eCPI

Definition

Effective cost per install is total marketing spend divided by all installs attributable to that spend, including organic uplift and referred installs, rather than only the installs a network claims.

CPI answers an accounting question: what did the invoice divide by the claimed installs come to. eCPI attempts a business question: what did each user actually cost. The two diverge most for channels that drive search and word of mouth, which is precisely where CPI misleads a budget decision most expensively.

Three costs that get called CPI

Almost every argument about install cost is three metrics being given one name. Naming them separately resolves it immediately.

Blended CPI is the most commonly quoted and the least meaningful: it credits spend with every organic install that would have arrived regardless, including users who came from an unrelated PR mention. It always looks best, which is why it appears in so many decks.

The last row is uncomfortable and usually closest to the truth. When incremental CPI comes out above network CPI, it means part of the claimed volume was installs you were getting anyway — see click spamming and view-through attribution for the two mechanisms that most often produce that gap.

Estimating organic uplift without a holdout

A geo holdout is the correct method and is not always available. The workable substitute is the organic-to-paid ratio observed during genuinely quiet periods, applied as a multiplier — imperfect, but explicit about its assumption, which blended CPI is not.

-- Step 1: baseline organic rate from weeks with negligible spend.
WITH baseline AS (
  SELECT AVG(organic_installs) AS baseline_organic_per_week
  FROM weekly_installs
  WHERE total_spend < 500          -- effectively unpaid weeks
),
-- Step 2: organic above baseline is treated as spend-driven uplift.
uplift AS (
  SELECT
    w.week,
    w.total_spend,
    w.paid_installs,
    w.organic_installs,
    GREATEST(w.organic_installs - b.baseline_organic_per_week, 0) AS organic_uplift
  FROM weekly_installs w
  CROSS JOIN baseline b
)
SELECT
  week,
  total_spend,
  paid_installs,
  organic_uplift,
  ROUND(SAFE_DIVIDE(total_spend, paid_installs), 2)                  AS network_cpi,
  ROUND(SAFE_DIVIDE(total_spend, paid_installs + organic_uplift), 2) AS ecpi,
  ROUND(SAFE_DIVIDE(total_spend, paid_installs + organic_installs), 2) AS blended_cpi
FROM uplift
ORDER BY week DESC;

Referrals belong in the same calculation. If your paid installs produce referred installs at a K-factor of 0.4, then 10,000 paid installs eventually yield about 16,700 users, and dividing spend by the larger figure is the honest comparison against a channel with no referral loop at all.

eCPI only matters next to value

A lower eCPI is not automatically better. Channels differ in the quality of users they deliver, and the cheapest install is regularly the least valuable one. The decision metric is the ratio between eCPI and the revenue those users produce over a fixed window.

Pair eCPI with a cohort table split by source. Cost per install is a single number at a single moment; the cohort's revenue and retention curves are what determine whether that cost was worth paying, and the two channels that look identical on eCPI frequently look nothing alike by day 30.

• Report network CPI, eCPI and blended CPI together, never one alone.
• State the uplift assumption next to any eCPI figure.
• Include incentive costs for referral channels; they are spend.
• Re-baseline organic quarterly and after any store feature.
• Validate the uplift factor against a holdout at least once a year.