Build contingency from the risks on this job, not from habit. List the specific things that could cost money, estimate the cost and likelihood of each, add a smaller amount for the unknown, and adjust for how complete the design is. Then track it as its own budget line and release it as risks pass.
Plan reviews connect field conditions, drawings, and the next decisions on a job. AI-generated illustration.
To estimate contingency, start with a list of specific risks on this project: unknown soil, an existing building with hidden conditions, a tight schedule, a volatile material, a sub you have never used. For each, estimate what it would cost if it happens and how likely it is. Multiply, add them up, and add a modest amount for the risks nobody has thought of. The result is a contingency you can explain line by line, which matters when an owner, a lender, or your own project manager asks why the number is what it is. A flat percentage applied to every job is either too much on simple work or not enough on risky work.
Types of contingency
Contractor's contingency: money inside the contractor's price for its own risks, such as estimating errors, productivity, buyout, and scope gaps. On a hard bid, it is part of your price and not shown separately.
Owner's contingency: money the owner holds for owner-side changes, design changes, and unforeseen conditions. Not part of the contractor's price.
Design contingency: money an estimator adds when drawings are incomplete, to cover detail that will appear as the design develops.
Contingency on GMP and cost-plus jobs: often a defined line in the contract, with rules about who can use it and what happens to the unused balance. On a GMP contract, read those rules carefully.
Contingency is not the same as an allowance. An allowance is a placeholder for a known item whose details are not chosen yet, like tile at $8 per square foot. Contingency covers things that may or may not happen.
Federal cost rules draw a similar line. FAR 31.205-7 defines a contingency as a possible future event or condition arising from presently known or unknown causes, the outcome of which cannot yet be determined. For estimates of future costs, it separates contingencies from known conditions whose effect can be foreseen within reasonable limits, which are included in the estimate, from those that cannot be measured precisely enough to be equitable, which are disclosed separately instead of built into individual cost items. That is the same idea as a risk-based contingency: price what you can, and be open about the rest.
Step by step: building a risk-based contingency
1. List the risks
Walk through the estimate with the people who will build it: estimator, project manager, superintendent. Ask what could cost money. Typical sources: existing conditions, soils and groundwater, weather, schedule pressure, material prices, new or thinly capitalized subs, incomplete design, permits and inspections, and owner decision speed. Record each in a simple risk register.
2. Remove what belongs somewhere else
Some risks should be handled another way: excluded in your bid clarifications, priced as an alternate, shifted by contract (an escalation clause, a unit price for rock), or transferred to a sub. Contingency is for the risk you keep.
3. Estimate cost impact and likelihood
For each remaining risk, estimate a realistic cost if it happens and a probability. Use your own job history where you can. Rough bands (10, 25, 50, 75 percent) are fine; false precision does not help.
4. Calculate expected value and add an unknowns allowance
Multiply cost by probability for each risk and add them. Then add a small general amount for what you have not identified. Keep it modest and say what it is for.
5. Adjust for design stage and job type
A budget from schematic drawings needs more design contingency than a bid on complete construction documents. Renovation of an occupied building needs more than new construction on a clean site. Write down why.
6. Decide how to carry it
On a hard bid, contingency usually sits inside the price, and competition limits how much you can carry. On negotiated, GMP, or cost-plus work, it is often a visible line. Either way, put it in its own cost code in the job budget so it is not quietly spent covering overruns.
Worked example: renovating a 1960s school wing
Say a general contractor is pricing a $2.6 million renovation of an older school building over a summer break. The risks and numbers are illustrative.
Example: risk-based contingency
Risk
Cost if it happens
Probability
Expected value
Hidden deteriorated framing behind plaster
$60,000
40%
$24,000
Unknown utilities in slab during trenching
$35,000
30%
$10,500
Overtime to finish before school opens
$48,000
50%
$24,000
Electrical sub default (new sub)
$90,000
10%
$9,000
Estimating and buyout gaps
$40,000
50%
$20,000
Unidentified risks allowance
$15,000
Total contingency
$102,500
That is about 3.9 percent of the job, built from reasons rather than habit. Hazardous materials abatement was left out on purpose: the contractor excluded it in the bid because the owner's survey was incomplete. If the owner pushes back on price, the contractor can see exactly which risks it is choosing to carry without coverage, rather than cutting a round number.
