Cost per thousand impressions
CPM
Cost per thousand ad impressions. In influencer marketing, it is calculated from the actual reach of a post, not the subscriber count: a channel with 200,000 subscribers may get anywhere from 30,000 to 90,000 views per post, which changes the CPM by a factor of three.
Example. A channel with 200,000 subscribers; the post got 46,000 views and cost $1,000. CPM = $1,000 / 46 = $21.70. Calculating from subscribers would give $5 — four times more optimistic and wrong.
Cost per target action
CPA
Cost of one action you consider a result: a signup, a deposit, an install. This is the only metric we use to cut a placement — reach and views do not factor into that decision.
Example. A flight cost $13,000 and delivered 840 verified signups. CPA = $15.50. If the cap is $13, the flight is unprofitable overall — but within it, four channels delivered $10 and two came in at $34. We cut the two, not the flight.
Revenue per unit of ad spend
ROAS
Revenue divided by ad spend. For long-cycle products (fintech, SaaS), measuring ROAS on day seven is meaningless: two months can pass before a deal closes, so we lock the attribution window upfront.
Example. Spent $9,000; revenue from acquired users over 60 days was $43,000. ROAS = 4.8x. On day seven it was 1.3x — if we had evaluated then, the channel would have been cut.
Lifetime value of a customer
LTV
How much revenue a customer generates over their lifetime. Sets the CPA ceiling: if LTV is $90, paying $100 for acquisition only works at a loss.
Example. Average order $27, purchases per year 3.1, margin 42%. LTV = $27 x 3.1 x 0.42 = $35. The CPA ceiling for one-year payback is $35, not $27.
Engagement relative to reach
ER
Share of the audience that reacted to a post. A high ER on a large channel more often signals fraud than genuine engagement — we look at ER together with the distribution of reactions over time.
Example. A post with 46,000 views, 1,200 reactions and 40 comments. ER = 1,240 / 46,000 = 2.7%. The number itself is fine; what matters is the pattern: if 900 reactions arrived in the first ten minutes, the engagement was purchased.
How many people saw it
Reach
Number of unique people who saw a placement. On Telegram it is post views over 48 hours; on YouTube it is video views; for stories it is unique viewers. These cannot be added together across platforms.
Where it breaks. A Telegram post got 46,000 views, a YouTube video — 120,000, stories — 18,000 unique viewers. Adding them up to 184,000 is wrong: some people saw both, and a story view and a completed video view are different events. They get added anyway, and the report says "reach 184,000" — a number that never existed.
How many times it was shown
Impressions
Total number of times an ad was shown, including repeat exposures to the same person. Always higher than reach. Confusing impressions with reach is the easiest way to inflate a report by a factor of two.
Where it breaks. The platform report says "1,400,000 impressions." Unique people behind them — 300,000. If the presentation labels the first number as reach, the figure grows almost fivefold without a single additional person actually seeing the ad.
How many times per person
Frequency
Average number of times one person sees the ad. Above four is usually wasted budget and irritation.
Example. Reach of 300,000 with 1,400,000 impressions gives a frequency of 4.7. Above four, response usually drops and annoyance rises — the budget is better moved to a new audience.
Share of viewers who clicked
CTR
Share of viewers who clicked. In influencer marketing, a low CTR does not mean failure: part of the audience arrives via search, not through the link.
Example. Of 46,000 views, 690 clicked through — CTR 1.5%. Yet sign-ups totaled 210, of which 90 arrived via a branded search. Counting clicks only would miss 43% of the result.
The same people counted twice
Audience overlap
Share of subscribers common to two channels. When buying across ten channels in the same niche, overlap can reach 40% — you pay twice for the same person.
Example. Ten channels in the same niche delivered a combined 610,000 views, but unique viewers totaled 380,000. Overlap of 38% — every third viewer was paid for twice.
The watch-through curve
Retention
Share of viewers still watching at each second. A drop at second fifteen says more about the script than the total view count.
Example. Of 10,000 who started the video, 4,100 made it to second 15 and 1,800 to the end. The drop between seconds 12 and 18 means the script loses the viewer there, and that is the segment to reshoot.
Cost per install
CPI
Price of one app install. In iGaming and crypto it means nothing on its own: half of the installs never reach registration, and cheap installs from incentivised traffic almost never reach a deposit. We look at CPI only next to the share of installs that reached the target action.
Example. Network A gives installs at $0.90, network B at $2.40. From A, 1.1% reach a deposit; from B, 6%. A deposit costs $82 via A and $40 via B. The cheap install turned out twice as expensive.
Average revenue per user
ARPU
Revenue for a period divided by the number of active users. It shows how much an average player or trader brings and, together with CPA, answers the main question: does acquisition pay off. Do not compute ARPU over all registrations — half of them never pay.
Example. In a month, 2,400 active players brought $61,000. ARPU = $25. A deposit costs $40, so payback comes no earlier than the second month, and keeping the channel makes sense only if retention confirms it.
Share of users lost in a period
Churn
The share of users who stopped being active during a period. The flip side of retention: if 30 of 100 players are still playing after a month, churn is 70%. Churn shows which source brings people for the long run and which — for a single bonus.
Example. Channel A: of 500 players, 190 are active after 30 days, churn 62%. Channel B: of 500, 60 are active, churn 88%. At the same CPA, channel B is three times more expensive per player who stayed.
A group of users from one week
Cohort
Users who arrived in the same period and are then counted together: how many of them pay on day 7, day 30, day 90. Only cohorts show whether traffic got better — total revenue grows from old players too.
Example. Week-one cohort: 1,000 registrations, 90 paying by day 30. Week-four cohort after new creatives: 1,000 registrations, 140 paying. Monthly revenue grew 20%, but it was the cohorts that showed the creatives did it, not the season.
The uplift that would not exist without the ads
Incrementality
The difference between what happened with advertising and what would have happened without it. Attribution credits a source with everything that passed through it; incrementality asks how much of that would have come anyway. Tested by switching a channel off in some geos or for part of the audience.
Example. Retargeting showed 1,200 deposits a month by attribution. Switched off in two of six countries: deposits there dropped by 140, not by the 400 expected from its share. The incremental contribution was a third of what was credited.
Gross gaming revenue
GGR
Bets minus player winnings for a period — what the operator earned before bonuses, taxes and payment fees. In affiliate programmes the partner’s share is calculated from GGR or NGR, so it matters to fix in advance what gets deducted.
Example. Players wagered $400,000 and won $352,000. GGR = $48,000. If the contract is a share of GGR, the base is $48,000; if of NGR, bonuses and fees are deducted first, and the base may be half that.
Net gaming revenue
NGR
GGR minus bonuses, gaming taxes, payment fees and sometimes anti-fraud costs. The base of most rev-share deals. Every deduction must be listed in the contract: “other operator expenses” without a list eats half the base.
Example. Monthly GGR $48,000. Bonuses $9,000, tax $6,000, payment fees $2,400. NGR = $30,600. A partner on 30% of NGR gets $9,180; on GGR they would get $14,400.
A share of revenue instead of a fixed price
Rev-share
A payment model where the source or partner receives a percentage of revenue from the players they brought, usually of NGR and for life. Cheaper at the start and more expensive over the long run; the contract must fix the calculation base, the term and the negative-balance rule.
Example. A network offered 35% rev-share instead of $60 per deposit. Its players bring $180 NGR a year on average: rev-share costs $63 per player in the first year and keeps being charged after. CPA is cheaper for a test; rev-share for a proven source with short LTV.
