How AI Speeds up Profitable Bids
There’s a moment in every estimator’s week that Deepti Yenireddy can describe precisely: the deadline is closing in, there are fourteen bids due, and the team is working evenings and weekends just to keep up with the volume.
That moment is exactly what she built Boon AI to solve.
Deepti is the founder and CEO of Boon, an AI platform for pre-construction that handles everything from early-stage takeoffs to bid leveling to conflict identification. Before that, she led product at Samsara, where construction was one of the largest customer segments. The thread connecting her career has always been the same fascination — the challenge of applying AI not to software problems, but to physical world problems, which she’ll be the first to tell you are significantly harder.
She joined this episode of Builders Budgets and Beers to talk about what’s actually happening at the frontier of AI estimating, why the labor math in construction makes AI adoption an urgent priority rather than a nice-to-have, and what advice she’d give to estimators who’ve tried generic AI tools and walked away unimpressed.
Why construction is one of the hardest problems for AI to solve
Most people’s mental model of AI is something like ChatGPT — a system trained on an enormous amount of text data that can reason and respond across a wide range of topics. That model works well for software problems, where data is abundant, structured, and easily captured.
Construction is different. The data problem alone is significant. There’s no vast, organized global library of construction plan sets. Architects design differently. Symbols vary. Building codes differ by county, state, and project type. And the spatial intelligence required to understand a three-dimensional structure from a two-dimensional drawing is a fundamentally harder problem than predicting the next word in a sentence.
“Within the world of construction and spatial intelligence with buildings, there’s limited data,” Deepti said. “We haven’t always been structured in designing buildings, capturing that data, making it available.”
To work around the limits of real-world training data, Boon has invested heavily in generating synthetic data — essentially creating realistic, code-compliant fictional plan sets to train its vision models on at a scale that the real world can’t provide. The goal is to teach the AI to recognize components, understand spatial relationships, and apply those learnings accurately enough that an estimator can trust the output.
It’s an enormous technical undertaking, and Deepti is candid that it’s ongoing work. But it’s also why vertical AI — AI built specifically for construction — produces meaningfully better results than pointing a general-purpose model at a plan set and hoping for the best.
The real pain driving AI adoption in estimating
When Deepti talks to customers, the conversations don’t start with technology. They start with pain.
The most consistent pain in construction estimating right now is a collision between two forces: a severe shortage of experienced estimators and an unprecedented surge in demand driven by the data center boom. The top five hyperscalers are spending half a trillion dollars annually. Projects need to be priced fast, with accuracy, by people who are already stretched thin.
“We came across customers who wanted to double their revenue in a year,” Deepti said. “When does that happen in construction?”
The bottleneck isn’t ambition. It’s capacity. Estimating is skilled, time-intensive work, and you can’t just hire your way out of the shortage — because the people who can do it well aren’t readily available. The construction industry is facing a version of the same math that’s driving automation across every skilled-labor industry: demand is growing faster than the workforce can.
What Boon demonstrated to those customers was a before-and-after that reframed the conversation entirely. Something that takes two days to complete manually can be finished in two minutes with AI. Even if the estimator spends fifteen minutes reviewing the output, the net time savings is transformative — and more importantly, it creates the capacity to take on work that would otherwise be impossible to pursue.
What an AI estimator looks like in practice
The most vivid illustration from the conversation is worth dwelling on.
A Boon customer — an electrical estimator — was at home on a Friday evening with his family. He had a project he needed to review. Using Boone’s agentic AI, he opened Slack, started a conversation with the AI agent (which Boone’s team calls Gandalf, a nod to Deepti’s affinity for the character), and walked through a series of requests: review the project, pull the specs, identify conflicts, start measuring feeders.
Over the course of three to four hours — while sitting on his couch — he worked through an entire project with the AI doing the heavy lifting on measurement and analysis.
“He did that all on a Friday evening on his couch while he was with his family,” Deepti said.
This is what the shift from tool to agent looks like in practice. Not a software feature that saves a few clicks. A system that takes instruction, executes across multiple steps, and returns meaningful work product that the human can review and refine. The estimator’s job became less about measuring and more about directing — which is, notably, a better use of a skilled estimator’s time.
The accuracy objection — and why it’s missing the point
The most common pushback Deepti hears from estimators who haven’t adopted AI is straightforward: the tools they tried weren’t accurate enough to trust.
