Asset-Based Finance Software: Who Controls the Borrowing Base?

Trace loan-level data into eligibility rules, borrowing bases and usable funding. Compare private-credit platforms by controls, integration and facility complexity.

The borrowing base is the link between data and usable funding

The September ABF manual adds the contractual mechanics missing from a generic private-credit platform comparison. Commitment is a ceiling; the borrowing base is collateral-supported capacity under the agreement. Neither is the same as cash available now. Data validation, eligibility, concentration, advance rates, reserves and draw conditions determine how much can be borrowed.

Reproduce the manual’s example in USD millions. Gross collateral is 500. Ineligible assets remove 40 and concentration deductions remove another 20, leaving 440 eligible. An 80% advance rate gives 352 of support. A 12 reserve leaves a borrowing base of 340. Against a commitment of 400 and existing drawings of 325, drawable headroom is min(340, 400) − 325 = 15. Nominal unused commitment is 75. The 60 difference is not missing cash the software can create; it is contractual capacity without sufficient current collateral support.

The deductions must be mutually consistent: the same exposure cannot be removed once as ineligible and again as an excess concentration unless the agreement calls for that treatment. If another 20 becomes ineligible without secondary effects, the base falls to 420 × 80% − 12 = 324. Drawings of 325 then exceed support by 1. A recognised repayment of 1, or 1.25 of additional fully eligible collateral at an 80% advance rate, cures the arithmetic shortfall. The contract determines whether the cure is accepted and when it is due.

Facility allocation is an economic problem with legal constraints

The source’s allocation case shows why sending assets to the cheapest facility first can be inferior. Facility A accepts only prime assets at an 80% advance; B accepts prime and near-prime at 75%, but near-prime cannot exceed 60% of its pool; C accepts specialty assets at 70%. With prime assets of 100, near-prime of 60 and specialty of 40, a naive allocation funds only 108: 80 in A and 28 in C, with near-prime stranded outside B’s composition rule.

Assigning 60 prime to A, 40 prime plus 60 near-prime to B, and 40 specialty to C produces 48 + 75 + 28 = 151. This maximises funding in the simplified source case, not necessarily economic value after interest, fees, transfer friction and liquidity buffers. A correct platform should show both the objective and binding constraints rather than label every improved number “optimisation.”

Compare providers on the same accepted output

The report’s four-provider framework is retained as a test design, not a universal ranking. Cardo AI’s public materials describe calculation-agent and asset-based-finance workflows. Its 2026 cash-flow announcement adds live-rate-curve integration, a vendor-reported capability to test rather than independent proof of performance. Finley’s documentation describes tape ingestion, validation and borrowing-base calculations linked to funding requests. Setpoint’s Q1 2026 update describes document-to-tape verification and cash reconciliation. These scopes overlap: labels alone cannot establish which firm should win a specific mandate.

The practical comparison is a controlled replay. Give each candidate the same anonymised tape, agreement, amendment, late correction and funding request. Test the accepted calculation, the exception queue and who signs it. Then change one eligibility term and require reproduction of both the old approved result and the new result without rewriting history. A capable architecture retains rule version, source evidence, operator action, approval and downstream funding status.

Responsibility should be explicit. The originator maintains asset data; an agreed party verifies evidence; the contractual agent or lender accepts the calculation under the agreement; operations sends the instruction; treasury monitors cash. A product may automate several activities, but the agreement governs authority. Comparison economics should therefore include the people and controls left behind after purchase. The source’s deployment sizes, scorecards and negotiated-price examples remain scenarios, not market-share or customer-spend facts.

The October follow-up clarifies the deployment model

The 1 October ABF follow-up explicitly treats 40 facilities, 65% software coverage and 26 managed facilities as a constructed planning case. It does not establish an institution’s deployment. Its USD 1.5 million software allowance divided by 40 gives USD 37,500 per facility; a USD 0.5 million targeted-operations allowance gives another USD 12,500. Linear scaling to 26 facilities produces USD 975,000 of software and USD 1.30 million total external cost. Implementation, retained staff and taxes are excluded. No matched-scope public quotations support interpreting the equal software allowance as the actual price of four providers.

The useful update is the hybrid operating model: connect data, calculation, exception management, approvals and reporting while retaining specialist valuation, servicing or accounting where they add value. The follow-up prioritises configurability, data quality, governance, implementation and scalability. Its numeric vendor scores express conditional fit judgment, including untested UI and implementation assumptions; they are not reproduced as objective performance rankings. A shared submission format may be justified where it reduces control friction, but the benefit must outweigh the counterparty’s implementation burden.

