72%
Overall accuracy
45
Polls completed
18
Programs covered
4,800+
Total votes cast
Since our first poll in December 2024, the AcceleratorAlumni community has conducted 45 structured prediction polls across 18 accelerator programs. Our overall accuracy — the percentage of polls where our top-ranked prediction was actually selected for the cohort — stands at 72%. Here is the full breakdown.
Accuracy by Program
| Program | Polls | Correct | Accuracy |
|---|---|---|---|
| ICON Spark | 4 | 3 | 75% |
| Y Combinator | 6 | 5 | 83% |
| Techstars (all) | 8 | 6 | 75% |
| 500 Global | 5 | 3 | 60% |
| Alchemist | 4 | 3 | 75% |
| Antler | 3 | 2 | 67% |
| Plug and Play | 4 | 3 | 75% |
| MassChallenge | 3 | 2 | 67% |
| Others (combined) | 8 | 5 | 63% |
Our highest accuracy is for Y Combinator (83%), which benefits from the largest respondent pool and the most available public signal data. Our lowest accuracy is for 500 Global (60%), where the international scope and larger cohort sizes make individual predictions harder.
Accuracy by Confidence Score
Our confidence weighting system is designed to identify higher- conviction predictions. Here is how accuracy correlates with the confidence score assigned to our top prediction.
| Confidence Range | Polls | Correct | Accuracy |
|---|---|---|---|
| 0.85+ (Very High) | 5 | 4 | 80% |
| 0.70-0.84 (High) | 12 | 9 | 75% |
| 0.55-0.69 (Moderate) | 18 | 13 | 72% |
| Below 0.55 (Low) | 10 | 6 | 60% |
The correlation is clear: higher-confidence predictions perform better. When our community converges with high conviction (0.85+ confidence), we have been correct 80% of the time. Low-conviction predictions (below 0.55) are essentially coin flips.
Notable Successes
YC W25 — NeuralFlow
Our community predicted NeuralFlow as the top YC W25 selection with 71% vote share and 0.82 confidence. NeuralFlow was indeed selected and went on to raise a $15M Series A post-demo day. This was our highest-profile correct prediction.
ICON Spark 2025 — ShieldAI
Alumni correctly predicted ShieldAI for ICON Spark 2025 with 68% vote share. Our ICON Spark alumni provided key insight into the committee's preference for teams with government security backgrounds, which proved decisive.
Techstars NYC 2025 — DataVault
We predicted DataVault for Techstars NYC 2025 with moderate confidence (0.61). Despite the lower confidence, the prediction was correct. DataVault went on to be one of the strongest performers in the cohort.
Honest Misses
We believe in full transparency. Here are predictions where our community got it wrong and what we learned.
500 Global Batch 33 — Missed HealthSync
We predicted PayFlow for 500 Global Batch 33, but the program selected HealthSync instead. Our community underweighted 500 Global's increasing focus on healthtech. Lesson: program priorities shift, and our alumni base may lag in recognizing those shifts for programs with less community representation.
Antler Singapore 2025 — Wrong Top Pick
Our top prediction for Antler Singapore 2025 was rejected. The issue: limited Antler alumni in our community meant lower confidence (0.48) and less program-specific insight. We have since expanded Antler alumni recruitment to improve future predictions for the program.
ICON Spark 2025 — Missed QuantumShield
While we correctly predicted ShieldAI as a top selection, we missed QuantumShield entirely. The company had weak public signals but strong internal referrals that our community could not observe. This represents a fundamental limitation of any prediction methodology based on public information.
Techstars Boulder 2025 — Overconfident Pick
We predicted AgriTech Pro for Techstars Boulder 2025 with relatively high confidence (0.72), but the company was not selected. Post-mortem analysis revealed the founding team had a prior relationship with a Techstars mentor who advocated against selection — an internal dynamic we could not have known about.
Improvement Over Time
Our accuracy has improved as the community has grown. Early polls (December 2024 through June 2025) averaged 65% accuracy with smaller sample sizes. Recent polls (July 2025 onward) average 76% accuracy, reflecting a larger, more diverse respondent pool and refined weighting methodology.
| Period | Polls | Avg Respondents | Accuracy |
|---|---|---|---|
| Dec 2024 — Jun 2025 | 18 | 62 | 65% |
| Jul 2025 — May 2026 | 27 | 148 | 76% |
The 11-point accuracy improvement correlates with both community growth (more respondents per poll) and methodology refinements (introduction of confidence weighting in v2 of our polling system). We expect continued improvement as the community grows and our weighting model accumulates more backtesting data.
Methodology note: Accuracy is measured as the percentage of polls where our top-ranked prediction (the startup with the highest weighted vote share) was confirmed as selected for the actual cohort. For programs that select multiple companies per cohort, a prediction is “correct” if our top pick was among those selected. Results are self-tracked and based on publicly announced cohort lists. Last updated May 10, 2026.