It’s a legitimate concern, and she doesn’t dismiss it. General-purpose AI tools — the large language models that most people interact with as chat interfaces — don’t have the construction-specific context required to produce reliable takeoffs. They hallucinate. They misidentify components. They don’t understand building codes or the spatial logic of a plan set. The skepticism those experiences generate is real and earned.
But there are two things that get lost in that conversation.
The first is the difference between horizontal and vertical AI. A general-purpose model trained on everything is not the same as a model trained specifically on construction drawings, plan sets, and estimating data. The accuracy problem that makes general tools unreliable is largely a data problem — and vertical AI addresses it directly by training on domain-specific data and building in the business logic that generic tools lack. The same principle applies in construction finance: a general tool that can read a PDF is not the same as a system that understands the relationship between a cost code, a job, a change order, and a billing cycle.
The second thing that gets lost is how quickly AI is improving. The timeline from “not quite accurate enough to trust” to “accurate enough to use in production” has compressed dramatically. Capabilities that weren’t viable six months ago are viable today. Waiting out the technology based on a bad experience from a year ago means waiting out one of the fastest-moving periods of improvement in the history of software.
“You can’t say I tried it six months ago and it didn’t work,” Reece said, “because this technology is getting better so fast.”
AI won’t replace estimators. It will make the best ones irreplaceable.
The question of whether AI will eliminate construction jobs comes up in almost every conversation about the technology, and Deepti has a considered answer.
Her view is grounded in a longer historical arc. Every major technological shift — industrial revolution, digital revolution, AI — has made humans more sophisticated, not less necessary. What changes is what humans are doing. The industrial revolution didn’t eliminate labor; it changed what labor looked like. The same pattern is playing out now.
In estimating specifically, the shift is from execution to strategy. The estimator who spends most of their time measuring beams and columns gets replaced — not by AI eliminating their job, but by AI handling the measurement so the estimator can focus on engineering questions, risk identification, and project strategy. The skill set evolves. The value of deep expertise doesn’t diminish; it gets applied at a higher level.
“Estimation is going to be very hard for AI to make redundant,” Deepti said. “Not now, not in many, many years.”
What’s also worth naming is the labor math on the other side of the equation. Construction is already facing a workforce shortage. The tradespeople, journeymen, and skilled craftspeople who actually build the things that get estimated aren’t being replaced by AI anytime soon — the physical world problems Deepti described at the start of the conversation haven’t been solved, and they won’t be for a long time. If AI makes estimating faster and construction companies more competitive, the demand for people who can actually do the work only increases.
The advice for estimators not yet using AI
For estimators who are watching from the sidelines — aware that AI is changing the industry but not yet sure how to engage with it — Deepti’s advice is practical and direct.
Don’t start with a generic tool. The hallucination problem that’s feeding skepticism is largely a function of using horizontal AI for a vertical problem. Find tools built specifically for construction estimating, where the training data and business logic are designed for the work you’re actually doing.
Invest a small amount of time to try it yourself. Thirty minutes of hands-on testing will tell you more than any amount of reading about it. Run a small takeoff through it. Review the output. See where it’s accurate and where it needs correction. That’s how trust gets built — not through a demo, but through doing.
And treat accuracy as a starting point, not a ceiling. The question isn’t whether AI is already perfect. It’s whether it can get you to a better outcome — in less time, with more capacity — than your current process. Two days to two minutes, with fifteen minutes of review, is a better outcome even if the AI isn’t perfect.
“The key thing to do,” Deepti said, “is try a few specialized systems and build trust in them.”
Keep the pace of learning
At the end of the conversation, Reece asked Deepti for the one piece of advice she’d leave with listeners. She took a moment, then gave an answer that had less to do with AI specifically and more to do with how to stay oriented in a fast-moving environment.
The only constant, she said, is your pace of learning. Technology is changing faster than anyone can reliably predict — her own perspective on AI’s trajectory shifts week to week as new capabilities emerge. What she can control, and what anyone can control, is how quickly they’re absorbing new information, testing new tools, and updating their understanding of what’s possible.
For estimators, for contractors, for anyone in the construction industry watching AI accelerate and wondering where it leaves them — the answer is probably not to have a settled position on it. The answer is to keep learning fast enough to stay current with what the tools can actually do.
Deepti Yenireddy is the founder and CEO ofBoon AI, an AI platform for pre-construction estimating and takeoffs. This episode of Builders Budgets and Beers is available wherever you listen to podcasts.