AI for Asset-Based Lending

1. ABL Is Growing Faster Than Its Back Office

Asset-based lending keeps compounding. The Secured Finance Network's market sizing puts US ABL commitments at $537 billion at the end of 2024, with growth every year since 2018, and its quarterly indexes through 2025 show the non-bank side growing fastest: non-bank outstandings jumped 12.6% in Q4 2025 alone. Private credit noticed. Macfarlanes counts five of the top 30 US private debt managers launching their first dedicated ABL funds in a single year, and a majority of surveyed investors prioritizing asset-backed strategies.

Every new ABL dollar carries an operational tail that cash-flow lending does not: the collateral has to be counted, tested, and re-margined continuously for the life of the facility.

Growth in commitments without growth in monitoring capacity means one of two things gives: the headcount budget, or the rigor. AI is the third option.

2. The Borrowing Base Is a Document Pipeline

Strip ABL to its weekly rhythm. The borrower submits a borrowing-base certificate: gross AR, minus ineligibles, times the advance rate, plus eligible inventory at the lesser of cost or market, times its advance rate, minus reserves, equals availability. Attached come the AR aging, the AP aging, inventory reports, and sometimes perpetual inventory detail.

At most lenders a portfolio analyst opens each package, spot-checks the borrower's arithmetic, eyeballs the agings, and books the certificate. On a desk with 60 borrowers reporting weekly or monthly, that is thousands of packages a year, reviewed at whatever depth the calendar allows. Depth flexes with workload, which means the thin reviews cluster in busy weeks, which is when problems like to arrive.

Each certificate is a small, structured, rules-based document set. That is the precise shape of work AI does perfectly, every time, at any volume.

3. What AI Can and Cannot Do

The boundary, stated plainly.

AI can recompute. Rebuild the borrowing base from the underlying agings and the actual eligibility rules in the credit agreement, and reconcile against the borrower's certificate line by line.

AI can test. Apply cross-aging, concentration, dilution, and foreign-account rules mechanically, and flag the accounts that flip status this period.

AI can watch. Track dilution, aging migration, inventory mix, and availability trends across every borrower, and rank the book by deterioration.

AI cannot set the structure. Advance rates, reserve levels, and the judgment calls on gray-zone eligibility (the disputed account, the related-party receivable) are credit decisions. They price risk the rules cannot see.

ABL's advantage over most credit niches: the rules are already written down in the credit agreement. There is less ambiguity for AI to mishandle, and more pure recomputation to win.

4. Borrowing-Base Certificates, Processed

The target state is simple to describe. A certificate package arrives by email or portal. An agent extracts the certificate and its support, recomputes availability from your rules (not the borrower's summary), ties every certificate line to the underlying reports, and posts the result to the loan system. Clean packages flow through with an audit trail. Discrepancies route to the analyst with the difference quantified and located.

Notice what the analyst's job becomes: instead of checking arithmetic on sixty packages, they investigate the four where the borrower's number and the recomputed number disagree. The work shifts from processing to exception handling, which is the work analysts were hired for.

Two design rules from hard experience. The recomputation must come from the credit agreement's actual definitions, encoded facility by facility, because eligibility language varies and the variance is the point. And extraction confidence must be explicit: a blurry scanned aging gets routed to a human, never silently guessed.

Reading those definitions out of the credit agreement is itself AI-assisted work, covered in our covenant review guide.

5. Ineligibles and the Rules Engine

Ineligibles are where borrowing bases go quietly wrong. Past-due accounts over 90 days. Cross-aging: when 50% of a customer's balance is past due, the whole customer goes ineligible. Concentration caps. Contra accounts where the borrower also owes the customer. Foreign receivables, government receivables, affiliates.

Borrowers compute these themselves, and the errors are not random: they skew toward availability. Sometimes it is honest spreadsheet error, sometimes wishful classification, occasionally fraud. The famous ABL frauds ran through the borrowing base for years because nobody recomputed from source.

A rules engine recomputes every ineligible category from the raw aging every period. It catches the cross-age the borrower missed, the concentration breach created by one growing customer, the contra that appeared when the borrower started buying from its own client.

The gray zone stays human: whether a disputed account is collectible, whether to carve out a strategic customer from concentration limits. The engine flags; the credit officer rules.

6. Collateral Trends: The Early Warning System

A single certificate tells you availability. The sequence tells you the future. Dilution creeping from 3% to 5% over two quarters says customers are disputing invoices or taking deductions, which often precedes revenue problems. Aging migrating rightward says collections are slowing before the borrower says anything. Inventory shifting from finished goods to raw materials can mean production problems; the reverse can mean sales problems.

These patterns are nearly invisible certificate by certificate and obvious in a trend line. An AI monitoring layer maintains the trend lines automatically, across every borrower and every collateral metric, and surfaces the moves that matter: "Dilution up 40% over three periods at Borrower X, driven by two customers, here are the invoices."

In ABL the collateral often deteriorates before the financial covenants trip. The borrowing base is your highest-frequency data feed on borrower health; treating it as paperwork wastes the best early-warning signal in the credit. The book-wide discipline is the same one in our portfolio monitoring guide, running on weekly data instead of quarterly.

7. Field Exams and the Verification Loop

Field exams stay. AI does not verify that the inventory exists; an examiner walking the warehouse does. What AI changes is what the examiner walks in knowing.

Exam prep today is gathering: pulling agings, prior exam reports, certificate history, and building test selections. An agent assembles that package in minutes and, better, makes the selections smarter: the accounts with unusual dilution, the inventory categories that moved oddly, the customers whose payment behavior changed. The exam tests where the data says to look, instead of sampling evenly across a book that is not evenly risky.

Exam findings then close the loop: ineligible classifications get corrected in the rules engine, reserve recommendations land in the file, and the next exam's baseline is already structured. The cadence stays annual or semiannual; the surveillance between exams becomes continuous.

8. The Tools

ABL has mature servicing platforms and a young AI layer. The stack usually combines both.

The servicing platforms hold the records but mostly trust the borrower's math. The recomputation-from-source layer, tuned to your eligibility definitions, is the piece that usually has to be built, and it is where the risk reduction actually lives.

9. The Human Line: Advance Rates Are Judgment

Everything above makes the collateral picture sharper. None of it decides what to lend against it.

Advance rates and reserves price the unknowable: how the collateral liquidates in a downside nobody has seen yet, how this borrower behaves under stress, what the exam could not test. Those calls belong to credit officers, and the AI's job is to make sure they are made with the true aging, the true dilution trend, and the true ineligible math in front of them.

And fraud deserves its own sentence. AI raises the cost of borrowing-base fraud substantially (recomputation from source defeats summary-level manipulation), but a determined fraud fabricates the source documents too. Field exams, customer verification, and skepticism remain load-bearing.

Borrower AR data is commercially sensitive, so the standing rule applies: no-training infrastructure, your environment, full audit trail. Details in our Security and Data Governance guide.

10. Where to Start

A practical sequence for an ABL or specialty-finance desk.

First. Pilot certificate recomputation on your ten largest facilities. Encode their eligibility definitions, recompute three months of history, and count the discrepancies you find. That number usually settles the business case by itself.

Second. Turn on trend monitoring (dilution, aging migration, availability usage) across the book, with a ranked weekly watchlist.

Third. Wire the output into exam prep and the credit file, so surveillance, exams, and annual reviews share one collateral record.

A Discovery Sprint maps your borrowing-base workflow, measures where the hours and the risk concentrate, and scopes the recomputation engine against your actual credit agreements.