Credit policy and precedent search, explained.

What a credit analyst looks up mid-memo: the written credit policy, its exceptions and the archive of past decisions

One acquisition loan, from banker's ask to Thursday's committee

1. A relationship manager brings a dealHollis & Crane wants $12,000,000 to fund an acquisition and its working capital. Committee meets Thursday.
2. An analyst lays out the financials and starts the memoEBITDA of the two companies combined, $4,230,000, against $14,500,000 of debt at closing.
3. The analyst checks the policy and the bank's past decisionsDo synergies count? Is 3.43x an exception? Who approves it? This is where an unlabeled exception is caught or missed.
4. A senior credit officer reviews and approves the exceptionWithin the authority the policy gives that officer: up to 0.50x over the leverage limit in this bank.
5. Committee decides; the exception is tracked for lifeReported to the board in aggregate, and read again at every annual review.

Step 3 is the subject of this guide. The bank, people, companies and figures in the example are fictional.

Exceptions are expected: what surprises new analysts

Credit policy, policy exception, precedent, leverage

Fourteen terms are defined in the one-page summary.

The five questions a credit analyst asks, and the documents that answer each one

In shortAn analyst's questions fall into five kinds: the rule, the bank's practice, the approver, the reviewer's habits and the test on the deal's own numbers. Each draws on a different shelf. The policy holds outer limits, the guides hold the method, the memos hold practice and outcomes, and the reviewers' comments hold what seniors care about. A search that covers only the policy answers one question in five.

Supervisors describe the layers precisely. In the OCC's 2026 lending booklet, policies "primarily impose standards" and are "a statement of the bank's credit culture and risk appetite"; procedures "describe how policies are put into practice", and "Policies often include outer limits, with procedures or guidelines including more granular instructions or inner targets".4 The same booklet expects an effective policy to describe "types of acceptable exceptions" and to include "guidelines for identifying, justifying, approving, and reporting exceptions".4

A number in the policy is a limit; a number in a guide is a target

The difference matters most to a new analyst. A number in the policy is a limit: going past it needs an exception approved by someone the policy names. A number in an underwriting guide is usually a target or a method: missing it needs an explanation in the memo. Treat a guide's target as a policy limit and an unnecessary exception goes to a senior, which costs a day. Treat a policy limit as a guide's target and a real exception reaches committee without a label, which is the failure examiners look for.

Seven changes between December 2025 and October 2026 that touch a commercial credit policy

In shortNo federal rule writes a bank's credit policy. Regulators require one, approved by the board, and examine whether the bank follows it. In ten months the texts behind it moved: leveraged lending guidance rescinded by two agencies, model risk guidance replaced, a new OCC lending booklet, a revised concentrations booklet, a narrower Regulation B, a new SBA procedure, and proposed third-party guidance.

What the rescission means for an analyst's question

The agencies called the old guidance "overly restrictive" and replaced it with principles, including that a bank should analyze "a leveraged borrower's past and current performance compared with projections".3 Two consequences reach the analyst's desk. First, the leverage limits, how acquisition synergies count and when an exception is needed are now entirely the bank's own words. Second, the comparison of projections with results that the agencies ask for is what a precedent shows once its later performance sits beside it.

What a bank loan policy has to state, and the supervisory ceilings above it

In shortA loan policy states the limits the bank will enforce, the lending authority of each officer and committee, and how exceptions are made and reported. For real estate it sits under federal loan-to-value ceilings, which cap the bank's own limits but are no promise that a loan at the ceiling is sound. The board approves real estate policies every year.

The OCC lists the topics most lending policies cover: lending authorities, "who is authorized to approve loans, make changes in terms, and grant exceptions to policy"; limits and diversification standards; the types of loans offered; underwriting standards; credit administration; credit risk review; problem loans; reporting; laws and regulations; accounting; and exceptions.4 The FDIC's list is longer and adds "Lending authority of each loan officer" and the trade area.5 None of them fills in a number. The OCC warns examiners to "be skeptical" of "off the shelf" or boilerplate policies.4

The federal loan-to-value ceilings

The guidelines expect each bank to "establish their own internal loan-to-value limits", which "should not exceed" these. They add that the limits "should not be interpreted to mean that loans at these levels will automatically be considered sound".14 Owner-occupied 1-to-4 family and home equity loans carry no supervisory limit.14

Above the bank's own limit, or above the federal ceiling: two sets of rules

How often the policy changes

Real estate lending policies must be "reviewed and approved by the FDIC-supervised institution's board of directors at least annually".14 The OCC says boards "typically review and approve policies annually or when there are material changes", and that a "high volume of exceptions indicates that many loan decisions are being made outside the policy".4 Each release creates a new version of every limit an analyst quotes. Quoting last year's version is one of the commonest errors in section 08.

