When we released our bracket building tool on Selection Sunday, our users got after it! In just three days, more than 28,000 brackets were completed. Half of them were manually completed by users and half were done using the WP autofill feature.
This gives us a huge sample of over one million picks to analyze, and we can see how human picks compared to WP autofill for every single game!
Humans Lead the Race 🧠
Through two rounds, manual human picks beat out the WP autofill.
Manual Picks: 73.5% correct
WP Autofill: 68% correct
This is about a 5.5% edge for the humans (this was not the case last season, our users are getting sharper).
WP Autofill Is Better at Calling Upsets 🔮
We defined an upset here as anytime a team two or more seeds below another wins.
Number of upsets predicted:
Manual Picks: 27.3%
WP Autofill: 29.7%
WP Autofill called 5 of the 7 upsets at higher rates than humans:
Upset | WP Autofill | Humans |
|---|---|---|
Texas over (6) BYU | 40.8% | 17.7% |
Tennessee over (3) Virginia | 38.8% | 22.8% |
Iowa over (1) Florida | 18.7% | 2.2% |
High Point over (5) Wisconsin | 17.9% | 12.0% |
Texas over (3) Gonzaga | 13.6% | 7.2% |
The Iowa (9) over Florida (1) upset really stands out here.
Nearly 1 in 5 WP Autofill picks had Iowa winning. Only 1 in 50 human picks did. When a No. 1 seed is on the line, humans trust the ranking - WP Autofill looks at the matchup.
Humans called 2 upsets better:
Upset | Humans | WP Autofill |
|---|---|---|
VCU over (6) North Carolina | 65.7% | 37.6% |
Texas A&M over (7) Saint Mary's | 63.2% | 40.7% |
Games That Stumped Everyone
The games that fewest brackets (manual or WP Autofill) got right:
Texas over Gonzaga (11 over 3) - only 10.4% of all brackets predicted this
Iowa over Florida (9 over 1) - only 10.4% predicted this
High Point over Wisconsin (12 over 5) - only 14.9% predicted this
Texas over BYU (11 over 6) - only 29.1% predicted this
Nebraska over Vanderbilt (4 over 5) - only 29.6% predicted this
Best Brackets So Far 🙌
Brackets are scored using ESPN tournament scoring: 10 points per correct Round 1 pick, 20 points per correct Round 2 pick, doubling each round through the championship (max 1,920 points for a perfect bracket).
Through 2 rounds (max 640 points):
3 brackets are tied for first at 570 points - 2 are fully WP Autofill, 1 is fully human
At 560 points: 12 brackets - 5 WP Autofill, 7 human
At 550 points: 32 brackets - 12 WP Autofill, 20 human
WP Autofill holds 2 of the 3 best brackets despite accounting for fewer than half of all picks. Humans dominate the next tiers and score higher on average, but WP Autofill has a clear edge at the top.
Part of the reason for this is that the WP autofill is usually optimizing for unique (larger) pool settings, and thus making riskier picks. A bracket that is more correct does you no good if it has no edge in a large winner-take-all pool.
Just look at the differences in Champion/Final Four picks. The difference in approach is very clear:
Champion picks:
Team | Overall | Human | WP Autofill |
|---|---|---|---|
Arizona | 28.4% | 25.1% | 3.3% |
Duke | 14.6% | 10.9% | 3.6% |
Michigan | 9.4% | 5.7% | 3.7% |
Florida | 4.9% | 1.9% | 2.9% |
Houston | 4.8% | 2.0% | 2.7% |
Final Four picks:
Team | Overall | Human | WP Autofill |
|---|---|---|---|
Arizona | 51.2% | 45.0% | 6.2% |
Duke | 38.9% | 32.1% | 6.8% |
Michigan | 35.5% | 28.6% | 6.9% |
Florida | 35.0% | 29.4% | 5.6% |
Michigan State | 24.5% | 21.0% | 3.5% |
