Desktop App Conversion
Review downloadable-software funnels, Steam wishlist sales ratios, refunds, and installation warnings with explicit dates and samples.
Written and reviewed by Eric Han · Co-founder & CTO, reopt
Key takeaways
- Most public desktop-software stage rates date from 2001–2010. Data from the 2020s are available mainly for Steam games.
- In 2007, the free Firefox browser had 70% visit-to-download, 57% download-to-install, and 49% install-to-daily-active after 30 days: about 20 of 100 visitors remained.
- A 2009 survey of 92 paid indie products had mean sales per visit of 0.99% and median 0.70%. Sales per download already includes noninstallers, so do not multiply by installation rate again.
- Estimated median Steam first-week sales ÷ wishlists was 10.5% for games with roughly 5,000+ launch wishlists in August–October 2024, versus about 20% in an opt-in survey. Sample and timing matter greatly.
- We found no public measurement of how signing/notarization warnings reduce legitimate-app installation. Do not substitute malware-warning figures.
Check the year first
These software-stage data include the dial-up and pre-stub-installer eras. They illustrate funnel structure and orders of magnitude, not current rates. We found no 2020s vendor benchmarks from Paddle, FastSpring, Setapp, Apple, or Microsoft.
Free-Product Funnel: Firefox
These are Mozilla primary data (B). A famous free browser is not a baseline for an unfamiliar indie app.
- Visit → download70%
- Download → successful install57%
- Install → daily active after 30 days49%
- Install → daily active after about 5 months42%
The 70% visit-to-download figure is from a Mozilla wiki image, on a page marked outdated. November 2008 data say 63 of 100 download starters completed installation; the source states 63%, although 79% × 79% is about 62%.
Page experiments also exist (A, arm-level samples in original tables): reducing load time 2.2 seconds increased downloads 15.4%; button wording gave 10.07% versus 9.73%; and none of 16 multivariate combinations significantly beat control.
Paid Products: The 2009 Indie Survey
Andy Brice's anonymous survey includes 92 responses, 8.1 million visits, 2.2 million downloads, and 110,000 sales (small-vendor skew, unverifiable, C). This book recalculated medians from public raw data.
| Metric | Mean | Recalculated median | Recalculated quartiles |
|---|---|---|---|
| Downloads ÷ visits | 28% | 21.4% | 12.1–38.9% |
| Sales ÷ downloads | 4.5% | 3.5% | 1.3–6.7% |
| Sales ÷ visits | 0.99% | 0.70% | 0.24–1.44% |
The author's outlier-excluded quartiles are 12.1–35.3%, 1.3–6.4%, and 0.28–1.39%. Right tails raise the means, so retain distributions when estimating sales. Mac 2.29% versus Windows 0.50% should carry little weight because the author notes a small Mac sample. Feature-limited 0.91% and time-limited 0.99% show no clear difference; no statistical test was reported.
Simulation Inputs
| Stage | Input | Evidence and conditions |
|---|---|---|
| Visit → sale | Mean 0.99%, | Fitted to Brice 2009 quantiles, C; order-of-magnitude only |
| Free product visit → download → install | 70%, 57% | Firefox 2007, B; do not transfer a famous brand directly to an unfamiliar app |
| Steam first-week sales ÷ wishlists | Median 10.5%, 25th percentile 4.8%, 10th percentile 1.9% | GameDiscoverCo estimates, C; calculator below |
| Steam lifetime refund rate | 9.5% | About 150 responses, C; applying it to first-week sales is approximate |
Do not use Docker/Cursor/Hancom Docs penetration, IE9's 95% warning figure, or malicious-page warning bypass rates as inputs.
Paid desktop-app intervals
Defaults fit the 2009 survey's sales-per-visit quantiles, giving a distribution median around 0.69%. Sales per visit already includes installation abandonment; do not multiply by another install rate.
8,000 runs. The 90% interval holds only rate uncertainty around the expected value; the actual count adds chance variation in counts. Vertical lines mark the 5th percentile, median (darker), and 95th. The last bin collects values above the 99th percentile. 2009 data; use only to check orders of magnitude.
