Lending teams at active debt funds are not running slow because they lack skilled underwriters – they are running slow because their skilled underwriters are doing the wrong work.
Industry estimates suggest analysts at commercial origination desks spend roughly 60 percent of their week on mechanical tasks that have nothing to do with credit judgment. Normalizing rent rolls. Flagging missing documents. Rebuilding models from inconsistently formatted PDFs. The credit judgment those analysts were hired to apply gets squeezed into whatever time remains.
With $700B in commercial loans maturing in 2026, the volume of packages hitting underwriting desks is not shrinking. The lending teams keeping pace are not hiring faster. They removed the data-entry layer. AI handles the mechanical work. The analysts do the actual underwriting.
4 Tasks an AI Layer Handles Before an Analyst Opens the File
The most direct impact is document extraction.
Borrower packages arrive in every format. Rent rolls are messy PDFs. T12 financials are formatted differently by every property manager. An AI extraction layer converts those inputs into normalized cash flow models in minutes. The analyst who spent three hours building a model from a raw rent roll now spends twenty minutes validating one.
Package QA works the same way.
Before a deal reaches an underwriter, something is usually missing: wrong entity on the insurance certificate, unsigned rent roll, incomplete schedules. AI flags every gap before the file gets assigned. The analyst gets a complete package on the first pass instead of spending two days chasing documentation.
Covenant parsing follows.
A 100-page loan document buries DSCR thresholds, cash management rules, and reserve requirements in dense legal language. Natural language processing surfaces those terms in seconds. The underwriter still interprets them. They do not have to find them first.
Committee memo drafting is where the savings compound.
AI generates the baseline asset description, submarket demographics, and standard comps. The analyst writes the risk narrative. That is the part that requires judgment. Assembly should not be the bottleneck.
Removing the mechanical work does not shrink the underwriter’s role – it clarifies it.
The Judgment Calls No Algorithm Can Make
The most dangerous misread of AI in underwriting is treating it as a credit decision tool. It is not. It is a data processing tool, and that distinction matters.
Sponsor risk cannot be automated. Whether a borrower will support a distressed asset, negotiate in good faith, or walk away under pressure is not in the rent roll. It comes from experience, pattern recognition, and direct conversation. No model replicates it.
Story risk is the same. AI reads that occupancy dropped from 90 percent to 70 percent. It takes a human underwriter to determine whether a tenant dispute, a repositioning plan, or a local market shift explains it, and whether that explanation survives committee scrutiny.
Complex structures still require judgment. Deals with layered capital, competing claims, and non-standard arrangements need someone thinking through incentives, not just extracting terms.
The final call on whether a deal belongs on the balance sheet is always human. AI provides the arithmetic. The underwriter makes the judgment.
3 Guardrails That Cannot Move
The primary failure mode in AI-assisted underwriting is not the algorithm. It is the analyst who trusts an output they cannot verify. Every number an AI model produces must trace back to a specific cell in a specific source document. No untraceable figures in the model.
The role of AI is extraction and anomaly flagging. The role of the underwriter is deciding what those flags mean. Junior analysts shifting from data entry to validation are not doing less valuable work. They are doing the right work, auditing the machine before the senior underwriter touches the file.
Adoption follows traceability. Teams that start with narrow, verifiable tasks and expand from there build confidence over time. Teams that try to automate judgment first lose trust in the tool fast. The sequence matters.
Where the Time Savings Are Largest
Small-balance multifamily is where AI-assisted underwriting produces the clearest results. These loans are structurally similar, which makes extraction models highly reliable across large submission volumes. A team processing 40 small-balance deals a month cuts substantial manual intake time before a human reviews a single file. Industry estimates put processing time reductions in standardized, high-volume loan categories at 30 to 50 percent once document extraction is automated.
Large portfolios are the second high-impact area. When a model flags concentrated tenant exposure across a 50-property pool in seconds, the underwriting team works from a complete picture from day one. Mid-process risk discovery is one of the most common reasons deals slow down or get retraded. Catching it at intake has a direct impact on execution speed.
The bottleneck in 2026 is not capital availability – it is processing speed. Every hour removed from mechanical work comes back as time in deal execution.
3 Objections and What Operators Actually Say
The traceability concern is the most common. If the model cannot show its work, it does not get used for that task. That is a configuration decision, not a technology limitation.
The concern about automation touching risk decisions is correct and should remain. AI handles extraction, not approval. When that line is respected, the underwriter’s role does not shrink. It sharpens. Analysts who spent their days on data entry are now reviewing for accuracy and catching errors before they reach committee.
On cost, the math works the same way across fund sizes. A team underwriting three deals in the time it previously took for one pays for the tool quickly. The real friction is not budget. It is redefining how junior analysts measure their contribution. That is a management decision, not a technology one.
The lending teams moving fastest right now are not the ones with the most sophisticated models. They are the ones that removed the most manual friction from their intake process and redeployed that capacity toward execution.
LoanBase’s structured deal data and ingestion pipeline give underwriting teams a cleaner starting point on every submission, so analysts spend their time on credit judgment rather than data recovery.