The wider private-credit operating model

Private-credit technology works best when accounting, analytical visibility and loan-level control have clearly assigned owners. Buying three platforms can be rational; allowing all three to maintain competing versions of the same loan cannot. The real purchasing question is whether each additional system removes a distinct operational constraint and whether its benefit survives the cost of integration, validation and maintenance.

This edition combines the original private-credit operating-model research, the September 2026 asset-based-finance manual and its 1 October follow-up. Unreconciled institution-specific cost assertions are removed from the current comparison. Explicitly illustrative build-versus-buy and facility examples remain teaching cases.

Three different operating problems

A fund-accounting platform records the economics of the investment vehicle. It maintains the general ledger, partnership allocations, capital calls, distributions, fee calculations and investor balances. Its value lies in producing a controlled financial record from which net asset value and investor reporting can be reconciled. In a credit strategy, it must also accommodate interest, payment-in-kind income, amendments, restructurings and vehicle-specific waterfalls.

A portfolio-analytics platform answers a different question: what is happening across investments? It collects company metrics, capital-account statements, qualitative commentary and exposure information, then normalises them for comparison. A fund accounting record can show the carrying value of an investment without revealing that several apparently unrelated funds share a customer, region or industry exposure. Cross-portfolio analytics is therefore a complement to accounting rather than a substitute for it.

A specialist credit platform follows the contractual mechanics beneath the portfolio total. It tests covenants, calculates borrowing bases, applies eligibility rules, monitors concentration limits and processes granular asset tapes. Its value becomes most visible in asset-backed portfolios, structured credit and leverage facilities where a small data error can change available liquidity or conceal a covenant breach.

Three operating layers and the records they control

The architecture should specify which layer originates each field, who can override it and how an approved correction reaches downstream systems. Otherwise integration becomes the movement of disagreements between databases.

Where integrated suites earn their place

An integrated core can reduce reconciliation between accounting, performance reporting and investor communications. That reduces the number of interfaces which must agree before a close can finish. It also makes it easier to trace an investor statement back to the same approved ledger used by finance and audit.

The historical operating discussion described nine-to-twelve-month implementation periods for broad modules at a large institution. This was a scope-specific assertion, including governance, migration, regulatory mapping and training; it should not be read as a universal deployment time. Its significance is the difference between installing software and creating a controlled production process.

Suite breadth is also an imperfect guide to module quality. A strong accounting engine can have a merely adequate monitoring tool. Procurement should test the actual workflow and exception cases rather than infer uniform depth from a product catalogue. Paying for a specialist layer can be justified if it avoids a material control gap, but not merely because its dashboard looks more modern.

The right implementation sequence begins with the information model: funds, vehicles, legal borrowers, facilities, instruments, currencies, dates and economic ownership. A clean interface cannot repair a disagreement about what one of those objects means. A borrower merger, for example, must preserve both the current legal identity and the historical covenant record. Replacing the name in a single table loses information required to explain old decisions.

Compare cost only after matching responsibility

The inherited workbooks contained an unverified institution-specific software case and inconsistent forward totals. Those figures are excluded from the current procurement comparison. A defensible price bridge separates subscription, implementation, retained operations, independent verification, data and exit costs, and identifies assets, facilities and legal entities covered. A ratio to assets under management is useful only after that service perimeter is fixed.

A facility-control purchase can include software alone or operational responsibilities. A lower license is not necessarily cheaper if the institution must still reconcile asset tapes, interpret legal tests and produce accepted reports manually. The later ABF manual makes this distinction explicit and supplies transparent collateral examples in place of unsupported institutional price benchmarks.

AI changes the cost composition

The internal-build estimates show why cheaper code does not automatically produce cheaper ownership.

Original build-and-operation estimates and workbook expectations

These are the original workbook estimates, not externally validated engineering budgets. They imply lower initial development costs but a larger ongoing consumption bill. The operational burden shifts from writing interfaces toward supervising extraction, maintaining permissions, resolving exceptions and paying for compute and supporting infrastructure.