The only trial-to-paid survey found was Soft•Letter, 2001: median 20%, 189 respondents, with an unclear “prospects” denominator. Different denominators and years mean this does not contradict Brice's 4.5%. Indie claims of 2–5% and about 5% are developer estimates.
Web Versus Desktop
The only public cohort comparison found was Bingo Card Creator. For the same product, people who signed up for the web version converted to purchase at about twice the rate of those who downloaded the desktop version, web-account customers generated about 78% of sales, and the downloadable version generated about ten times more support requests. This is a developer estimate for a low-price educational product, not a randomized comparison (D). The +25% associated with removing downloads was an initial expectation, not a result.
Paid Penetration Is Not Conversion
| Product | Figure | Why it is not conversion | Evidence |
|---|---|---|---|
| Docker | 7–10% | External estimate; mandatory large-company licensing | Sacra, D |
| Cursor | About 50%, calculated | Funding-round coverage; may include enterprise seats | The Next Web, D |
| Hancom Docs | About 5%, calculated | About 100,000 paid conversions and two million free signups at end-January 2023; brokerage NDR recap | Eugene Investment, D |
Installation Warnings
| Figure | What it measures | Evidence |
|---|---|---|
| 95% deleted after warning | Previously undetected malware flagged by IE9 SmartScreen | Threatpost, C via press |
| Warning bypass: Chrome 23%, Firefox 7% | Malicious web-page warnings | Almuhimedi et al., B, secondary citation within paper |
Neither measures abandonment for unsigned legitimate apps. Even signed new files may show warnings until reputation builds. We found no before/after measurement; use your installation logs. Evidence linking size and abandonment was found only for Android, not desktop.
Games: Steam
GameDiscoverCo is a data vendor publishing external estimates (C). Most figures are ratios of first-week sales, including buyers without wishlists, to wishlists at launch. They are not true wishlist conversion rates.
- Filter: 5,000+ wishlists and 500+ sales22.0%D
- All opt-in survey respondents20.0%C
- All games with 25,000+ wishlists15.0%C
- Broad sample, mostly 5,000+ wishlists: median10.5%C
- Same sample: 25th percentile4.8%C
- Same sample: 10th percentile1.9%C
- The 22% comes from about 700 games filtered for at least 500 first-month sales (summary, D). The 10.5% covers August–October 2024 releases with GDCo Hype above 250.
- All games with 25,000+ wishlists: 0.15×; above $10, 0.10×. The author describes a 10–20-fold outcome range.
- Following only wishlisters from a specific date gives 1–4%, four cases, D.
- Lifetime refund rate has a sales-volume-based median of 9.5%, 8% for full releases and 12.4% for early access, about 150 responses.
- First-month revenue is 1.39 times first-week revenue; first-year revenue 3.16 times (survey, revenue-based).
Watch secondary citations: some coverage calls 1.9% the “top 10%,” while the original means the 10th percentile, or bottom 10%.
Combinations Not to Multiply
| Combination | Why |
|---|---|
| Sales ÷ downloads × download → install | Sales/download already includes noninstallers |
| Steam ratio × refund rate × demo → wishlist | The ratio includes purchases without wishlists and samples differ. Subtracting refunds alone is acceptable as an approximation |
| Docker/Cursor penetration → trial-to-paid | Different denominators and mechanisms |
Missing Evidence
Within our search scope, we found no post-2015 downloadable-trial-to-paid benchmark, vendor-level desktop rates, paid-app install-to-activation, Electron/web cohort comparison, company conversion figures for JetBrains, Raycast, Obsidian, or 1Password, VS Code extension conversion, or Hancom annual-report conversion figures.
App Subscription Conversion
Review download-to-trial-to-paid conversion by category, region, and price, plus hard paywalls, AI-app retention, and Korean trial-rule effects using transaction logs.
Expert Forecasts and Calibration
Reduce optimism with reference-class forecasts and small expert ensembles, calibrate simulations with pilots, and track the gap between predictions and outcomes.