Accuracy benchmarks are computed from consenting Procore connections. Data will appear here once enough projects have been analyzed.
Aggregate statistics will appear here once enough consenting partners have contributed data.
Accuracy varies by project scale. Buckets with fewer than 5 projects are hidden.
Not enough data in each size bucket yet.
Numbers come from contractors who have connected their Procore accounts and opted in to anonymous benchmarking. Only closed projects with actual cost data are included — no in-progress projects, no estimates without actuals.
Before comparing against actual costs, ProBid AI generates a shadow estimate using only the project metadata that would have been available at bid time: trade type, location, scope, and project size. The actual cost is hidden from the AI during this step.
Error % = |ProBid Estimate − Actual Cost| ÷ Actual Cost × 100. This is an absolute (non-directional) percentage — it measures how far off the estimate was in either direction.
We report P50 (median) and P80 error. Medians are more robust than means for skewed data. A lower number means tighter, more accurate estimates.
Any breakdown category with fewer than 5 projects is hidden rather than shown. Small samples produce unreliable statistics. We show "Insufficient data" rather than mislead.
No project names, company names, or individual line items are ever shown publicly. Only aggregate statistics are included. Participation is opt-in; contractors can change their preference any time in their Procore settings.
Procore-connected accounts (read-only OAuth). Actual costs pulled from Procore budget views after project close.
Benchmarks are recomputed daily from all consenting connections. The last update timestamp is shown above.
Business plan subscribers can opt in to anonymous benchmarking in their Procore settings. Your data improves the benchmarks for everyone.
Connect Procore →Connect your Procore account and we'll run ProBid's estimates against your own closed projects. You get a private accuracy report — and can choose whether to contribute anonymized stats to this page.
Run Your Own Accuracy Test