What is AI construction contingency sizing? AI construction contingency sizing is the use of AI tools to build a construction contingency from the bottom up, pricing each identified risk in a project budget instead of applying one flat percentage across the whole job. The output is a defensible number you can put in front of a lender and an equity partner, along with the specific risks that justify it. This is one of the more consequential calculations in our complete guide to AI CRE finance and capital markets, because contingency is the line that decides whether an overrun is absorbed by the budget or becomes a capital call.
Key Takeaways
- A flat 5 percent contingency is a habit, not an analysis. Sized bottom-up against real project risks, most 2026 budgets land closer to 8 to 12 percent.
- Contingency should scale with design completeness. Drawings at 60 percent carry far more undefined scope than a fully bought-out guaranteed maximum price.
- Escalation belongs on its own line. Apply it only to scope not yet bought out, over the actual months remaining until buyout.
- Track contingency burn against percent complete. Spending 60 percent of the buffer at 30 percent complete signals exhaustion long before the money is gone.
- AI sizes and monitors the number. It does not know your subcontractor market, and a local general contractor's judgment still overrides the model.
Why a Flat 5 Percent Contingency Fails
A flat 5 percent contingency fails because it prices the budget rather than the risk. Two projects with identical hard costs can carry wildly different exposure: one with permitted drawings and every trade bought out, another at design development with steel and electrical gear unpriced. The same percentage applied to both is either wasted equity on the first or an underfunded budget on the second. Contingency is a risk estimate, and risk estimates should be built from the risks.
The 2026 environment punishes the shortcut harder than usual. Costs are still climbing, though at a slower pace than the recent past, and the pressure is uneven rather than general. The Urban Land Institute's 2026 construction costs outlook identifies skilled labor availability and craft wage rates as the central concern for contractors, notes that imported materials and those competing with imports face outsized increases, and reports data center construction forecast to rise 7 percent to $195 billion, which pulls electricians, switchgear, and generator capacity out of every other project's market. Uneven pressure is precisely what a single blended percentage cannot capture.
The Four Inputs AI Uses to Size Contingency
Sizing contingency with AI means feeding the model four specific inputs and having it price each separately, then sum them. Give it the drawing set status, the buyout schedule, the geotechnical and existing-condition reports, and the trade-level budget breakdown, and ask for a line-by-line contingency build rather than a single number.
- Design completeness gap: the scope that is not yet drawn. Schematic design carries the widest gap, design development narrows it, and construction documents narrow it further. This is the largest single driver on most ground-up projects.
- Escalation on un-bought-out scope: the cost of time between today and the date each trade is locked. Applying escalation to already-bought-out scope double counts and inflates the budget.
- Site and existing conditions: subsurface unknowns, undocumented utilities, abatement, and on renovation work the findings from a property condition assessment. The ASTM E2018 Standard Guide for Property Condition Assessments is the common baseline, though ASTM sells the standard rather than publishing it free.
- Trade-level volatility: exposure concentrated in tariff-sensitive and supply-constrained categories such as steel, copper, and electrical distribution equipment, rather than spread evenly across the budget.
Models including Claude Opus 5, GPT-5.6, and Gemini 3.1 Pro handle this well because it is structured arithmetic over documents you already have. The analytical companion is our guide to AI construction cost estimation and bid analysis, which covers getting the base budget right; contingency sizing assumes that base number is already sound.
A Worked Example: Sizing Contingency on a 120 Unit Build
Take a 120 unit multifamily project with a hard cost budget of $28,800,000, or $240,000 per unit, at 60 percent design development drawings with roughly 70 percent of scope still to be bought out over the next 18 months. A bottom-up build looks like this:
- Design completeness gap: 4.0 percent of hard costs, or $1,152,000, reflecting scope that will appear as drawings finish.
- Escalation: 5 percent annualized applied to the 70 percent not yet bought out, over 18 months. That is $28,800,000 times 0.70 times 5 percent times 1.5 years, or $1,512,000.
- Subsurface and existing conditions: 1.5 percent, or $432,000, based on the geotechnical report's flagged variability.