Now test it against the market. If similar school renovations in the area are going to bidders at tight margins, the contractor may not be able to carry the full $102,500 and still win. The choice then is explicit: carry less and accept more risk, push more items into clarifications, or walk away. What it should not do is shave the contingency without knowing which risk it just left uncovered. A short note in the bid file recording that decision helps the project team later, because they will know which risks the price does not cover.
Managing contingency during the job
Track contingency as its own budget line and log every draw against it with a reason.
Review it monthly with the cost to complete: as risks pass (framing exposed, slab trenched), the matching contingency can be released to projected profit.
Do not use contingency to cover owner changes. Those are change orders.
Watch the burn rate. A job that has used 80 percent of its contingency at 40 percent complete needs attention now.
Mistakes to avoid
One percentage for every job. It hides risk on hard jobs and costs you bids on easy ones.
Counting risk twice, once in inflated unit prices and again in contingency.
Using contingency to absorb owner changes instead of writing change orders.
Spending it silently. Without a log, nobody learns which risks actually happen.
Never releasing it. Unused contingency held to the end understates profit on the WIP report and can distort billing.
What percentage should construction contingency be?+
There is no single right percentage. It depends on design completeness, job type, existing conditions, and contract type. Build it from the specific risks, then compare the result with your history on similar jobs.
What is the difference between contingency and allowance?+
An allowance is a set amount for a known item whose details are not yet chosen. Contingency covers events that may or may not happen.
Who gets unused contingency?+
On a lump sum job, unused contractor contingency becomes profit. On GMP and cost-plus jobs, the contract decides, and savings are often returned to the owner or shared.
Should contingency be shown to the owner?+
On hard bids it is usually built into the price. On negotiated, GMP, or open-book work it is commonly shown as a line, with rules for its use.
Maurice Francois writes and edits construction operations guides for Smart Business Automator. More from Maurice on the Scaling Legends podcast.
Related listening · Scaling Legends
Y Combinator's 2026 Construction Cohort: How the AI-Agent Startups Betting on 'Construction Intelligence' Could Reshape How You Estimate, Schedule, and Bid
Y Combinator's 2026 real estate and construction cohort is betting big on AI agents and construction intelligence — and MarketScale reports 38% of contractors now claim measurable impact from AI. With Autodesk and Silicon Valley Bank both flagging an autonomous tipping point, here's which categories matter for contractors and which are hype.
Starts muted when the player comes into view.
Read the full episode transcript
Published transcript from Scaling Legends. It may contain transcription errors. View the source transcript.
As always, this content is for educational and informational purposes, only it is not legal, financial, or professional advice. We're currently accepting beta testers for Smart Business Automator. Sign up at smartbusinessautomator .com. Beta testers get first access to new features and updates before public rollout. So if your livelihood depends on moving dirt, pouring concrete or, you know, putting up steel, your back office is about to be fundamentally disrupted. Oh, absolutely disrupted. It's not even a question anymore. Right. Because Y Combinator, the Silicon Valley accelerator behind massive tech juggernauts like Airbnb and Stripe, they've officially unveiled their 2026 construction cohort. Yeah. And they aren't just dipping their toes in. They are pouring hundreds of millions of dollars. into back office
automation. Hundreds of millions. Yeah. And so the goal of this deep dive today is to explore exactly how these new AI agents are going to completely reshape how you estimate, how you schedule, and how you bid. Exactly. And we're going to give you a highly actionable step -by -step plan for what you can do, like this week, to capitalize on this wave. Yeah, because when Y Combinator points its capital and its top engineering talent at a specific sector, that sector transforms, period. Silicon Valley is officially betting the bank on what they call construction intelligence. And we aren't talking about consumer apps or just some minor software updates. No, not at all. This is a structural shift in how a construction business actually operates. To really understand
