Wishlists to recoup: 200,000 simulated launches

The question#

Studios ask us some version of this every week: we have N wishlists; is that enough for the deal we want? The honest answer is a probability, not a yes or no. A wishlist count doesn't fix your sales; it sets the centre of a wide distribution. This post builds that distribution from one public benchmark plus a few explicit assumptions, then asks how likely a typical small publishing deal ($600k of funding and marketing, recoupable) is to earn out.

The model is deliberately small. Everything is in research/launch_monte_carlo.py and runs in about two seconds. Where we had to assume something, we vary it and show how much the answer moves.

The model#

Lifetime units sold are launch wishlists $W$ times two uncertain ratios: the first-week conversion $c$ (first-week units divided by launch wishlists) and a tail multiplier $L$ (lifetime units divided by first-week units):

Net receipts per unit $n$ are the price after the store fee $f$ and refunds plus sales tax $t$. At a $20 price, $f=0.30$, $t=0.12$:

Products of lognormals are lognormal, so $\ln R$ is normal with

which gives a closed form for the probability that receipts cover a recoupable budget $B$, and its inverse: the wishlists needed for a target probability $p$:

The exponential term is the uncertainty tax. At $p=0.5$ it vanishes, because the median outcome just needs to equal the budget. Every percentage point of confidence above 50% multiplies the wishlists you need by $e^{\sigma {\Phi }^{-1}(p)}$, and with realistic $\sigma$ that factor grows fast.

Finally, the studio's take under Ammighty's standard terms (15% of net receipts during recoup, 70% after):

Inputs and where they come from#

The one number we take from public data is the conversion median, and it's a moving target. GameDiscoverCo has published several cuts of it, and the figure shifts with price, genre and how big the launch was. We picked the cut that most resembles a funded indie launch, and the sensitivity table shows what happens if your genre converts differently.

Results#

Line: closed form. Dots: Monte Carlo with 200,000 draws per point; the two agree within 0.05 percentage points.

Three things stand out.

The break-even point is higher than most studios expect. A $600k deal has even odds of recouping at about 155k launch wishlists. With 40k wishlists, which is a good result for a small team, the chance is 7%. The median game at 40k sells about 12,574 copies in two years, and 12,574 × $12.32 ≈ $155k doesn't come close to $600k.

Confidence is expensive. Going from a 50% to an 80% chance of recoup requires 2.17× more wishlists (335k). That multiplier is $e^{\sigma {\Phi }^{-1}(0.8)}=e^{0.92\times 0.84}$, and it doesn't depend on the budget at all, only on how uncertain launches are.

Means and medians tell different stories. At 80k wishlists, the median studio take is $46k but the mean is $143k. The average is carried by a minority of breakout launches. If you plan your studio's runway on the average, most of the time you will be disappointed.

The whole surface: wishlists × budget#

The chart above fixes the budget at $600k. Letting the budget vary too turns the curve into a surface. Every point below is the closed-form $\mathrm{Pr}(R\ge B)$ with the same parameters, from 5k to 500k wishlists and $100k to $2M of recoupable spend. The black line is the 50% contour, $W=B/(n\tilde{c}\tilde{L})$, a straight line in log-wishlists. The pin is this post's central case. Drag to rotate.

How much the assumptions matter#

Required wishlists scale linearly with $B/(n\tilde{c}\tilde{L})$, so halve the budget and you halve the wishlists needed. The two parameters we had to assume move things more than any we cited:

The median requirement depends only on the tail multiplier: 103k to 232k wishlists across a plausible range. The 80% requirement depends on both, from 182k (a predictable, long-tailed game) to 631k (a volatile, front-loaded one). Long-tail behaviour is the parameter most worth estimating for your own genre. Comparable titles' review counts over time are a decent proxy; see the review-multiplier method.

What we do with this#

We don't use this model to reject pitches. We use it to size deals. If a game has 30k wishlists six months before launch, we size the recoupable budget so the deal has a fair chance of earning out, and move more of the risk into non-recoupable spend. That's one reason Ammighty's inference cover sits outside recoup. For the studio, the useful inversion is:

With 60k wishlists, $p=0.5$ and our central assumptions, $B_{\max }\approx$ $233k. If you're negotiating a recoupable budget well above that number, expect most outcomes to never pay a back-end. That's fine if you knew it going in, and the 15%-from-day-one clause matters a lot more than the headline 70/30.

Limitations#

• Conversion isn't independent of the tail. Games that convert well at launch often keep selling, so the true joint distribution is positively correlated. Correlation widens the spread of outcomes, so the 80% figures here are, if anything, optimistic.
• One platform. The model covers Steam only. Console, other storefronts, bundles and subscription deals add revenue the model ignores. They usually make recoup easier, sometimes much easier.
• Wishlists aren't exogenous. Publisher marketing changes $W$ itself. The model takes wishlists at launch as given.
• The benchmark is a snapshot. It covers three months of 2024 releases above 5k wishlists. Re-run the script with your own cohort's numbers before relying on any output.

Reproduce it#

The full script is research/launch_monte_carlo.py (Python 3.10+, numpy, scipy; fixed seeds, so you'll get the same numbers). This is the core of it:

c = MEDIAN_CONV * np.exp(SIGMA_CONV * rng.standard_normal(N))   # first-week conversion
L = MEDIAN_TAIL * np.exp(SIGMA_TAIL * rng.standard_normal(N))   # lifetime / first week
R = W * c * L * NET_PER_UNIT                                      # net receipts
p_recoup_mc = np.mean(R >= BUDGET)

# closed form, used to validate the simulation
mu = np.log(W * NET_PER_UNIT * MEDIAN_CONV * MEDIAN_TAIL)
p_recoup = norm.sf((np.log(BUDGET) - mu) / np.hypot(SIGMA_CONV, SIGMA_TAIL))
w_needed = BUDGET / (NET_PER_UNIT * MEDIAN_CONV * MEDIAN_TAIL) * np.exp(-norm.isf(p) * sigma)

Run it with python3 research/launch_monte_carlo.py; results land in research/out/, and the site's tables and charts are rendered from that JSON at build time. To model your own launch, change BUDGET, PRICE and, above all, MEDIAN_TAIL to match your genre's comparables. An earlier draft of this post had the sign of the inverse-normal term flipped, which made the 80% requirement smaller than the 50% one. The Monte Carlo rows caught it.

Frequently asked questions#

Sources#

1. Game World Observer, "Wishlist-to-sales ratio on Steam: GameDiscoverCo benchmark", 6 Dec 2024. gameworldobserver.com/2024/12/06/wishlist-to-sales-ratio-steam-gamedis
2. GameDiscoverCo newsletter, "The state of Steam wishlist conversions". newsletter.gamediscover.co/p/the-state-of-steam-wishlist-conversions
3. Neowin, "Valve introduces new revenue share tiers for Steam" (30% / 25% above $10M / 20% above $50M), 2018. www.neowin.net/news/valve-introduces-new-revenue-share-tiers-for-steam