How to document a loan policy exception: the limit, the gap, the approver, the mitigants, the precedent

In shortThe OCC's 2026 lending booklet sorts exceptions into three kinds: underwriting exceptions granted at approval, financial exceptions found during the life of a loan, and credit administration exceptions such as missing documents. The one an analyst writes up is the first. A good write-up states five things, puts a label on page one of the memo, and names the approver the policy names.

The OCC notes that banks may use other names for these kinds. It adds that "Examiners should not criticize a loan solely because it has an exception", and that exceptions which weaken the bank's ability to protect its interest in a loan "are critical".4

The five things an approvable write-up states

The arithmetic of one exception, worked

In the example bank, Kestrel Precision Machining wants to buy the 46,000 square foot plant it has leased since 2014. The price is $3,400,000 and the appraisal $3,450,000, so the base is the price. The request is $2,720,000, 80%.

Fictional example. Global debt-service coverage is $822,000 of cash flow over $568,000 of payments ($258,000 on the new loan and $310,000 on equipment notes), 1.45x against a 1.25x minimum.

Who approves: the authority ladder

Regulators ask for a "designated authority" and leave the title to the bank.1 The OCC describes laddered authority in which "individuals have escalating levels of approval based on the size of the loan, exceptions to policy, borrower history, risk rating, and other factors", and notes that an independent approver "such as a senior credit officer ... whose primary goal is credit quality can introduce more objectivity".4

The example bank's ladder for leverage is typical in shape. For a credit rated grade 5 on the bank's risk scale, leverage up to 3.25x needs no exception. Up to 0.50x over, to 3.75x, a senior credit officer may approve. Beyond that, credit committee decides.

Working the leverage test on an acquisition, with and without the borrower's synergies

In shortLeverage is funded debt divided by EBITDA. On an acquisition, the borrower adds expected cost savings to EBITDA, and the ratio improves before anything has happened. A careful policy lets documented synergies count up to a cap and also asks for the ratio with none. The gap between the two is where most acquisition exceptions sit.

The bank's policy asks for the last reading too: "Every acquisition memo shall also show leverage with no synergies", and above the limit the credit "is a policy exception ... whatever the adjusted figure". So the honest answer to "do the synergies count?" is "partly, and the deal is still an exception".

Fixed-charge coverage, the other test, asks whether cash flow left after capital spending, taxes and owner distributions covers the year's principal and interest. It is worked without synergies: $4,230,000 less unfinanced capital spending of $480,000, cash taxes of $360,000 and distributions of $90,000 leaves $3,300,000, against $2,521,300 of principal and interest, 1.31x against a 1.25x minimum.

Why lenders discount projected synergies

The caution has support in the supervisory texts. The OCC's 2026 lending booklet lists an "Elevated leverage ratio" and repayment "highly dependent on projected" cash flows among the structural weaknesses examiners look for in underwriting.4 Synergies are projected cash flow by definition: savings the two companies expect to make once they are combined. The example bank learned the same lesson from its own file, when only $80,000 of Bramwell's $350,000 of planned synergies arrived.

Other public guidance points the same way. SBA's SOP 50 10 8.1, in force since October 1, 2026, asks 7(a) lenders for a quality-of-earnings report on business acquisitions at a purchase price of $3 million or more.13 A policy cap on credited synergies, and a required figure with none, turn the OCC and FDIC's request to compare performance "with projections" (section 03) into a rule an analyst can apply at approval.3

Credit precedent search: five questions to ask of any past approved exception

In shortA precedent tells you that an argument persuaded an approver under a particular version of the policy, and what followed. It binds no later approver, and the policy may have changed because of how it turned out. Read every past decision for its version, its approver, its mitigants, its later performance and whether it is one of a pattern. Include the declines.

