CPA vs CPI vs CPM: How to Choose the Right Pricing Metric (2026)
CPA vs CPI vs CPM: what each metric prices
CPA vs CPI vs CPM is the standard comparison media buyers use when deciding how a campaign should pay for attention, installs, or outcomes. The three acronyms look similar; they price three different layers of the funnel.
|
Metric |
Full name |
You pay when… |
Typical use |
|---|---|---|---|
|
CPM |
Cost per mille |
1,000 impressions are delivered |
Reach, brand, upper funnel, auction diagnostics |
|
CPI |
Cost per install |
A user installs the app (or equivalent first-open event) |
App UA, store growth, MMP-optimized campaigns |
|
CPA |
Cost per action / acquisition |
A defined conversion fires (lead, purchase, signup) |
Performance, lead gen, web or app post-install goals |
CPM is inventory-priced. CPI is install-priced. CPA is outcome-priced. Mixing them in one dashboard without labeling the denominator is how “cheap” traffic looks expensive after finance closes the books.
How the math connects
Start from spend:
- CPM = spend / (impressions / 1,000)
- CPC (often the bridge) = spend / clicks
- CPI = spend / installs
- CPA = spend / conversions
Useful identities when volumes are honest:
- CPI ≈ CPM / (impressions per install × 1,000) — or via CTR × install rate from click
- CPA ≈ CPI / install-to-action rate (app) or CPM / (impression → conversion rate × 1,000) (web)
If install-to-purchase is 5%, a $2 CPI implies a $40 CPA before creative quality changes. That single conversion rate usually explains more than switching bid types.
When to prefer each model
Prefer CPM when
- You need predictable delivery against a reach or frequency plan
- You are testing creative or placements and care about auction CPM / quality signals first
- Brand lift or view-through matters more than last-click CPA this week
CPM alone does not prove efficiency. Pair it with viewability, brand study, or downstream CPA so cheap impressions do not become expensive customers.
Prefer CPI when
- The product is an app and the first success event is install / first open
- You are scaling UA with an MMP and need install volume for learning
- Post-install events are sparse early, so optimizing CPA would starve delivery
CPI is not “free” of fraud or accidental installs. Validate with retention (D1/D7) and a post-install CPA or ROAS guardrail. Related DeepClick reading: Target CPA vs Target ROAS.
Prefer CPA when
- You have a stable conversion event and enough volume for the platform to learn
- Finance cares about leads or purchases, not impressions or installs
- You can pass clean conversion APIs / pixels and exclude junk events
CPA can hide weak funnel stages: a “cheap” lead CPA with 2% sales accept rate loses to a pricier CPA with 20% accept. Define the action carefully (qualified lead vs raw form fill).
Side-by-side decision guide
Ask four questions:
- What does finance call a win? Impression, install, or revenue/lead → maps to CPM / CPI / CPA.
- Do you have enough conversion volume? If not, start CPM or CPI (or maximize clicks) and graduate to CPA / tCPA / tROAS.
- Is the surface web or app? Web lean CPA/ROAS; app often CPI first, then in-app CPA.
- Can you trust the event? Server-side / MMP events beat brittle pixels when comparing CPA vs CPI vs CPM across channels.
Also watch account controls that clip delivery, such as ad account spending limits and prepaid vs postpaid billing modes—billing friction can fake a metric problem.
Common mix-ups
- Reporting CPI while optimizing a CPA bid strategy (or the reverse)
- Comparing Meta CPM to Google Search CPA without noting auction and intent differences
- Treating SKAN / modeled conversions as identical to last-click CPA
- Using view-through CPM “efficiency” to justify poor click CPA
- Forgetting currency and timezone when stitching MMP CPI to ad-platform CPA
Practical rollout
- Instrument one source of truth for installs and for the money event.
- Baseline current CPM, CTR, install rate, and conversion rate for 7–14 days.
- Choose the bid/pricing model that matches the scarcest reliable event you can feed the algorithm.
- Add a secondary guardrail (e.g., CPI campaign with max CPA or min D1 retention).
- Revisit weekly; as volume grows, shift weight from CPM/CPI toward CPA/ROAS without rewriting history in the same chart.
Bottom line
CPA vs CPI vs CPM is not a beauty contest—it is a choice of which funnel step you are willing to pay for. Use CPM to buy attention, CPI to buy installs, CPA to buy outcomes, and always translate between them with conversion rates so “cheap” does not silently mean “unprofitable.”
