Predicting Heavy Civil Jobs Using AI
About five years ago, we launched Thunderbolt Predict at our startup, and it still stands out as one of the most ambitious things we built. The model was trained on a large volume of publicly available DOT bidding data and refined through months of careful feature engineering and rigorous backtesting.
On paper, it did exactly what we hoped. It consistently identified bid strategies that could improve win rates for DOT paving contractors. In practice, however, the response surprised us. We did not get a single customer willing to adopt it. AI just wasn’t as “sexy” as it is now…
At the time, the model operated as a true black box. For many estimators, the idea that a system could outperform experienced competitors without hands-on estimator input felt unsettling rather than exciting. Trust was the real hurdle, not accuracy.
We even integrated the model directly into a beta customer’s B2W on-prem database with minimal effort, simply by adding a link that exposed the predictions inside their existing workflow. Technically, it worked exactly as designed.
Looking back, it feels like a product that arrived too early. With today’s expectations around AI, explainability, and estimator-assisted workflows, I often wonder how differently Thunderbolt Predict would be received if we were building it right now!
Want to learn more? Are you a losing jobs and want an AI advantage? Curious what part of your workflow could be automated next? We’re building tools like this every day at AutomateBetter.ai … reach out and let’s chat about how we can automate some of your most time-consuming processes … or how we can help you win more bids with greater margins!