Capital Projects Over Budget? Keeping a Nonprofit Building Budget Honest With AI
A new headquarters, a renovated shelter, a clinic build-out, a food bank warehouse, or a restored performance hall is often the largest financial commitment a nonprofit will ever make, and it is usually managed by a finance team that has never run a construction project. Budgets drift for predictable reasons: soft costs nobody priced, contingency spent in the first month, change orders approved by email, pay applications that bill ahead of the work, and pledge payments that arrive years after the contractor needs to be paid. This article explains how capital budgets are built and how they slip, and shows where AI can help a small team compare bids, track every change, check every invoice, and tell the board the truth about variance. It is just as clear that your owner's representative, architect, counsel, and auditor own the approvals.
Most nonprofit capital projects do not go over budget because of a single dramatic event. They go over budget through accumulation. The architect's fee was based on an earlier scope. The permit took four months longer than planned and the general conditions kept running. The contractor found asbestos behind a wall, which was a legitimate change, and then a dozen smaller changes followed, each one reasonable on its own. Furniture, signage, IT cabling, moving costs, and the security system were never in the construction contract at all. By the time anyone adds it up, the project is 15 or 20 percent over the number the board approved, and the capital campaign has already been sized to the old figure.
The organizations that avoid this are not necessarily the ones with bigger budgets or more experienced staff. They are the ones that keep a single, honest version of the budget and update it every time something changes, so that the gap between what was approved and what is being spent is always visible. That sounds simple, and in principle it is. In practice it means reading contracts, bids, schedules of values, change order requests, and monthly pay applications that arrive in inconsistent formats from several parties, then reconciling them against each other and against a cash flow plan that depends on donors paying their pledges on time. A finance director doing this on top of payroll, audit, and grant reporting will fall behind.
AI is useful here because so much of the work is reading and reconciling documents. A capable model can lay three bids side by side, summarize a 60-page construction contract into the clauses that matter for cost, keep a change order log current, flag a pay application that bills 70 percent of the framing when the site photos show 40 percent, and turn a variance report into a board update that a non-specialist can follow. None of that replaces the professionals on the project. It gives a small team the capacity to keep up with them and to ask better questions.
This article covers how a capital budget is structured and why each line matters, the case for an owner's representative, how to read bids and schedules of values, how to control change orders and review pay applications including retainage, how to match construction cash flow to pledge collections and when bridge financing or New Markets Tax Credits come into the picture, the strings that come with federal and state capital money, and how to report variance to the board and capitalize costs properly when the building is done. Several neighboring topics have their own articles. If you are still deciding whether to build at all, start with our guide to the lease-or-buy decision. If you are raising the money, see our article on running capital campaigns with AI. This piece is about the stretch in between: keeping the project budget honest once the money is committed and the work begins. It is general information, not legal, tax, or construction advice.
Building a Budget That Can Survive Contact With the Project
A capital project budget has two broad halves. Hard costs are the physical work: demolition, site work, foundations, structure, roofing, mechanical, electrical, plumbing, finishes, and the general contractor's overhead, fee, and general conditions. These are what a construction contract covers. Soft costs are everything else needed to deliver a usable building: architectural and engineering fees, surveys, environmental and geotechnical studies, permits and impact fees, legal and accounting, builder's risk insurance, financing costs, owner's representative fees, testing and inspections, furniture, fixtures and equipment, technology and security systems, signage, moving, and temporary space while the work happens. Soft costs are where nonprofit budgets most often prove incomplete, because the board approved a construction number and the soft costs were treated as an afterthought.
Two more lines decide whether a budget survives. Contingency is money set aside for things you know will happen but cannot yet price. Good budgets carry more than one kind: a design contingency that shrinks as drawings are completed, a construction contingency for unforeseen conditions and owner-requested changes, and sometimes a separate owner's contingency held by the organization rather than the contractor. Renovations of older buildings need more contingency than new construction because of what hides in walls, floors, and soils. The most important rule about contingency is that it is drawn down on purpose, with a record of each use, not quietly absorbed into other lines.
