AcceleratorWatch

Our 14-Signal Prediction Model

A data-driven approach to predicting accelerator cohort selections.

AcceleratorWatch uses a proprietary 14-signal model to predict which startups will be accepted into top accelerator programs. Each signal is weighted based on historical correlation with acceptance outcomes across 340+ predictions. The model is recalibrated quarterly as new cohort data becomes available.

The 14 Signals

1

Founding Team Depth

Backgrounds, domain expertise, prior exits, and complementary skill sets. The single most predictive signal across all accelerators.

20%
Weight
100
2

Technical Differentiation

Novelty of approach, proprietary IP, architecture advantages, and defensibility of the technology stack.

15%
Weight
75
3

Customer Traction

Paying customers, revenue metrics, growth rate, and quality of customer logos.

12%
Weight
60
4

Market Timing

TAM growth trajectory, regulatory tailwinds, industry adoption curves, and macroeconomic alignment.

10%
Weight
50
5

Fundraising Velocity

Investor quality, round size relative to stage, terms, and speed of fundraising.

8%
Weight
40
6

Competitive Positioning

Market gap identification, defensibility against incumbents, and clarity of differentiation.

7%
Weight
35
7

Advisor Network

Mentor connections, industry endorsements, and quality of advisory board.

5%
Weight
25
8

Media Coverage

Press mentions, conference talks, industry award nominations, and podcast appearances.

5%
Weight
25
9

Team Growth Rate

Hiring pace, talent attraction from notable companies, and team expansion trajectory.

4%
Weight
20
10

Problem Urgency

How pressing is the problem they solve? Measured by buyer intent signals and market demand indicators.

4%
Weight
20
11

TAM Size

Total addressable market size and realistic serviceable market within 3-5 years.

3%
Weight
15
12

Demo Quality

Pitch deck clarity, product demo assessments from public events, and presentation quality.

3%
Weight
15
13

Prior Accelerator Interest

Other programs' interest signals, previous applications, and cross-program referrals.

2%
Weight
10
14

Patent/IP Filings

Filed or granted patents, trade secrets, and intellectual property moat.

2%
Weight
10

Backtesting Results

Our model has been backtested across 340+ predictions with an overall accuracy of 78%. Accuracy varies by accelerator type and program characteristics.

Program TypePredictionsAccuracy
Cybersecurity & Enterprise8882%
General Tech (YC, 500 Global)14579%
Deep Tech & Hardware5275%
Consumer-focused55+71%

Our model performs best for cybersecurity and enterprise programs (82%) and less well for consumer-focused programs (71%). We believe this is because enterprise and cybersecurity accelerators tend to weight technical differentiation and customer traction more heavily — signals that are easier to measure from public data. Consumer programs rely more on subjective assessments of market insight and viral potential, which our model captures less effectively.

Limitations

No model can perfectly predict human selection committees. Our signals are based on publicly available data and may miss internal accelerator dynamics, personal referrals, or subjective assessments that selection committees weigh. We also cannot capture late-breaking changes like pivots, team departures, or private negotiations. Acceptance probabilities represent our model's confidence, not guaranteed outcomes.

Frequently Asked Questions