The monitoring load is rising faster than the deal flow
The Secured Finance Network's Q1 2026 index, published on July 1, described a quiet quarter for new business and a busy one for existing books. Total commitments were unchanged for banks and up about 1% for non-banks, while new outstandings fell 26.0% for banks and 70.4% for non-banks. Utilization rates rose for both groups as borrowers drew more heavily on facilities they already had. Lender sentiment split: bank confidence fell seven points to 55, and non-bank confidence rose nine points to 67, its highest reading in more than three years.
That combination is specific and it matters operationally. Fewer new deals means less closing fee income and fewer new files. Higher utilization on existing facilities means more availability being used, thinner cushions, and more sensitivity to a mis-stated borrowing base. SFNet's year-end 2025 survey had already noted criticized loans and non-accruals rising while write-offs stayed stable, which is the profile of a book that needs closer watching rather than one that is failing.
So the work goes up while the thing that pays for the work goes sideways. Portfolio managers cover more credits, field exam calendars get stretched, and the certificate that arrives on Monday gets a lighter look than it did a year ago. That is the pressure automation is being asked to relieve, and it is worth being precise about which parts of the job it can take.
The reporting rhythm, and what each layer is actually answering
Asset-based lending monitoring is not one activity. It is four or five overlapping cycles running at different frequencies, each answering a different question. Confusing them is how a lender ends up with a well-administered calendar and a badly understood collateral position.
| Cadence | What arrives | The question it answers |
|---|---|---|
| Daily | Cash receipts through the lockbox or blocked account, collections applied, new sales and credit memos | Is availability moving the way the borrower said it would? |
| Weekly or monthly | Borrowing base certificate, accounts receivable aging, accounts payable aging, inventory report | How much is the borrower entitled to draw today? |
| Monthly or quarterly | Financial statements, covenant compliance certificate, rolling fixed charge coverage | Is the underlying business still performing? |
| Semi-annual, annual, or triggered | Field examination and its workpapers | Do the reported numbers tie back to the borrower's own books? |
| Annual or triggered | Inventory and machinery appraisals, net orderly liquidation value | What is the collateral actually worth if the facility has to be repaid from it? |
The layers check each other. The certificate is a borrower assertion. The field exam is the test of that assertion. The appraisal is the test of the value the exam assumed. Daily cash application is the continuity check that tells you whether the last certificate is still approximately true. Remove any one layer and the others keep producing numbers that look fine.
The OCC's Comptroller's Handbook on asset-based lending makes the same point in supervisory language: administering and monitoring these facilities is time and cost intensive, and the product is particularly vulnerable to borrower fraud, especially when the business comes under strain. That vulnerability is not a reason to avoid the product. It is the reason the checking has to actually happen rather than being nominally scheduled.
Four places the borrowing base goes wrong
Most availability errors are not fraud. They are the same handful of mechanical failures, repeated quietly over months. Each is a good test of whether a monitoring tool is doing arithmetic or doing the job.
Ineligibles applied loosely
Past due accounts, cross-aging, concentration caps, contras where the customer is also a supplier, related-party balances, foreign accounts, bill-and-hold, consignment, and progress billings. A common supervisory benchmark treats an account as ineligible once it is past due by three times its terms, so 90 days on 30-day terms, and most agreements then deem the rest of that customer's balance ineligible too.
Dilution measured on a convenient window
Dilution is the gap between what was invoiced and what was actually collected: credit memos, returns, discounts, allowances, write-offs. It drives the advance rate cushion, so measuring it over a short or flattering period quietly raises availability without anyone deciding to raise it.
Reserves that are set once and never revisited
Rent reserves for locations without a landlord waiver, tax and payroll reserves, work-in-process, customer deposits. These are established at closing against a picture of the business that changes, and they are the first thing to go stale on a facility nobody has re-examined in a year.
Timing and roll-forward gaps
The certificate is as of a date, cash applies continuously, and the roll-forward from one certificate to the next has to reconcile to the general ledger. When sales, collections, and credits do not bridge the two balances, the difference is the finding, and it is invisible if each certificate is only read on its own.
None of those require intuition to detect. They require someone to recompute the base from the underlying aging rather than accept the summary line, and to compare this week's computation with the last twenty. That is exactly the shape of work that scales badly with people and well with software.