the scale of this, we just got to look at the numbers. According to Smart Business Automator's Market Intelligence Report, drawing on recent findings from MarketScale, 38 % of contractors are already reporting measurable AI impact in their daily operations. Wait, almost 40 %? Yeah. I mean, that crosses the line from early adoption right into mainstream integration. Yeah. We're way past the theoretical phase here. Oh, we're definitely past the theory. Institutional heavyweights like Autodesk and Silicon Valley Bank, they're going on the record calling this an autonomous tipping point. Yeah. Yeah, they aren't treating this as some temporary trend. They're signaling that the industry is permanently tipping over into autonomous workflows. OK, let's unpack this. An autonomous tipping point. I hear that phrase
and I immediately think about, well, the history of the job site itself. Sure. Is this transition like, say, when power tools finally replaced hand tools? That's a really good way to look at it. Because when that happened. It wasn't just a minor upgrade, right? It wasn't just, oh, my arm is less tired. It completely changed the speed, the scale, and the fundamental economics of the job. Exactly. You could build things you just couldn't build before. Right. Faster. And with drastically fewer people turning screws by hand. But I do want to push you on this a bit. What exactly is Y Combinator funding to cause this specific tipping point? Because when people hear AI in construction, I think they still picture those Boston Dynamics robot dogs walking around a job site with a clipboard. Right, yeah. You're picturing the robot dog because
that's what gets clicks on social media. But Y Combinator isn't really looking at hardware right now. They are funding the invisible heavy lifting. They're targeting the estimating desk, the scheduling matrix, the project management workflows. Because think about it, the job site itself has actually become highly optimized over the last 20 years. Yeah, that's true. We have laser levels, GPS guided, earth moving equipment, incredibly advanced materials. But the back office of a $50 million commercial contractor, it often still looks exactly like it did in 1998. Oh, completely. It's just spreadsheets, highlighters, massive PDF plan sets, and project managers working 70 -hour weeks just to keep track of endless email threats. They're completely drowning in
unstructured data. And that's exactly it. YC recognized that administrative friction is the true bottleneck of the entire construction industry. So they are funding AI agents designed specifically to process that data and eliminate that friction. Here's where it gets really interesting, because these agents actually DDo the work. The barriers to entry for software companies just plummeted, right? Oh, absolutely. Suddenly, anyone can slap an open AI wrapper on a generic tool. and call it a construction agent. And that creates a massive new problem for contractors. A huge problem. Because contractors are exhausted by software that gives them more homework. We all have these dashboards that promise to solve everything, but they just end up being another tab we leave open and ignore. So if I already ignore my expensive
dashboards, why won't I just ignore an AI agent? Well, we have to break down the fundamental mechanical difference between an AI agent and a dashboard. A dashboard is passive. It just shows you data. It's essentially a mirror reflecting your project back to you. It says here are your RFI statuses, meaning the log of all your outstanding requests for information from the architect. And here is your schedule variance. Now, human, you figure out what to do about it. So it's still on me. Exactly. It relies entirely on your input, your time, and your cognitive load to function. Like you said, it assigns you homework. Here are all the problems you need to fix today. Good luck. Yeah, exactly. But an AI agent is active. It interacts with the data directly. Let's look at how this changes the estimating process. Okay.
An estimator doesn't want a dashboard showing them how many pages are in a spec book. They want someone or something to actually pull the quantities. Right. A traditional software tool might let you hit CTRIAL -LF to search a PDF for the word door, but an AI agent uses large language models, or LLMs. This means it actually reads the context of a 500 -page architectural plan set. So instead of just searching for a keyword, it has semantic understanding. Exactly. It recognizes that a door schedule on page 4 connects to a specific fire rating specification buried all the way over on page 400. Yes. The agent can ingest that entire plan That's wild.
