AI Underwriting for Contractor and Construction Lending

Contractor lending looks adjacent to standard C&I underwriting, and at a glance the borrowers run like other small businesses: an operating company, a guarantor, sometimes a related real-estate entity, sometimes a related equipment LLC. The line on the call report is similar. The work in the file is not.

Contractors operate on long-term contracts. Under IRC Section 460, the default tax treatment for those contracts is percentage-of-completion, which means the borrower recognizes revenue as costs are incurred against the total estimated cost of the job. Smaller contractors below the indexed gross-receipts threshold can elect the completed-contract method and defer revenue until substantial completion. Either way, the trailing tax return alone is a poor mirror of operating reality. The work-in-process schedule is where the story actually lives.

Layer in retainage, a balance sheet that carries cranes and excavators rather than office furniture, a surety relationship that gates how much backlog the contractor can take on, seasonal cash flow swings tied to weather and bid cycles, and the file starts to fight standard underwriting templates. Two analysts looking at the same contractor return can produce wildly different EBITDA depending on how they treat percentage-of-completion adjustments, equipment depreciation, and operating-lease payments.

This is where AI belongs. Not on the credit call. On the operational work of extracting the WIP schedule, reconciling billings against revenue, surfacing retainage timing, normalizing EBITDA across the contractor add-back framework, tying equipment debt to UCC filings, and assembling a memo with bonding capacity and backlog framing the underwriter can actually defend. For the broader workflow context, the AI-Assisted Underwriting Playbook covers the document intake, spreading, and memo layers that every commercial file shares. Contractor lending applies that workflow to a borrower type where the documents disagree more often than not.

Why the Files Fight Standard Templates

Contractor underwriting looks templated and almost never is

The same handful of issues drag on every contractor file. They are not exotic. They are the predictable consequences of how contractors recognize revenue, manage receivables, and carry their balance sheets.

Percentage-of-completion creates a wedge between billings and revenue

A contractor on percentage-of-completion recognizes revenue based on cost incurred against total estimated cost. Billings follow the schedule of values negotiated with the owner, which is rarely the same curve. Jobs that are billed ahead of progress produce billings in excess of costs and estimated earnings, which sits on the balance sheet as a liability. Jobs billed behind progress produce costs and estimated earnings in excess of billings, which sits as an asset. Both lines can swing materially between reporting periods. An analyst spreading a year-end statement without the matching WIP schedule will get the numbers in the right boxes and the story wrong.

Retainage looks like a receivable, behaves like a long-dated bet

Owners and general contractors typically withhold a portion of each progress payment until substantial completion. The withheld amount sits on the balance sheet as retainage receivable. State prompt-payment statutes govern release timing, and on a multi-phase job the retainage on early work can sit unbilled for a year or more. Treating retainage as if it converts on a normal A/R cycle inflates working capital. Ignoring it understates collateral. Underwriters who see contractors only occasionally tend to default to one of the two extremes.

Bonding capacity caps the deal, but rarely appears on the spread

For any meaningful public or commercial job, the contractor needs a surety to issue bid, performance, and payment bonds. The Federal Acquisition Regulation requires performance and payment bonds on most federal construction contracts above the statutory threshold, and most state and large private owners have parallel requirements. Surety capacity is set by the surety on the contractor's working capital, net worth, equipment, banking line, and contract history. If the bank extends a working line that lets the contractor bid work the surety will not bond, the line goes unused. If the surety expands capacity beyond what the bank line can fund through the work-in-process cycle, the contractor runs out of cash mid-job. The two facilities have to be sized together, and that conversation rarely shows up in a generic credit memo.

Equipment add-backs are not optional, and they are not free

Construction firms carry capital-intensive balance sheets. Depreciation, equipment loan interest, and operating-lease payments are all real cash outflows over time even when accounting allows them as add-backs in a given year. A clean EBITDA add-back of depreciation overstates sustainable cash flow if the equipment replacement cycle is not funded. The honest version of the spread runs fixed-charge coverage with new payments layered on top, includes a replacement reserve, and reconciles existing equipment debt against UCC filings and the interest expense line on the return. The same discipline applies on a dedicated equipment file, and the equipment financing solution page walks through the add-back framework end to end.

Workflow

Where AI actually helps in a contractor lending workflow

1. WIP schedule extraction

Most contractor packages include a work-in-process schedule produced from the accounting system, sometimes formatted in spreadsheet form, sometimes printed as a PDF report with a row per active job. The fields are consistent: contract value, change orders, revised contract amount, costs incurred to date, estimated cost to complete, total estimated cost, percent complete, billings to date, revenue earned to date, and the resulting over- or under-billing. Good extraction reads each row, ties the totals to the financial statement balance, and surfaces the jobs with the largest billing variances first. The analyst then judges which variances reflect timing and which reflect a job in trouble. That judgment cannot be automated, and should not be.

2. Spreading with contractor-specific normalization

A contractor spread runs on fixed-charge coverage rather than raw DSCR, because operating leases on equipment, existing equipment debt, and the proposed new payment all matter together. The spread should walk EBITDA to EBITDAR by adding back operating-lease payments, then layer in existing debt service, equipment-loan principal and interest, and the proposed facility. Percentage-of-completion adjustments need to land consistently across years so that a job completing in year three does not look like an earnings windfall. The same spreading discipline shows up on every commercial file the financial spreading software solution page covers, with the contractor add-back framework layered on top.