Escalation is the allowance for prices rising between the time the budget is set and the time the work is bought. A budget prepared during a feasibility study may not be bid for a year or more, and construction prices have not been standing still. The producer price index for inputs to nonresidential construction rose 3.3 percent from December 2024 to December 2025, with contractors and the Associated General Contractors attributing much of the increase to tariffs, as Facilities Dive reported, and materials prices continued climbing through 2026. A budget without an explicit escalation line is a budget priced at yesterday's costs. Your cost estimator or contractor should state the assumption, and the board should see it as its own line.
AI can help you pressure-test a draft budget before it goes to the board. Give a model your line items and a plain description of the project, and ask it to list soft costs that commonly appear in similar projects but are missing from yours, to identify where contingency is absent or thin, and to show which numbers come from an estimate, a quote, or a guess. It will not know local permit fees or prevailing construction costs in your market, and it should not be used to produce the estimate itself. Its value is as a checklist that never gets tired and a reader that will notice when the furniture line is zero.
Hard cost lines
Usually inside the construction contract
- Demolition, abatement, and site work
- Structure, envelope, and roofing
- Mechanical, electrical, plumbing, and fire protection
- Interior construction and finishes
- General conditions, contractor fee, insurance, and bonds
Soft cost and reserve lines
The lines most often missing from nonprofit budgets
- Design, engineering, surveys, testing, and inspections
- Permits, impact fees, legal, accounting, and financing costs
- Furniture, fixtures, equipment, IT, security, and signage
- Moving, temporary space, and program disruption
- Design and construction contingency, owner's contingency, and escalation
Owner's Representative or Going It Alone
On every construction project, the architect works for the design, the contractor works for the contract, and someone needs to work only for the owner. In a large institution that person is a facilities director with project experience. In most nonprofits nobody on staff has managed a construction project, and the executive director or finance lead ends up doing it by default, often while running the organization through the disruption of the project itself. An owner's representative, sometimes called an owner's project manager, is a consultant hired to fill that gap: reviewing the budget and schedule, managing the procurement of the contractor, attending site meetings, reviewing change orders and pay applications, and telling the organization what it needs to decide and when.
The fee is real, typically structured as a percentage of construction cost or a monthly retainer, and it belongs in the soft cost budget. The argument for paying it is that an experienced representative tends to save more than they cost by catching scope gaps in bids, pushing back on change order pricing, and keeping the schedule from slipping. The argument for going without one is stronger on small, simple projects, such as an interior fit-out of leased space with an experienced contractor and a modest budget, and weaker on anything involving new construction, a historic building, public funding with compliance strings, or a capital campaign whose donors will expect careful stewardship. Many funders and lenders on larger projects will expect to see one.
AI changes this calculation at the margins, not at the center. It can make a staff-led project more manageable by doing the document work: summarizing site meeting minutes, tracking open decisions, drafting requests for information, and keeping the change order log. It can make an owner's representative more efficient and give staff a way to understand and check that representative's recommendations. What it cannot do is walk the site, judge whether a contractor's claim of unforeseen conditions is legitimate, or negotiate a change order with someone who knows the local market. If the honest answer to "who on our side has done this before?" is nobody, AI is not a substitute for hiring someone who has.
Whichever path you choose, put roles in writing. A simple responsibility matrix should state who can approve what: which changes the owner's representative may approve up to a dollar limit, which require the executive director, which require the board or its building committee, and who signs pay applications. That matrix is what keeps a well-meaning program director from agreeing to an extra doorway on a site walk and creating an unpriced change. It also reflects the same principle behind segregation of duties in a small finance office: the person who requests a change should not be the only person who approves and pays for it.
When an owner's representative is worth the fee
The more of these that apply, the stronger the case
- No one on staff has managed a construction project of comparable size
- The project involves new construction, an older or historic building, or significant site work
- Federal, state, or local capital funds carry procurement, wage, or reporting requirements
- The financing structure involves a construction lender, bridge loan, or tax credit investors
- Programs must keep operating in or next to the construction zone
Reading Bids and Schedules of Values Side by Side
When bids come in, the instinct is to compare the bottom lines and pick the lowest. That is how many nonprofits buy their first budget overrun. Bids are only comparable if they price the same scope, and they rarely do. One contractor includes the site utility work and another excludes it. One carries an allowance of a few thousand dollars for hardware where another carries a realistic number. One lists a page of exclusions and qualifications that quietly moves risk back to the owner. The low bid with the longest exclusion list is frequently the most expensive project by the time it is finished.