So when the human estimator arrives with their morning coffee, the agent has already completed the takeoff. The tedious part. Right. The incredibly tedious process of measuring out all materials needed from the plans. The human is now editing, verifying and applying their strategic expertise rather than spending 14 hours manually clicking the corners of a PDF with a digital measuring tool. So the agent is essentially an incredibly fast. perfectly caffeinated junior estimator who doesn't sleep and never gets distracted. Pretty much. Or take scheduling. A traditional system gives you a gaunt chart. You know, those cascading bar charts showing project timelines. Yeah, everybody loves a good gaunt chart. Right. And it highlights a delay in red. But an agent actively flags contextual conflicts. It says,
hey, I noticed the drywall delivery is scheduled for Tuesday, but the electrical rough -in inspections aren't cleared yet. And historically, this specific municipal inspector takes three days to come out. Oh, wow. That's real foresight. Yeah. And we see this level of proactive problem solving heavily featured in the automation protocols of Smart Business Automator. It's a prime example of how this technology is engineered to completely eliminate back office friction rather than just adding another layer of management for the contractor to deal with. Doing versus showing. It's a massive paradigm shift. It really is. But... Because these LLMs make it so easy to build tools that do things, the market is about to be flooded. I mean, every SaaS company is going to stick the word AI on their website tomorrow. Oh, they
already are. Right. And the contractor's inbox is going to be slammed with aggressive sales pitches. So what does this all mean? How do we build a filter to separate the real revenue generating tools from the useless toys? That is the million dollar question. Because protecting margins is everything right now. The cost of capital is high. Materials are still fluctuating. And labor is incredibly tight. You just cannot pay your crews with a cool software demo. No, you can't. So you build a ruthless evaluation process we call the contractor's filter. Okay, the contractor's filter. I advise everyone listening to evaluate any new AI tool based on two specific revenue -centric metrics. Number one. Number one, does it improve your bid -win rate? And number two, does it help you capture change orders? If a
piece of software cannot draw a straight, undeniable line to one of those two outcomes, you skip it. Skip it completely. Completely. Well, bid -win rate makes complete sense. The bidding process is a numbers game, but it's also a precision game. If an agent is doing the heavy lifting on the quantity takeoffs, my estimators aren't fatigued. They can bid 10 jobs a week instead of three. And because the AI doesn't get tired at 4 p .m. on a Friday, the precision goes up. It's not missing that buried clause requiring a specific expensive type of fire caulking. Yes. So your volume goes up and your bids are more accurate, which directly increases your win rate while protecting you from underbidding a job and just losing your shirt. And it also drastically impacts the bid leveling process. Oh, right. Where you are painstakingly lining up subcontractor
quotes apples to apples to make sure no one missed a scope of work. That takes forever. A human estimator can spend days leveling bids. An AI agent can instantly cross -reference five different electrical bids against the master spec book and flag exactly which subcontractor missed the low voltage wiring. That's huge. OK, let's talk about change orders, because to me, that's the silent killer. Oh, yeah. Contractors bleed out on uncaptured scope creep. But how does an AI agent actually catch a moving wall on a job site if it doesn't have physical eyes on the project? Well, let's walk through a real -world scenario. Imagine you're doing a commercial office build -out. The HVAC contractor is installing ductwork, but they hit an unexpected plumbing line that wasn't on the as -built drawings. Happens all
the time. Right. The project manager is exhausted walking the site, and the HVAC foreman says, hey, we need to reroute this ductwork around the plumbing. It's going to take an extra day of labor and some custom fabrication. The PM, under pressure to keep the schedule moving, just says, just get it done so we don't hold up. the framing. Yeah. And then the guys do it, but the PM has a hundred other fires to put out. So the formal paperwork never gets filed. Exactly. Boom. You just did $10 ,000 of free unbilled work. Ouch. And this is where the data from Smart Business Automator's market intelligence report becomes critical. How so? The company seeing real bottom line ROI from AI are using it to catch exactly this type of scope creep. The AI agent monitors the daily site logs, the field notes, and even
email communications. It reads the PM's brief daily log entry, which might just say, HVAC rerouted duct work due to plumbing conflict in Sector B. The AI agent uses its semantic understanding to instantly cross -reference that daily input against the original contracted scope of work. So it connects the dots. It realizes re -routed ductwork and plumbing conflict were not in the original contract that it ingested. It flags it immediately. It sends an alert to the PM saying, anomaly detected in sector B labor and materials expended outside of contracted scope. That is incorrect. It practically drafts the change order for you, attaches the relevant log entry, and queues it up for review. Now that is... a tool that touches revenue. Yeah, that pays for itself. Exactly. If it's just a cool 3D demo where you
