From Day-2 to LTV: A Cohort Model You Can Build in a Spreadsheet

Retention · 2026-07-13 · 9 min read · By CROCO Games

Why day-2 retention is the earliest honest predictor of player LTV, and a step-by-step spreadsheet cohort model — columns, formulas, sensitivity math — using CROCO's published benchmarks.

Every operator wants to know a player's lifetime value, and most wait months for the answer their day-2 retention already gave them. The link between early retention and LTV is not mystical — it is arithmetic: LTV is the area under the retention curve times the value of an active day, and the retention curve's shape is largely determined by its first points. This article builds the full chain in a spreadsheet: from D2 to a fitted curve to an LTV estimate to sensitivity analysis, using CROCO's published portfolio benchmarks — 13.78% day-2 and 26.89% day-7 retention — as the worked example.

Why day 2, specifically

Day-2 retention — the share of a cohort that returns the day after first play — is the earliest point where the retention curve has a slope, and slot retention curves are well-behaved: they fall steeply, then flatten, in a shape closely approximated by a power law. Once the first points pin the curve down, the tail follows. A cohort that starts at 10% D2 and one that starts at 14% D2 are not 4 points apart at day 90 — they are on different curves, and the gap compounds through every later day. That is why D2 is a leading indicator rather than a vanity stat: it is measurable 48 hours after an acquisition campaign or a game launch, while D30 makes you wait a month to learn the same thing.

One definitional note before the math: fix your convention (calendar-day return vs 24-48h window, per-game vs per-casino) and never mix conventions across cohorts. CROCO's published figures — 13.78% D2, 26.89% D7 — are game-level portfolio averages; D7 exceeding D2 tells you these are cumulative return metrics (returned at least once within N days), not day-N snapshots. Both conventions are legitimate; comparisons across them are not.

The spreadsheet, column by column

One row per daily cohort (or weekly, for smaller volumes). Columns:

  1. A — cohort date. First-play date of the cohort.
  2. B — cohort size. New players that day.
  3. C..N — active on day d for d = 1, 2, 3, 7, 14, 30, 60, 90. Raw counts from your game-round log.
  4. Retention r(d) = active(d) / cohort size — one column per tracked day.
  5. Curve fit. Fit r(d) = a·d^(−b): with only a spreadsheet, take logs — ln r(d) = ln a − b·ln d — and use SLOPE() and INTERCEPT() over your observed (ln d, ln r) pairs. The exponent b is your decay rate; smaller b = stickier product.
  6. Projected active days over horizon H: sum a·d^(−b) for d = 1..H (a plain SUMPRODUCT over a helper column of day indices). This is the area under the curve — expected active days per new player.
  7. Value per active day. From your own P&L: net revenue per player-active-day, by segment if you have it. (For scale: CROCO's portfolio average bet is €1.77; combine with your observed spins-per-session and hold to sanity-check the revenue-per-day figure your finance data gives you.)
  8. LTV(H) = projected active days × value per active day. Report at fixed horizons — LTV(90), LTV(180) — never as an unbounded number; power-law tails flatter infinity.

That is the whole machine: eight column groups, two statistical functions, no Python. Its power is not precision — it is that it turns "retention improved" into "projected active days per player went from 4.1 to 4.9, which at €X per active day pays for the campaign twice over."

The sensitivity table: what +1pp of D2 is worth

The model's most useful output is the derivative, not the level. Build a small table: recompute the curve fit with r(2) nudged up one percentage point (holding the other observed points), and read off the change in projected active days. Because early points dominate the fit, a +1pp improvement at D2 typically drags the whole curve up and adds several percent to area-under-curve — the exact multiplier is your number, and that is the point of owning the model. Now every proposal has a common currency: a game swap that adds 0.5pp D2, an onboarding flow worth 1pp, a crash campaign that spikes day-1 but decays faster — all become euros over horizon, comparable on one axis.

Run the same table on value-per-active-day to keep yourself honest: retention work and monetisation work trade in the same currency, and sometimes the cheaper point is on the other axis. Auer and Griffiths' 2023 field study (43,731 players, five countries) is the caution here — game structure moves session length far more than bet size, so content levers mostly buy you the retention axis; bet-size gains have to come from the player mix your acquisition brings in.

Where a provider actually moves the needle

Be suspicious of anyone who claims content alone transforms LTV — but the retention axis is genuinely content-sensitive, and the mechanisms are specific:

CROCO publishes its numbers — 13.78% D2, 26.89% D7 across the portfolio — precisely so operators can put them in column form and check the compounding themselves rather than take "engaging content" on faith. Ask any provider you evaluate for the same, and for the player-LTV definitions behind their claims.

Failure modes of cohort models

Four ways this spreadsheet lies to you if you let it: mixed acquisition quality — a cheap-traffic campaign drags cohort curves down and reads as "product got worse"; segment your cohorts by source before concluding anything. Small cohorts — retention on a 40-player cohort is a coin flip; aggregate to weekly cohorts until r(d) stabilises. Promo contamination — a reactivation blast on day 6 puts a bump in r(7) that the power law will happily extrapolate; annotate campaign dates in the sheet. Horizon greed — fitted tails beyond your observed window are hypotheses, not assets; never report LTV at a horizon longer than ~3× your oldest observed cohort.

Frequently asked questions

Why does day-2 retention predict LTV? Because LTV is the area under the retention curve, and slot retention curves follow a steep-then-flat power-law shape pinned down by its earliest points. Cohorts that differ at D2 are on different curves, and the gap compounds daily.

What is a good day-2 retention rate for slots? It depends on definition and market, which is why published, definition-stated benchmarks matter. CROCO's portfolio averages 13.78% D2 and 26.89% D7 (cumulative-return convention, game level) — use them as a reference point, not a universal bar.

How do I calculate player LTV from retention data? Fit r(d) = a·d^(−b) to observed retention via a log-log linear fit, sum the fitted curve over your horizon to get expected active days, then multiply by net revenue per active day. Report at fixed horizons like LTV(90).

Can better game content increase player LTV? Mostly through the retention axis: field data shows game structure strongly affects session length but barely affects bet size. Content and lobby choices buy you active days; value-per-day comes largely from player mix.

Key takeaways