Basics

Ad-performance terms, decoded: ROAS, CPA, CTR, holdout, incrementality

The handful of terms you keep seeing in ad reports — ROAS, CPA, CTR, conversion rate, holdout, incrementality, attribution. One line each on what they mean, why they matter, and why incrementality is the one that actually tells you if your ads worked.

·อ่าน 5 นาที·ทีมวิศวกร DEXUN AdWhiz

เนื้อหานี้มีเฉพาะภาษาอังกฤษและภาษาจีนตัวย่อ จึงแสดงเป็นภาษาอังกฤษ

Ad reports are full of jargon, but only a handful of terms matter day to day. Here’s one line each — plus the one concept most people skip and shouldn’t: incrementality.

ROAS (return on ad spend)

= revenue from ads ÷ ad spend. ROAS 4 means $4 back per $1 spent. Caveat: it measures correlation, not necessarily causation — some conversions would happen even without the ad.

CPA (cost per acquisition)

= ad spend ÷ conversions. Lower is better, but read it against order value: high-AOV businesses can tolerate a higher CPA.

CTR (click-through rate) & conversion rate

CTR = clicks ÷ impressions (how appealing the creative/targeting is); conversion rate = conversions ÷ clicks (how well the landing page/product converts). High CTR + low conversion rate usually points at the landing page or an audience mismatch.

Attribution

Which ad touch a conversion gets "credited" to. Different models (last-click, data-driven, …) give different numbers — which is why platform self-reported ROAS tends to look optimistic.

Holdout (control group)

A randomly held-out slice of audience or campaigns that gets NO ads / no optimization, as a control. It’s the key tool for turning correlation into causation.

Incrementality

The conversions the ads actually added — the "ran vs. didn’t-run" difference, measured with a holdout. This is the gold standard for whether ads worked: a campaign with a great-looking ROAS can have low incrementality (those people would have bought anyway).

DEXUN AdWhiz applies the same incrementality logic to its own work: it holds out a random ~15% of your campaigns from its recommendations as a control group and compares the two groups on your dashboard, with an uncertainty range — and no number until there is enough data. Per-change savings are labelled as estimates.

อยากลองใช้สิ่งที่บทความนี้อธิบายไหม

ทดลองใช้ฟรี 7 วัน ยกเลิกได้ทุกเมื่อ

เริ่มทดลองใช้ฟรี

บทความที่คล้ายกัน