can wear a VR headset and walk through a digital lobby, but it doesn't touch the estimating desk or the billing desk, it's a toy. And construction is not a game. So we have this signal. YC is pouring money into back office automation. We have the concept, LLM -powered agents that read context and do work versus passive dashboards that just give you homework. And we have the filter, focusing ruthlessly on bid -win rates and change orders. Now we really need the action plan. We promised the listeners something they could do this week. Yes, we did. How does a contractor actually test one of these agents without blowing up their entire company? By running a 30 -day pilot. But not just any pilot. There is a golden rule for adopting construction intelligence. Which is? You must never bet a live, active job
on unproven software. Yeah, that makes total sense. I mean, you wouldn't test a brand new saw blade on a piece of premium custom milled finished carpentry. Definitely not. You don't take that risk with the good stuff. Yeah. You test it on scrap wood first to see how it cuts. to see if it splinters the grain. You need to know exactly how the tool behaves before you let it touch your livelihood. I love that analogy. So what is the scrap wood in this scenario? Right. It is your past project data. The exact strategy you should implement this week is a shadow run. A shadow run. You find an AI agent that passes the revenue filter, maybe an estimating agent. You do not use it on the hospital wing you are bidding tomorrow. No way. Instead, you pull up a project you completed six months ago. A project
where you already know exactly how the bidding went, exactly what the quantities were, and exactly how the schedule played out. Okay, I see where this is going. You feed that past architectural plan and spec book into the AI agent and you run it in parallel in the shadows against what your human estimator originally did. Oh, it's a digital bake -off. For 30 days, you let the AI agent estimate the historical job. Then you sit down with your chief estimator and you lay the AI's output next to your human -created spreadsheet. Yep. Did the AI catch the same amount of steer? Did it miss a tricky elevation change? Did it finish the takeoff in four hours when it took your team four days? Exactly. And the primary benefit of the shadow run is the psychological safety it provides. Oh, absolutely. You are proving
the return on investment with absolute zero risk to your current clients. Your estimating team isn't terrified that some algorithmic glitch is going to cost the company a million dollars because the job is already built and paid for. And honestly, it completely reframes the introduction of the technology. How so? When your veteran estimator sees the AI accurately pull 80 % of the rote data out of a complex plan set, they stop seeing it as a threat. Yes. They realize they can spend their time on value engineering, finding cheaper, more efficient materials or alternative methods without sacrificing quality. They can spend their time negotiating better rates with subcontractors rather than counting toilets on a floor plan. It elevates them from data entry clerks back into strategic builders. Spot on. The 30 -day shadow pilot is how you
cross that autonomous tipping point without losing your balance. If it works on the scrap wood, you bring it onto the main floor. If it splinters, you cancel the subscription, move on, and all it costs you is a few hours of setup time. That's brilliant. All right, this has been an incredibly dense and valuable deep dive. Let's do a quick recap of the journey we just took. Sounds good. We started with the massive signal coming from Silicon Valley. Y Combinator's 2026 construction cohort is making a historic bet on construction intelligence, proving that back office automation is the next frontier. We broke down the vital shift from passive dashboards. you know, the software you're exhausted by and ignore, to active AI agents powered by large language models that read context and do the heavy lifting of pulling
quantities and flagging schedule conflicts. Yep. We gave you the contractor's filter. If a tool doesn't directly increase your bid -win rate or help you capture missed change orders, It doesn't belong in your tech stack. Skip it. And finally, we laid out the 30 day shadow pilot strategy, taking your past project data, running it through the AI and comparing it against human performance to prove the ROI with absolutely zero risk. to a live job. It's a comprehensive playbook for sure. But, you know, as we move toward this autonomous tipping point, this raises a really important question. Okay. It's something I want everyone listening to really ponder. We have spent this entire discussion talking about optimizing your side of the table. Right. But the general contractors, the developers, the owners, they're adopting these agents too. Oh,
man. Yeah. So if your new AI agent is perfectly optimizing your bids to win more jobs and the general contractor's AI agent is perfectly optimizing how they select and level those bids are we rapidly approaching a future where your ai is simply negotiating with their ai and if so what happens to the human relationships the handshakes the trust the late night phone calls that the construction industry was fundamentally built on man That is a heavy thought to leave off on, an AI standoff in the bidding room. We will definitely have to tackle the future of industry relationships in an upcoming deep dive. But for now, thank you so much for joining us and spending your time with us today. Stay curious, keep building, and keep scaling.
