Pacific Northwest startups most likely to have a successful exit, according to PitchBook
PitchBook launched a tool that forecasts the chances of venture-backed startups achieving a successful exit.
Exit activity has seen a sharp decline over the last year.
We analyzed the top 50 startups in the GeekWire 200 using the PitchBook tool.
PitchBook unveiled a new machine learning tool this week that predicts the likelihood of a VC-backed startup having a successful exit.
The VC Exit Predictor uses PitchBook’s database — which includes more than 120,000 VC-backed companies; financing rounds; and investors — to train its algorithm.
The tool calculates a startup’s “opportunity score,” a percentile that reflects the probability of a profitable return on investment. Additionally, it forecasts the company’s exit path — acquisition, going public, or remaining self-sufficient or shutting down.
We analyzed the top 50 startups in the GeekWire 200, our ranking of the top privately held Pacific Northwest tech companies, using the VC Exit Predictor. See below for the results.
Exit type percentages indicate the most probable means of exit for a startup.
“N/A” is assigned to startups that received fewer than two rounds of financing.
PitchBook makes clear the new tool is not intended to replace regular due diligence. The model does not consider factors such as comprehensive financials, business models, and founder intangibles — all metrics commonly used to evaluate startups.
Machine learning algorithms can also be slow react to macroeconomic trends or industry-specific sentiments. For instance, the VC Exit Predictor maintains a positive outlook on crypto-related companies, TechCrunch reported.
Market intelligence platform CB Insights offers a similar tool called Mosaic Score. The feature, released in 2021, faced criticism after a Tech Brew report found it was susceptible to bias. The Mosaic Score reportedly took into account a founder’s previous employers, educational background, academic achievements, and “network quality.”
PitchBook’s model does not use information related to the personal characteristics of founders, PitchBook Senior Quantitative Research Analyst Andrew Akers said during a webcast for reporters Thursday. However, he acknowledged there is “potential for bias in the output.”
Akers said PitchBook conducted a series of tests to minimize potential biases, and discovered no “statistically significant difference” in projected success rates between startups with male and female founders.
PitchBook’s new tool predicts Seattle data analytics startup MotherDuck is in the 96th percentile on expected return on investment.
VC Exit Predictor’s release coincides with a significant decline in startup exit activity brought on by rising interest rates, geopolitical tensions and waning investor outlook on the tech industry.
Total exit deal value in the U.S. was $71.4 billion in 2022, marking a decline of more than 90% from the previous year, PitchBook reported. That’s the first time since 2016 that annual exit value didn’t surpassed $100 billion, per the report.
No companies from Washington state went public via IPO last year, while two — cannabis platform Leafly and photo giant Getty Images — went public via a special-purpose acquisition company (SPAC).
The lethargic exit climate could continue to drag down the VC sector. A lag in exits could lead to longer hold periods, reduced returns, decreased distributions to limited partners, and slower fundraising, PitchBook said.
“To mitigate these affects and achieve liquidity, VC investors need to understand the quality of their assets and evaluate their appetite for taking on more risk,” the company said. “Meanwhile, startups need to showcase their path to profitability and justify their valuations to maximize shareholder value.”
In order to get an exit prediction, companies must’ve received at least two rounds of venture financing. PitchBook’s model is based on 46,000 observations from startups with known exit outcomes. The tool was tested on more than 11,000 observations that were not included in the model’s training, and it demonstrated a 75% accuracy rate in its outcomes.