A simple, undiscounted five-year comparison would total $27 million for the initial build case: $7 million plus five years at $4 million. The later case totals $27.5 million: $5 million plus five years at $4.5 million. That arithmetic is illustrative and excludes timing, migration and residual value. It also does not establish that internal build and the $750,000 purchased package cover identical functionality. Comparing them without a scope bridge would be a serious procurement error.

The relevant question is which internal capability creates an advantage. A proprietary portfolio construction method may justify internal ownership. Maintaining standard accounting allocations, document ingestion and routine investor notices usually creates less strategic differentiation. Selective internal tools can sit above a purchased controlled record without attempting to reproduce the entire administration system.

AI should accelerate discovery before approval

Private-credit documents are particularly difficult inputs. Definitions change between facilities, amendments override earlier clauses, and financial statements can use borrower-specific adjustments. An extraction tool may correctly read a number while assigning it to the wrong covenant or period. Fluent explanation is not evidence that the economic interpretation is sound.

A production process should retain the source passage, extraction confidence, transformation rule and approval status for every consequential field. The test of an assistant is not whether it can answer an easy question about a clean agreement. It is whether it flags conflicting amendments, missing certificates and ambiguous borrower definitions before a decision is made.

The case materials expected AI to add to software expenditure through new functionality and broader deployment. That prediction can coexist with declining cost per extracted page. An institution may process many more pages, cover additional asset types or increase surveillance frequency. Unit productivity and total spending measure different things.

Measure operational value where it occurs

Benefits should be attached to a baseline and a control outcome. Useful measures include elapsed close time, unresolved reconciliations, late covenant certificates, false alerts, borrowing-base adjustments, investor-query turnaround and analyst capacity. A claim that a tool saves hours should distinguish hours eliminated from hours moved to another team.

Borrowing-base work offers a particularly clear example. A system that identifies eligible assets more accurately can release previously unused facility capacity; a system that identifies ineligible assets sooner can prevent overborrowing. Both are valuable, but the first is a funding-efficiency outcome and the second is a risk-control outcome. They should not be added together as if both were guaranteed cash savings.

A practical acceptance test uses a representative portfolio containing amendments, multiple currencies, overdue information, unusual waterfalls and disputed values. Accounting, credit, operations and model-validation staff should independently reconcile the outputs. The target is a repeatable process whose exceptions have owners, not a demonstration with no exceptions.

Renewal and exit need the same discipline

The case workbooks allocated expected incremental spending

The workbooks’ forward allocation of incremental spending

Those figures are a respondent’s forward allocation, not a supplier revenue forecast. They suggest four separate budget approvals rather than one undifferentiated renewal increase.

Each approval needs a different measure of success.

Each budget approval needs its own success measure

Counting every increase as an AI premium would misrepresent the original budget logic.

The allocation also offers a way to challenge duplication. If the same document-extraction capability is sold through the accounting core, analytics layer and credit specialist, procurement should identify which system performs the authoritative extraction and whether the others require separate licences. Overlap can be deliberate for independent checking, but it should be purchased consciously and tested as a control.

Switching an accounting core carries different risks from replacing a search or visualisation layer. Historical ledgers, investor allocations and audit evidence require continuity. Analytical layers can often move more easily if the institution owns its taxonomy and can export the underlying data in a usable format.

Commercial negotiations should therefore separate the core record, optional modules, professional services, usage charges and exit support. Growth-based pricing should specify its denominator and avoid charging twice for the same asset through multiple vehicles. AI charges should identify whether the bill depends on users, documents, tokens, workflows or successful outputs.

The strongest architecture is selective and explicit. It preserves a controlled accounting record, adds specialist credit depth where contract complexity requires it and uses analytics to make the whole portfolio intelligible. The historical cost case supports continued demand for software even as AI reduces some development effort, while its unresolved arithmetic shows why every procurement forecast needs a reconciled model before it becomes a budget.

Primary sources checked for this edition

• Finley — Borrowing-base ingestion, validation and funding workflow documented. Source date: current page checked 2026-10-08. https://docs.finleycms.com/core-capabilities/automated-borrowing-base-calculation
• Cardo AI — ABF cash-flow tool announced with Bloomberg live curves. Source date: 2026 announcement. https://cardoai.com/announcements/cardo-ai-launches-cash-flow-modeling-tool-for-asset-based-finance-with-live-rate-curves-powered-by-bloomberg-data/
• Setpoint — Document-to-tape and cash verification scope. Source date: Q1 2026. https://new.setpoint.io/announcements/what-we-built-in-q1