The total is $3,096,000, which is 10.75 percent of hard costs. A flat 5 percent would have carried $1,440,000, leaving a $1,656,000 gap. On a project capitalized with $9,000,000 of equity, that gap is an 18.4 percent capital call to the partnership, arriving at the worst possible moment in the schedule. The number is also the argument: a lender reviewing a 10.75 percent contingency with a line-item derivation behind it treats it very differently from a round percentage with nothing underneath.
Contingency sits inside the loan budget, so it interacts directly with proceeds. Our guide to AI CRE loan sizing and maximum loan amount covers the tests that cap those proceeds, and once a project is financed the buffer's adequacy feeds straight into the stabilization test discussed in our guide to modeling the construction to perm conversion. An exhausted contingency and a missed conversion covenant tend to arrive together.
Tracking Contingency Burn Against Percent Complete
Sizing the contingency is a one-time exercise; monitoring the burn is what actually saves the project. The single most useful metric is a burn ratio: contingency spent as a share of contingency total, divided by percent complete. A ratio near 1.0 means you are consuming the buffer at the pace the project is progressing. Burning 60 percent of contingency at 30 percent complete gives a ratio of 2.0, which means the buffer is on pace to be gone at roughly the halfway mark.
This is straightforward to automate. Point the model at the monthly pay application, typically an AIA G702 with its G703 continuation sheet, plus the current schedule of values, and have it compute the ratio each cycle and classify every contingency draw by cause: design gap, differing site condition, owner-directed change, or escalation. The cause tagging is what makes the trend readable. A buffer consumed by owner-directed upgrades is a discipline problem with a straightforward fix. A buffer consumed by differing site conditions in the first three months is a forecasting problem, and the honest response is to re-forecast the remaining contingency upward now rather than at 70 percent complete when no options remain. The tooling cost for this is small relative to the exposure, as our breakdown of AI implementation cost for real estate firms lays out. Sponsors who want a burn tracker wired into their monthly draw process can get hands-on implementation support from The AI Consulting Network.
What AI Gets Wrong About Construction Contingency
The most frequent error is double counting between contingency and escalation. Models asked for "a contingency" will often fold a general escalation allowance into the percentage and then add an escalation line separately, inflating the budget by the same risk twice. Instruct the model explicitly that escalation covers known price movement on unpriced scope and contingency covers unknown scope, and require it to show which line each risk lands on.
Second, models default to national averages when your market is what matters. National escalation figures are a starting point, not an input; a general contractor with live bid data from your submarket will beat any model on the actual number. Use AI to build the structure and the arithmetic, then let the GC correct the assumptions.
Third, models tend to conflate owner contingency with GC contingency. They are separate pools with separate control rights: the GC's sits inside the contract sum for the contractor's own risks, while the owner's covers scope changes and owner-side decisions. Merging them produces a budget where nobody can say who controls the money. Developers who want this built as a repeatable model rather than a one-off spreadsheet can connect with The AI Consulting Network for hands-on implementation support.
Frequently Asked Questions
Q: What is a normal construction contingency percentage in 2026?
A: The traditional 5 to 10 percent range still applies to well-defined projects with complete drawings and most trades bought out. Longer schedules, material-heavy scopes, and tariff-exposed trades increasingly justify 10 to 15 percent, and projects with unknown existing conditions or complex phasing can support more with a documented risk assessment.
Q: Can AI actually size a construction contingency, or just describe one?
A: It can size one, provided you give it real inputs: drawing status, buyout schedule, geotechnical findings, and a trade-level budget. The arithmetic is structured and the model handles it well. What it cannot supply is local subcontractor pricing, which has to come from your general contractor.
Q: Should escalation and contingency be separate line items?
A: Yes. Escalation covers known price movement on scope not yet bought out and is calculated from time and market data. Contingency covers unknown scope and unforeseen conditions. Combining them hides which risk is driving the number and usually results in double counting.
Q: What happens when the contingency runs out mid-project?
A: The remaining overrun typically becomes an equity requirement, since lenders rarely increase proceeds mid-construction. This is why burn tracking against percent complete matters more than the initial number: catching the trend at 30 percent complete leaves room to value engineer, while catching it at 80 percent leaves only a capital call.