cohort in the MST Golf sample falls in the band that must be pooled over 90 days rather than measured per campaign — and we surface that limit in the product, not in a footnote.
ROI Tracker
And the edge that cuts back
detail
detail
90-day rolling difference
holdout 15%
+7.4%
95% confidence interval · ±2.1 pts
- Revenue from the sent group
- ฿2,418,000
- Expected from the holdout
- ฿2,251,300
- Countable difference
- ฿166,700
- Cost per returning customer
- ฿148
An illustration of the display, not a measured result from a live customer
What it actually prints
Including the row that says it does not know yet
| horizon | per head · sent | per head · held | difference | 95% CI | verdict |
|---|---|---|---|---|---|
| T+7 | ฿87 | ฿74 | +฿13 | −4.1 to 22.6% | Not enough yet |
| T+30 | ฿239 | ฿201 | +฿38 | 2.8 to 27.1% | Worked |
| T+90 | ฿604 | ฿512 | +฿92 | 6.2 to 24.9% | Worked |
The T+7 row is the point. Its interval crosses zero, so the console reports that it cannot conclude — rather than rounding it up into a win. Figures illustrate the display; they are not a live customer's result.
What separates this from the market
what most tools report
Total revenue from the group that got the message
A big number that looks good on a slide, and quietly includes everyone who was going to buy anyway. Asked what you would have earned by sending nothing, it has no answer.
what parallax reports
The difference against a group that got nothing
Far smaller, and the only number that holds up under questioning — the only one a budget decision can actually rest on.
The Proof layer is what makes churn hard — not through a contract, but because cancelling means explaining the number you gave up.
เริ่มจากไฟล์ยอดขายที่คุณมีอยู่แล้ว
หนึ่งไฟล์ หนึ่งบรีฟ และวิธีวัดผลที่เตรียมไว้ให้ ไม่ต้องติดตั้งอะไรเพิ่ม