The schedule of values is the tool that makes comparison possible. It breaks the contract sum into line items, usually organized by trade or by specification division, and it becomes the basis for every pay application later. Asking all bidders to submit a schedule of values in the same format, using the same line items, turns a stack of inconsistent proposals into something you can lay side by side. Industry forms such as the AIA G703 continuation sheet are widely used for this, and your architect or owner's representative will have a preferred template.
This is a task where AI pays for itself quickly. Upload the bid documents and ask the model to build a comparison table with one row per scope item and one column per bidder, flagging items that one bidder priced and another excluded, allowances that differ sharply, and line items that are unusually high or low relative to the others. Ask it to extract every exclusion, qualification, and assumption from each bid into a separate list, and to identify which of them would become owner costs or likely change orders. Then ask it to draft the clarification questions to send each bidder before award. The model will occasionally misread a scanned table, so check its extraction against the source pages for any number that drives the decision. The analysis itself should go to your owner's representative or architect, who will know which differences reflect real scope and which reflect pricing strategy.
Front-loading deserves special attention. A contractor may assign more value to early line items like mobilization, general conditions, and site work than those items really cost, which lets them bill more cash early in the project. It is common, sometimes tolerated, and occasionally a warning sign about a contractor's own cash position. Because the schedule of values drives every future payment, it is worth asking AI to compare early-phase line items across bidders and flag any that look heavy, then having your representative challenge them before the contract is signed. Once the schedule of values is agreed, it is much harder to change. The same discipline applies to the contract itself; our article on AI-assisted contract review covers how to have a model summarize payment terms, retainage, change order procedures, liquidated damages, and warranty clauses before counsel reviews them.
Bid leveling prompts that work
Give the model the documents, then ask specific questions
- Build a table of every scope item, showing which bidders included, excluded, or carried an allowance for it
- List every exclusion and qualification by bidder and say which would likely become an owner cost
- Compare allowances and flag any that look unrealistically low for the described scope
- Flag early-phase schedule of values lines that look front-loaded relative to other bidders
- Draft clarification questions for each bidder to answer in writing before award
Change Orders: Where Budgets Quietly Leak
Every construction project has changes. Some come from conditions nobody could see, like rotten joists or an undocumented utility line. Some come from design errors or omissions. Some come from the owner, when staff realize the intake room needs a second door or the kitchen needs a different hood. Each category has different implications for who should pay, and a good contract spells out the process: the contractor submits a change order request with pricing and any schedule impact, the architect and owner's representative review it, and an authorized person approves or rejects it in writing before the work proceeds, except in true emergencies.
Budgets leak when that process breaks down. Verbal approvals on site walks, changes approved by email without pricing, work done first and priced later, and "no cost" changes that later turn out to have a cost are all common. So is the slow drift of many small changes, none large enough to trigger board attention, that together consume the contingency. The defense is a change order log: a single list of every potential, pending, approved, and rejected change, with the originator, reason category, requested amount, negotiated amount, schedule impact, approval date and approver, and the budget line or contingency it draws from. A log maintained weekly gives you a running answer to the most important question on the project: how much contingency is left, and how much is already spoken for by pending changes?
AI can do much of the clerical work of keeping that log current. Feed it each change order request as it arrives, and ask it to summarize the change in a sentence, extract the price and schedule impact, classify the likely cause, and check whether the pricing breaks out labor, materials, and markup the way the contract requires. Ask it to compare the request against earlier changes and the original bid to see whether the work might already be in the base scope, which is one of the most common sources of improper change orders. Ask it to recalculate remaining contingency after each approval and to project where contingency will land if pending changes are approved at their requested amounts. These are all things an owner's representative should be doing; AI lets a staff member follow along and notice when something is not being done.
Keep the boundary firm. The model can say a change order looks like it duplicates base scope. It cannot decide that it does. Determining whether a condition was foreseeable, whether a design omission is the architect's responsibility, and whether a price is fair requires people who know the drawings, the contract, and the market. Treat every AI flag as a question for your representative and architect, and record their answer in the log.