What software should not do is decide the answer. Whether a concentration cap should be waived for a particular customer, or whether slow-moving inventory is genuinely salable, is a credit judgment. The distinction runs through the rest of this article.
The field exam, split into mechanical work and judgment
A field exam is where the borrower's reporting meets the borrower's books. It is also the most expensive recurring item in an ABL relationship, usually billed to the borrower, and the one most often deferred when calendars get tight. Splitting it into what a machine can prepare and what a human has to conclude is the practical way to think about automating it.
Tie-out of the aging to the general ledger and the certificate
Does the accounts receivable aging foot, does it agree to the trial balance, and does the certificate agree to the aging it claims to summarise? This is arithmetic and reconciliation across three documents that a system can perform completely rather than on a sample, and it is where a surprising share of findings originate. Fully mechanical.
Recalculation of the ineligibles the borrower applied
Re-run the eligibility rules from the credit agreement against the raw aging: past due tests, cross-aging, concentration limits, contras, foreign and related-party accounts. A system can apply the whole rule set to every account instead of testing a sample. The edges, an unusual customer relationship or a disputed balance, still need a person. Mechanical with judgment at the margin.
Invoice sampling, proof of delivery, and cash application testing
Pull the sample, match invoices to shipping documents and to the cash that eventually arrived, and flag the ones that do not reconcile. Selection and matching are automatable; deciding what an unmatched item means is not. This is also where invoice fraud patterns first show as inconsistencies rather than as anything dramatic. Mechanical preparation, human conclusion.
Dilution, turnover, and roll-forward analysis over the review period
Compute dilution consistently over a defined window, calculate turnover, and bridge each certificate to the next through sales, collections, and credits. Doing this across every period rather than a selected few is where trend detection actually comes from. Fully mechanical.
Quality of earnings, terms of sale, salability, and the reserve recommendation
What the customer concentration means for this borrower, whether terms of sale have quietly loosened, whether the inventory would move in a liquidation, what reserve the file now needs, and whether the advance rate still fits. This is the examiner's actual product and it should stay with the examiner. Automation earns its place by giving them the first four items already done.
What stays human, and what a regulator expects to see
Nothing here should be read as removing the field examiner or the portfolio manager. It removes the assembly work that consumes their week and leaves the conclusions with them, better evidenced. The supervisory expectations for asset-based lending are not satisfied by a tool having run. They are satisfied by a lender being able to show what was tested, on what, and what was decided.
- Every figure traced to its source. A recomputed availability number should link back to the aging line, the invoice, and the page it came from. Without that, verifying the machine's number costs as much as producing it by hand.
- Overrides retained with reasons. When a credit officer waives a concentration, releases a reserve, or accepts an over-advance, the prior value, the new value, the reason, the approver, and the timestamp all have to survive. Over-advances and reliance on illiquid collateral are exactly what examiners look at.
- Rules that match the credit agreement, not a template. Eligibility definitions vary facility by facility. A monitoring system applying a house default rather than this agreement's language is producing a number that does not correspond to anything contractual.
- The certificate and the recomputation both preserved. When the borrower's stated availability and the lender's calculation differ, keep both, identify what drives the gap, and record which one the lender is relying on. Storing only the borrower's figure records an assertion rather than a test.
- Vendor governance documented. Which model version processed a file, where documents are handled, and whether your borrowers' data trains anyone's models. The questions in SOC 2 Type II for commercial lending AI cover the half a security report does not answer.
Where Uptiq fits
This is the shape of work Uptiq's agents are built for. Borrower reporting is ingested as it arrives, whether it is a clean export or a scanned aging, normalised to the facility's own definitions, and the borrowing base is recomputed from the underlying detail rather than accepted from the summary. Movements in ineligibles, dilution, concentration, and reserves are tracked across periods instead of read one certificate at a time, every figure carries a citation back to the page it came from, and each override is retained with its reason and approver. The same engine that spreads the financials runs the covenant tests, and it sits alongside the existing servicing and origination systems through 100+ integrations rather than replacing them.
How to sequence this without stalling
Asset-based lending teams are small and busy, and a programme that tries to automate the whole monitoring stack at once turns into a systems project. Take one layer, prove it against files you already know the answers to, and expand.
