Originator guide
The Diligence Process
Draft for editorial review. Technical and final editorial approval are pending. Examples with invented numbers are labelled fictional.
Diligence usually runs as several parallel reviews with different reviewers, different evidence and different failure modes. Treating it as one process with one contact on your side is a reliable way to lose time, because a single coordinator becomes the bottleneck for questions they cannot answer. The five tracks below are a working framework rather than a fixed industry structure, and providers organise their reviews differently. Assign an owner per track, expect the tracks to move at different speeds, and manage the interfaces between them.
Five tracks running at once
Each track answers a different question and produces a different piece of the final structure. Understanding what a track produces tells you where to invest your preparation effort.
Exhibit 1
Diligence tracks, evidence and what each one produces
| Track | What it tests | Evidence you supply | Typical output |
|---|---|---|---|
| Commercial | Whether the product, volumes and growth plan fit the mandate | Business plan, origination forecast, pricing and unit economics | Facility size and structure |
| Credit | Whether underwriting is repeatable and performance is understood | Credit policy, exception log, vintage and static pool data | Advance rate and eligibility drafting |
| Data | Whether your numbers can be reproduced and reported monthly | Loan tape, field dictionary, reconciliations, sample reports | Reporting schedule and data covenants |
| Operational | Whether the book can be serviced, collected and controlled at scale | Procedures, systems list, staffing plan, continuity and transition testing | Servicing terms and backup arrangements |
| Legal | Who owns the receivables and what restricts transfer or servicing | Entity chart, material contracts, consents, licences | Conditions precedent and document architecture |
The outputs matter more than the questions. A credit reviewer who cannot see how cohorts season may draft tighter eligibility rather than simply asking again. An operational reviewer who cannot confirm who produces the monthly reporting pack may push for a backup arrangement you then have to pay for. Weak evidence can stop a process outright where it goes to ownership or reproducibility. More often it shows up as narrower terms or as conditions you have to satisfy before funding.
A bank-funded counterparty brings a predictable shape of questions: it performs its own analysis rather than relying on the seller, wants the information needed to underwrite and monitor the assets, expects transfer, servicing and any recourse arrangements to be documented, and plans to monitor performance after funding.
Sampling: what a file review can and cannot tell you
File review is where preparation meets reality. A reviewer selects loan files and checks whether the documentation, approvals and data match what your policy and tape claim.
In a fictional review, 50 files are selected and 4 are missing a required signed document. That is 8 percent of the sampled files carrying that specific defect. It is not an 8 percent pool default rate, not an 8 percent eligibility failure rate, and not a projection of anything at portfolio level. The distinction matters because samples in diligence are often chosen deliberately rather than randomly: largest exposures, most recent vintages, files with policy exceptions, delinquent accounts. A risk-based sample is designed to find problems, which makes it good at detection. Inference from it to the whole pool is limited, and any inference at all requires a documented selection design and weighting rather than a simple percentage read across.
Agree the selection method in writing before the sample is drawn. If the provider wants a defensible pool-level estimate rather than a targeted hunt, that requires a designed sample with stated selection probabilities and an agreed approach to precision. There is no single correct file count, because the number depends on the defect rate you need to detect and the precision the provider wants. Settling this early prevents a familiar argument, where a targeted sample result gets translated into a portfolio assumption nobody intended.
A sampled defect rate is a statement about the sample
Four missing documents out of 50 deliberately selected files tells you a control failed and how often it failed within that selection. A defensible pool estimate requires a sampling design that supports population inference; knowing the selection rules or attaching weights to a targeted sample is not enough by itself. It also says nothing about how those loans will perform. Ask that the finding is recorded as a documentation defect with a remediation plan, rather than as a pool statistic that migrates into the credit assumptions without anyone deciding it should.
Classify every finding before you argue about it
Findings arrive as a list of complaints unless somebody classifies them. Classification determines the response, and the response determines whether the finding costs you money.