Fields in a useful change order log
One row per change, from first mention to final decision
- Number, date raised, originator, and a one-sentence description
- Cause category: unforeseen condition, design error or omission, owner request, code or regulatory requirement
- Requested amount, negotiated amount, and schedule impact in days
- Status: potential, pending, approved, rejected, or withdrawn
- Approver, approval date, and the budget line or contingency it draws from
- Funding source, especially if a grant restricts what the change can be charged to
Pay Applications, Percent Complete, and Retainage
Contractors are usually paid monthly through a pay application, often on industry forms like the AIA G702 application and G703 continuation sheet. Each application lists every line of the schedule of values, the value of work completed this period and to date, materials stored on or off site, the percent complete for each line, the amount retained, prior payments, and the current amount due. The architect typically certifies that the work has progressed to the point indicated, and the owner then pays. For a nonprofit, this monthly document is the moment of greatest financial exposure on the project, because once money is paid ahead of the work it is very hard to get back.
Retainage is the portion of each payment held back, commonly a set percentage of completed work, until substantial or final completion. It protects the owner by keeping leverage for punch list items, closeout documents, warranties, and lien releases. Retainage terms are set by the contract and, for many public and some private projects, limited by state law, so the percentage and the conditions for releasing it vary. Your contract and counsel decide the terms. Your job each month is to confirm the right amount was withheld, that retainage was not released early, and that the total retained matches what the contract says.
Reviewing a pay application well means checking it against three things. First, against itself: the math across every line, the carried-forward totals from the prior application, and the retainage calculation. Second, against the contract: the schedule of values lines match what was agreed, approved change orders appear and unapproved ones do not, and stored materials are documented the way the contract requires. Third, against reality: the percent complete billed for each line is consistent with what is actually built, as shown by site observations, photos, inspection reports, and the architect's certification. The first two checks are mechanical and well suited to AI. The third requires eyes on the site.
Ask a model to reconcile each new pay application with the previous one and the schedule of values, to recompute every total, to flag lines where billed percent complete jumped sharply in one month or exceeds the schedule position, and to flag any line billed beyond 100 percent or any change order billed before it was approved. Give it the owner's representative's site report and ask it to identify lines where the billing seems ahead of the reported progress. That flag, "the application bills 70 percent of framing but the site report describes the second floor as not started," is exactly the question that should be raised before the check is signed. Pair each payment with conditional and unconditional lien waivers from the contractor and major subcontractors and suppliers, as your counsel advises, and ask AI to keep a tracker that shows which waivers are outstanding for which payments.
Monthly pay application checklist
AI handles the mechanical checks; people confirm the work exists
- All line math, prior totals, and the retainage calculation reconcile
- Schedule of values lines match the contract, and only approved change orders appear
- No line is billed over 100 percent, and no line jumps far ahead of reported site progress
- Stored materials are documented, insured, and located where the contract allows
- Architect certification is attached and the owner's representative recommends payment
- Lien waivers for the prior payment are received from the contractor and key subcontractors
Matching Construction Cash Flow to Pledge Collections
A project can be fully funded on paper and still run out of cash. Capital campaigns raise much of their money in multi-year pledges, with payment schedules that often stretch three to five years. Construction spending follows a very different curve: modest during design, then a steep climb once work starts, peaking in the middle months of construction and tapering at the end. Plotted on the same chart, the two curves almost never line up. The gap between them is a financing problem, and it needs to be planned before the contractor is under contract rather than discovered when the first large pay application arrives.
Building the cash flow plan starts with two schedules. The spending schedule comes from the contractor's projected billings by month, plus soft costs timed to when they will actually be invoiced. The collections schedule comes from the pledge records, adjusted for reality: not every pledge will be paid on schedule, some will be paid early, and a small share may never be paid. Our article on gift agreements and pledge documentation covers how to get pledge terms recorded clearly enough to forecast from. AI is a useful partner here: it can assemble both schedules into a single month-by-month model, apply collection assumptions you choose, show the cumulative gap at each month, and run scenarios such as a three-month construction delay, a 10 percent cost overrun, or a major pledge paid a year late. The broader techniques are covered in our guide to AI cash flow forecasting. The point for a capital project is that the finance committee should see the trough, the month when cumulative spending most exceeds cumulative collections, and know how it will be covered.