Start with the certificate, not the field exam
The borrowing base certificate is the highest-frequency artifact and the one whose errors compound fastest. Automating its ingestion and independent recomputation gives you a weekly signal from the first month, where field exam automation only pays back on the exam calendar.
Backtest against exams you have already completed
Run the last four field exams through the system and ask whether it would have surfaced the findings the examiner actually raised. This is a far better test than a vendor demo, because you know the answer and you know which findings were subtle.
Encode one facility's rules properly before doing twenty
Take a single credit agreement, encode its eligibility definitions, reserves, and advance rates precisely, and confirm the output matches what your team would have calculated. Rule fidelity on one facility is worth more than partial coverage across the portfolio.
Give it the reporting you actually receive
Not a clean CSV. The photographed aging, the spreadsheet with a manually inserted subtotal row, the inventory report that changed format when the borrower switched systems. Monitoring automation lives or dies on messy recurring inputs.
Baseline first, then measure the right things
Before anything changes, capture hours per certificate cycle, elapsed days from receipt to approved availability, field exam preparation hours, and how often a finding was first raised at exam rather than in between. Afterwards, watch that last measure most closely. Catching things between exams is the actual point.
The adjacent pieces in this cluster are covenant monitoring best practices for the obligations running alongside the collateral, what financial spreading software does for the engine underneath, and AI agents for commercial lending workflows for how the stages connect.
Frequently asked questions
What can AI actually automate in asset-based lending monitoring?
The mechanical layer: ingesting borrowing base certificates, agings, and inventory reports whatever format they arrive in, normalising them to the facility's definitions, recomputing eligible collateral and availability from the underlying detail, testing dilution and roll-forward across periods, and flagging exceptions and trends. For field exams it can prepare the tie-outs, the ineligible recalculation, and the sampling. The reserve recommendation, the salability call, and the advance rate decision stay with the examiner and the credit officer.
Does automation replace the field exam?
No, and a lender should be sceptical of anything sold that way. A field exam is a physical and investigative exercise as much as an analytical one, and its value is in the examiner's conclusions about quality of earnings, terms of sale, collateral salability, and the reserves the file needs. What automation changes is how much of the exam week goes into tie-outs and recalculation that could have been done continuously, and how much goes into judgment.
What are the most common borrowing base errors?
Ineligibles applied inconsistently, particularly cross-aging and concentration caps; dilution measured over a window that flatters the advance rate; reserves set at closing and never revisited as the business changes; and roll-forward gaps where sales, collections, and credits do not bridge one certificate to the next. Most are not fraud, they are drift, and they are found by recomputing rather than by reading.
How often should a field exam be performed?
It depends on the facility, the borrower's condition, and your credit policy rather than a single standard. A common approach is to set the base frequency by risk and define performance triggers that increase it, so deteriorating trends pull the next exam forward rather than leaving it on the calendar. Fully monitored facilities are typically examined more often than hybrid or lightly monitored ones.
Can a monitoring system detect collateral fraud?
It can surface the inconsistencies that fraud tends to produce: invoices without matching delivery evidence, cash that never arrives against specific customers, roll-forwards that will not reconcile, unusual credit memo patterns near period end. Verification and fraud controls belong inline in the monitoring pipeline for that reason. What follows from a flag is an investigation, not a conclusion, and supervisory guidance is clear that this product carries elevated fraud risk when a borrower is under strain.
Do we need this if our ABL book is small?
Possibly not. A disciplined analyst with a good spreadsheet can hold a handful of facilities with standard eligibility definitions. The economics change when the definitions vary facility to facility, when reporting arrives in formats nobody controls, when utilization is high enough that availability errors matter quickly, or when findings are consistently first raised at exam rather than in between.
Market figures are drawn from the Secured Finance Network's Q1 2026 Asset-Based Lending and Confidence Indexes and its 2025 year-end survey, and are subject to revision. Supervisory descriptions reflect publicly available guidance including the OCC's Comptroller's Handbook on asset-based lending as of September 2026. Eligibility definitions, advance rates, reserve policy, and field exam frequency are set by your own credit policy and the individual credit agreement. Nothing here is legal, accounting, or supervisory advice.
Send us last quarter's borrowing base certificates
With the agings behind them. We will recompute availability from the underlying detail, show the movement in ineligibles and dilution across the period, and cite every figure to the page it came from.