- Documentation defect: the evidence is missing or mislabelled. Often the underlying transaction is sound and the fix is retrieval plus a control change. Do not assume that, though. A missing signature or an unexecuted assignment can be a substantive legal problem rather than a filing failure, so establish with counsel whether the document exists and cannot be found, or was never properly executed.
- Data defect: the extract, definition or mapping is wrong. Fixed by a corrected file and a dictionary entry.
- Process gap: the control does not exist or is not evidenced. Fixed by a procedure and proof it operates.
- Credit exception: a loan sits outside stated policy. Producing the approval record and the authority behind it shows the exception was a decision rather than an oversight, which is necessary but not sufficient. The provider still has to accept the exception, either by treating the loan as eligible or by agreeing a cap on exceptions in the pool.
- Legal or consent issue: something restricts transfer, servicing or use of data. Routed to counsel, not to operations.
The worst outcome is an unclassified list that both sides read differently. Your team treats an item as closed because a file was uploaded, and the provider still carries it as open because the underlying control never changed.
Run one decision log, not five inboxes
One log, numbered, with both owners named and a defined evidence standard for closure. Update it on a fixed rhythm and circulate it before each call so the call is about exceptions rather than status.
Exhibit 2
Fictional decision log entries and how each one closes
| Finding | Classification | Evidence to close | Effect if unresolved |
|---|---|---|---|
| 4 of 50 sampled files missing a signed contract copy | Documentation defect | Retrieved copies plus a control change preventing recurrence | Affected loans may be excluded from eligible collateral, depending on what the retrieval produces |
| Two files underwritten outside stated income verification policy | Credit exception | Approval records showing authority and rationale, plus exception reporting and provider acceptance | Tighter eligibility drafting or a policy covenant |
| Delinquency field blank on paid off loans | Data defect | Corrected extract and a dictionary entry defining null values | Recurring reporting disputes after closing |
| Servicing contract silent on file access for a replacement servicer | Legal and operational gap | Amended contract terms and a tested data extract | Condition precedent, or a blocker if access cannot be agreed |
| Collections procedures undocumented for accounts past 90 days | Process gap | Written procedure plus an agreed period of evidenced operation | Operational covenant with monitoring |
Blockers and conditions are different animals
A blocker is a matter the provider will not fund around in its current state. You cannot demonstrate that the originating entity owns the receivables. The tape cannot be reproduced from source systems. A material contract prohibits assignment and the counterparty will not engage. Blockers are not managed into a closing checklist as they stand. Either a path to resolution is agreed, in which case the item can become a clearly scoped condition, or the process stops.
A condition is a matter capable of resolution by or at funding, documented as a condition precedent with a defined deliverable. Corrected files delivered. Account control arrangements executed. A named servicing deputy appointed. Conditions belong in the closing checklist with owners and dates.
Which category an item falls into is the provider's judgement, exercised through credit approval or a documented risk acceptance, not something you can classify unilaterally. What you can control is the record. Push every open item into one category or the other before you agree a timetable. Items parked in the middle, where the provider says it will get comfortable and nobody writes down what comfort requires, tend to resurface close to signing when there is little time to fix them.
Retesting and the path to first funding
When a fix is agreed, agree the retest at the same moment. Scope, sample size, who performs it and what evidence closes it. A documentation fix usually needs a fresh pull across the affected population rather than a re-check of the original four files, since those four have now had attention no other file received.
Retest results feed the closing checklist, which in turn feeds the conditions in the executed documents. That handoff is where processes lose time, because the person who ran diligence is often not the person assembling the checklist. Keep the decision log numbering intact through documentation so a condition can be traced back to the finding that created it, and read legal documentation alongside the log so the conditions you accept match the evidence you can actually produce.
Running your side of the process
- Name one coordinator and one owner per track, with a deputy for each.
- Agree the sample selection method and file count before selection begins.
- Publish the decision log weekly, including items you consider closed and the evidence that closed them.
- Escalate silence. An unanswered question on your side becomes a term on theirs.
- Keep the room and the log in step, so every closed item points to a document somebody can find.
The evidence base that feeds all five tracks is assembled in preparing your data room, and the closing sequence that picks up from the final decision log is set out in getting to close.