3. Retainage and billings reconciliation

Retainage receivable, billings in excess of costs, and costs in excess of billings should all reconcile to the WIP schedule. AI should pull the three balance-sheet lines from the financial statement, sum the matching columns from the WIP schedule, and flag the gap. A clean tie gives the underwriter confidence that the WIP schedule and the financials describe the same business. A meaningful gap is one of the first signs that the contractor's accounting system and reporting are out of sync, which is almost always a finding the bank wants to know about before closing.

4. Equipment lien and debt tracing

Contractor-heavy borrowers carry a stack of equipment loans, leases, and floor-plan-style lines. The self-reported debt schedule is almost always incomplete. The interest expense line on the return, the UCC-1 filings, and the equipment depreciation schedule all have to be reconciled against each other. AI should match tradelines on the credit bureau against the self-reported schedule, surface UCC filings that do not appear in either, and flag interest expense that cannot be explained by the reported debt stack. Missed equipment debt is one of the most common post-close surprises on contractor files, and the file should not produce it.

5. Bonding capacity and backlog framing

The credit memo should describe the surety relationship explicitly: who issues the bonds, the current single-job and aggregate capacity, the largest open contract, the total work-on-hand backlog, and the contractor's bid pipeline. A contractor with bonded capacity well above current backlog has room to grow. A contractor with backlog pushing capacity is one big bid loss away from a revenue gap. None of this is exotic. It is just usually not on the page the credit committee sees, because the system that produced the spread did not know to ask. The same global-cash-flow reasoning that the global cash flow automation guide describes applies to backlog framing: pull every relevant data point into one view, then let the underwriter weigh it.

6. Credit memo assembly

The contractor memo has more moving parts than a typical C&I memo. The spread, the WIP schedule reconciliation, the equipment debt schedule, the bonding context, the backlog summary, and the proposed structure all have to tie. When that memo is written from scratch while the exhibits are still being cleaned, the narrative drifts fast. Done right, AI provides a stronger starting frame: the system pulls the spread outputs, the WIP reconciliation, the equipment schedule, and the bonding context into one place so the underwriter writes the judgment part rather than re-keying the factual part. The same division of labor runs through AI credit memo generation: machine assembles support, lender writes the recommendation.

Worked Example

A general contractor with a multi-year backlog and an equipment-heavy balance sheet

Consider a regional general contractor running $45M of trailing-twelve-month revenue, primarily K-12 and municipal work. The borrower is on percentage-of-completion. The trailing tax return shows $2.1M of net income with $1.8M of depreciation and $620K of interest expense. The balance sheet carries $14M of equipment at cost, $7.2M of accumulated depreciation, $4.1M of equipment debt, $2.3M of retainage receivable, and $1.6M of billings in excess of costs. Bonded backlog is $62M across nine active contracts; single-job bonding capacity is $25M; aggregate capacity is $80M.

None of this is hard analysis. It is hard logistics. The work that takes a senior analyst the better part of a day on this file is not deciding whether to approve. It is getting the WIP schedule reconciled against the financial statement, the equipment debt reconciled against the UCC filings, the retainage classified against active jobs, and the bonding context into the memo before the credit committee asks. The technology that earns the seat in this workflow is the technology that does that logistics work cleanly and shows its work.

Review Standards

What loan review and examiners look for on a contractor file

Reviewers reading a contractor credit memo look for the same things they look for on any commercial file, plus a few that are specific to construction. The contractor-specific list is short and predictable.

• WIP schedule tied to the financial statement. The retainage receivable, costs in excess of billings, and billings in excess of costs lines on the balance sheet should reconcile to the corresponding columns on the WIP. A gap should be explained in the memo, not buried.
• Consistent treatment of percentage-of-completion adjustments. If the bank normalizes EBITDA across POC swings, every file in the portfolio should do it the same way. Drift across analysts is what loan review notices first.
• Equipment debt that reconciles against UCC and tax-return interest. If reported interest expense cannot be explained by the disclosed debt schedule, the schedule is incomplete.
• A documented bonding relationship and current capacity. The memo should name the surety, describe the program, and state current single-job and aggregate capacity. Reviewers should not have to email the relationship manager to find this.
• A memo that matches the exhibits. Numbers in the narrative should match the spread, the WIP reconciliation, and the equipment schedule. Drift between narrative and exhibits is the fastest way to lose trust in the file.

The discipline overlaps with the broader category map in the best commercial lending software guide. Contractor files raise the cost of doing this badly, because reconciling the WIP schedule, the financial statement, and the debt and equipment exhibits is the work, not a side task.

How to Implement It

The right rollout for a contractor book is the boring one

The best place to start with AI on a contractor book is document classification and spreading. Let the system normalize EBITDA and fixed-charge coverage across analysts. Move WIP extraction and retainage reconciliation in next, since those are the highest-error steps in the manual workflow. Save memo assembly and bonding-context generation for last, after the extraction work has earned trust. A parallel run on twenty recently closed contractor deals (especially the ugly ones with rework, change-order disputes, or a job that finished underwater) is the cleanest validation.

That is also the most defensible way to introduce Aloan into a contractor lending team. Aloan's commercial lending workflow fits contractor underwriting because it is built around document traceability, spreading, reconciliation, and memo support rather than pretending to replace credit judgment. The file gets cleaner. The lender stays in charge.

For commercial lending teams that are growing their contractor book on the same analyst capacity that worked at half the size, that is the practical answer. Take the manual work off the parts of the file that do not reward manual effort. Keep the credit judgment where it belongs.