Common ways to cover the gap include a bridge loan or line of credit secured by pledges receivable, a construction loan secured by the property, a temporary draw on reserves with board approval and a repayment plan, or phasing the project so that later phases wait for collections. Each has costs: interest and fees on borrowed money, covenants that require regular reporting, and in some cases lender requirements for an owner's representative, a fixed-price contract, or a minimum percentage of pledges collected before closing. Lenders familiar with nonprofit capital projects, including community development financial institutions, will often want to see the very cash flow model described above. Whether any financing is appropriate is a decision for the board with advice from your auditor and counsel.
For projects in qualifying low-income communities, New Markets Tax Credits are another possible source. The program, administered by the Treasury Department's CDFI Fund, gives investors a federal tax credit in exchange for investing through Community Development Entities that then finance qualifying projects. Nonprofits do not apply to the CDFI Fund directly; they work with a Community Development Entity that has received an allocation. Community facilities such as health centers, schools, child care, and arts venues are a significant share of the program's activity, according to IFF's primer for nonprofit leaders, and a well-structured transaction can leave the project with a meaningful amount of capital it does not have to repay. The trade-off is complexity. These transactions involve multiple entities, a seven-year compliance period, specialized legal and accounting work, and fees that make them impractical for smaller projects. AI can help a finance team understand term sheets and prepare for conversations, but this is firmly a specialist's area.
What the finance committee should see before construction starts
A cash plan, not just a funding plan
- Month-by-month projected spending, hard and soft costs combined
- Projected pledge collections with stated assumptions for late and unpaid pledges
- The cumulative cash gap at its deepest point, and the month it occurs
- How the gap will be covered: bridge loan, construction loan, reserves, or phasing
- Scenario results for delay, overrun, and slow collections
The Strings Attached to Public Capital Money
Federal, state, and local capital grants can make a project possible, and they also change how the project must be run. Accepting public money for construction often brings procurement rules, wage requirements, bonding, reporting, and long-term restrictions on the property itself. These obligations attach to the project when the award is accepted, and they are much easier to plan for than to retrofit. Read every award agreement before bidding, and have counsel confirm how each requirement applies to your specific funding.
Prevailing wage. The Davis-Bacon Act requires payment of locally prevailing wages and fringe benefits on federal construction contracts over $2,000, and many federal assistance programs extend similar requirements to the construction they fund through what the Department of Labor calls "Related Acts," as described in the Wage and Hour Division's fact sheet. Whether a particular grant triggers the requirement depends on the program's own statute, so confirm it with the funding agency. If it applies, the correct wage determination must be included in the bid documents and contract, contractors must submit certified payrolls, and someone on the owner's side must review them. Prevailing wage requirements affect bid prices, and a budget prepared without them can be materially low. Many states have their own prevailing wage laws that may apply to state-funded projects as well.
Procurement and bonding. Federal awards subject to the Uniform Guidance require recipients to follow the procurement standards in 2 CFR 200, including competitive procurement and documented cost or price analysis above certain thresholds. For construction contracts above the simplified acquisition threshold, 2 CFR 200.326 sets minimum requirements of a 5 percent bid guarantee and performance and payment bonds for 100 percent of the contract price, unless the awarding agency accepts the recipient's own bonding policy. Our guide to 2 CFR 200 compliance with AI covers the broader rules.
Federal interest in the property. Real property acquired or improved with a federal award is held in trust for the purposes of the program, and under 2 CFR 200.316 the agency may require the recipient to record a lien or other notice of record showing the federal interest. In practice this can restrict selling, mortgaging, or changing the use of the building for a period that can last decades, and disposing of the property later may require compensating the agency for its share. That restriction matters to lenders, who may need the agency's consent or a subordination before they will lend against the building, and it matters to your long-range planning. Requirements vary considerably by program, so verify the specifics for each award. State and local capital grants frequently carry their own use restrictions, recapture provisions, and reporting periods.
AI is helpful for organizing these obligations, not for interpreting them. Ask a model to read each award agreement and extract every construction-related requirement into a compliance matrix: what is required, when, by whom, and what evidence you will need to show it was met. Ask it to cross-check the bid documents and draft contract against that matrix to catch missing wage determinations or bonding language. Ask it to build a schedule of reporting deadlines that extends past the end of construction for the full restriction period. Then send that matrix to counsel and the funder's program officer to confirm. A model reading a grant agreement can miss a requirement incorporated by reference, and the consequences of missing one, such as disallowed costs or wage restitution, can be severe.
Public funding requirements to confirm before bidding
Verify each one with the funder and counsel; requirements differ by program
- Whether federal Davis-Bacon or a state prevailing wage law applies, and which wage determination to use
- Procurement method, documentation, and cost or price analysis requirements
- Bid guarantee, performance bond, and payment bond requirements
- Recorded notice of federal or state interest, use restrictions, and their duration
- Environmental review, accessibility, and historic preservation steps that must precede construction
- Prior approval requirements for budget changes and the reporting calendar after completion
Telling the Board the Truth About Variance
Boards approve capital projects at a number, and that number becomes the reference point for everything that follows. The most damaging pattern in nonprofit capital projects is not an overrun itself. It is an overrun the board learns about late, all at once, after the options for dealing with it have narrowed. A board told in month four that contingency is being consumed faster than planned can reduce scope, delay a phase, extend the campaign, or approve financing. A board told in month eleven can only pay.
A good monthly project report is short and consistent. It shows the approved budget by major category, the current forecast at completion for each category, the variance between them, and an explanation for any variance above a threshold the board has set. It shows contingency remaining, both after approved changes and after pending changes. It shows the schedule against the baseline, the cash position against the cash flow plan, and the top risks the project team is watching. The forecast at completion is the key column, because it reflects not just what has been spent but what the project is now expected to cost when it is done. A report that only shows spending to date can look healthy right up to the point where it is not.
AI makes this report much easier to produce and keep consistent. Given the budget, the change order log, the latest pay application, and the cash flow model, a model can assemble the variance table, calculate contingency positions, and draft the narrative explanations in plain language. Ask it to write two versions: a one-page summary for the full board and a more detailed version for the building or finance committee. Ask it to compare this month's report with last month's and highlight what changed, which is often more informative than the absolute numbers. Our article on preparing board meeting packets with AI covers the broader process. The finance lead and owner's representative should review every number and every explanation before it goes out, because a board update with a wrong figure does more harm to trust than a late one.
Be careful with tone. Models default to reassuring language, and project reports are a place where reassurance can mislead. Instruct the model to state variances plainly, to avoid words like "minor" or "on track" unless the numbers support them, and to say clearly when a decision is needed from the board. The goal is a report that a board member could read in five minutes and correctly answer three questions: are we going to finish within the approved budget, if not by how much, and what do you need from us?
A one-page monthly capital project report
Same format every month, so changes stand out
- Approved budget, forecast at completion, and variance by major category
- Contingency remaining after approved changes and after pending changes
- Schedule status against the baseline, with reasons for any slip
- Cash position against plan, pledge collections to date, and any draws on financing
- Top risks, what changed since last month, and any decisions needed from the board
Closing Out and Capitalizing the Building
When the building is finished, the budget work is not. Closeout involves the punch list, final lien waivers, release of retainage, warranties, operation and maintenance manuals, as-built drawings, and the final reconciliation of the contract sum with every approved change. Each of these should be tracked to completion before final payment, because leverage disappears once the last check is written. AI can build the closeout checklist from the contract and track what has been received, but the architect and owner's representative determine when the work is actually complete.
Then the accounting. Costs of acquiring, constructing, or improving a building are generally capitalized as fixed assets rather than expensed, and the project ledger needs to be translated into entries in the fixed asset register. That means deciding which costs are part of the building, which belong to separately depreciated components like the roof or mechanical systems if your policy uses component depreciation, which are furniture and equipment with shorter useful lives, and which are period expenses that should not be capitalized at all. Interest during construction, grant-funded costs, and the treatment of donated materials or services each raise questions your auditor will want to settle. Our article on maintaining a fixed asset register with AI covers the register itself; the point here is that a project ledger kept cleanly by category from day one makes this step far easier.
Ask AI to map every project cost to a proposed asset category and useful life based on your capitalization policy, and to flag costs whose treatment is uncertain. Then take that mapping to your auditor before year-end close, not after. If federal funds paid for any part of the project, the asset records also need to reflect the federal share and any recorded interest, and the project will likely be examined in your next audit; our guide to audit preparation with AI applies directly. Finally, hand the building over properly to whoever will operate it. Warranties, maintenance schedules, and system documentation belong in your facilities records from the first day, and our article on AI for facility management picks up where this one ends.
Where AI Helps, and Who Owns the Approvals
Across every stage of the project, the pattern is the same. AI is very good at reading many documents, extracting the numbers and terms that matter, reconciling them against each other, and producing clear summaries for different audiences. That is the work that overwhelms a small finance team during a capital project, and doing it consistently is most of what keeps a budget honest. AI is not able to observe the site, judge the legitimacy of a claim, interpret a contract or regulation with authority, or take responsibility for a decision. Those belong to the professionals and officers whose job it is.
Some practical guardrails help. Keep a single source of truth for the budget, change order log, and cash flow model, and have AI work from it rather than from scattered email attachments. Check any figure the model extracts from a scanned document before it drives a decision. Record every human decision, with name and date, alongside the AI analysis that informed it. Be thoughtful about what you upload: construction contracts, bids, and financing documents are confidential business information, and they should go only into tools your organization has approved for that kind of material. And remember that the model has no stake in the outcome, which is useful for spotting problems and a reason it should never be the one that signs off.
Good uses of AI on a capital project
Reading, reconciling, and drafting
- Checking a draft budget for missing soft costs and thin contingency
- Leveling bids and extracting exclusions, allowances, and front-loading
- Summarizing contracts and change orders and keeping the log current
- Reconciling pay applications and flagging billing ahead of progress
- Building cash flow scenarios, compliance matrices, and board variance reports
Decisions that stay with people
Owner's rep, architect, counsel, auditor, and the board
- Selecting the contractor and approving the contract and schedule of values
- Approving or rejecting change orders and judging their legitimacy
- Certifying work in place and authorizing payment or retainage release
- Interpreting grant conditions, wage requirements, liens, and financing terms
- Approving budget increases, borrowing, and accounting treatment
A reminder on scope: Construction contract terms, retainage limits, lien rights, prevailing wage obligations, federal interest requirements, and accounting treatment all depend on your state, your contract, and the specific terms of each funding source. Use this article to organize your work and prepare good questions, then rely on your owner's representative, architect, attorney, auditor, and funders for the determinations themselves.
Conclusion
Capital projects drift over budget through ordinary causes: incomplete soft costs, contingency spent without a record, bids compared on the bottom line, changes approved casually, pay applications paid without checking them against the site, and cash plans that assume pledges will arrive on time. None of these require bad intentions or bad luck. They require only that the people responsible for the budget fall behind the volume of documents the project generates, which is exactly what happens to small nonprofit finance teams in the middle of construction.
AI changes how much a small team can keep up with. Bid leveling, change order tracking, pay application reconciliation, cash flow scenarios, compliance matrices, variance reports, and closeout mapping are all document-heavy tasks that a capable model can handle quickly and consistently, giving staff the time and the information to ask the right questions of the professionals on the project. What it does not change is who is accountable. The owner's representative, architect, counsel, auditor, and board each have decisions only they can make, and an honest budget depends on those decisions being made deliberately, on good information, and on time.
Your donors gave to a building because they believe in what will happen inside it. Keeping the budget honest, and telling the board the truth about it every month, is how you show them that their gifts were stewarded as carefully as they were given.
Keep Your Capital Project on Budget
We help nonprofits set up AI-assisted workflows for bid comparison, change order logs, pay application review, cash flow planning, and board reporting, so a small team can keep pace with a big project. Your professionals make the calls. We help you arrive at every decision with the facts in front of you.