1. Which policy version was it decided under? A decision under v11 is evidence about v11. In the example bank the cap on synergies arrived in v13 and the required figure with no synergies in v14.
2. Who approved it, and with what authority? An exception granted by committee at 0.70x over the limit is no guide for a senior officer whose authority stops at 0.50x.
3. What carried it? Read the mitigants and the reviewer's words. A request with the same ratio and none of the same support is a different request.
4. How has it performed since? Annual reviews and grade changes hold the answer, and most memo archives never link them to the memo.
5. Is it alone, or one of a pattern? One exception is a judgment. Twenty on the same point are a change in appetite that nobody wrote down.

The bank's own practice on one rule

The two pictures read differently. On owner-occupied plants the practice is narrow and steady: six exceptions in seven years, all in the same band, on the same mitigants, all performing. That is a policy working as designed.

On acquisitions, nine of 29 approvals went above the limit, and three loans were downgraded. That is the kind of pattern the OCC tells a board to look at: "If the volume of exceptions is high or the performance of pools of loans with exceptions is poor", the board may "revise the lending policy" or limit the exceptions it permits.4 The example bank did that, with the 10% cap.

Same ratio, opposite outcomes

Fictional example bank. A senior's comment on the Ashgrove memo became the v14 rule: "Show me leverage with zero synergies. If it's over policy at zero, call it an exception and say why we live with it."

Postmortems turn a bad loan into a rule

The OCC describes postmortem reviews "to understand how problem loans and losses develop", which "can inform strategic execution and changes in risk selection". Its list includes "Comparing each loan's terms and characteristics with underwriting guidelines" and "Assessing the effectiveness of the loan and covenant structure".4 A precedent search that shows the postmortem's finding beside the old memo hands that lesson to the next analyst who faces the same question.

Nine policy-answer errors that reach credit committee, and what catches each one

In shortMost errors come from quoting the wrong thing: an old version, the wrong base, a guide in place of the policy, a misread precedent or the wrong approver. They are easy to make in an analyst's first year and cheap to catch with the source open. The costly one is the exception nobody labeled, which examiners find later.

Unidentified exceptions, in the regulator's words

When credit risk review, audit or examiners find an exception that staff missed, it is an unidentified exception. The OCC says isolated ones "are not necessarily a cause for concern", but "A large volume of unidentified exceptions ... could be indicative of less than satisfactory exception processes and generally merits examiner scrutiny and determination of the root cause". It names the causes: "errors, miscalculations, breakdowns in the pre- or post-closing review process, or system failures".4 Each of the nine errors above is a way to produce one.

What inconsistent exceptions risk: exam ratings, fair lending findings and leveraged credit

In shortExaminers rate how a bank identifies, approves, tracks and reports exceptions. Fair lending reviews of commercial lending look first at vague criteria and undocumented exceptions, and the 2026 change to Regulation B left disparate treatment untouched. And the largest pool of criticized bank credit is leveraged, which is where acquisition exceptions live.

How examiners grade policy exception tracking

The weakest column in that table reads: "Policy exceptions may not receive appropriate approval, significant policy exceptions may be approved but not reported individually or in aggregate, or their effect on portfolio quality is not analyzed".20

The same booklet calls it "prudent for the bank to consider aging of all exceptions with sufficient stratification to identify trends in volumes, loan officer, and types".20 Industry guidance adds a habit and a question: rank exceptions by severity, with more senior approval for the serious ones, and ask whether loans granted exceptions performed as well as comparable loans without them.21

The FDIC tells its examiners to work out "whether troublesome loans result from inadequate lending and collection policies and practices or merely reflect exceptions to basically sound credit policies".5

Fair lending: exceptions are where discretion shows

The interagency fair lending examination procedures tell examiners reviewing commercial lending to limit the risk review to a short list of factors, including U4 to U8.22 Three of those concern exceptions: "Lack of clear guidance on making exceptions to underwriting criteria", "Lack of clear loan file documentation regarding reasons for any exceptions", and "Relatively high percentages of either exceptions to underwriting criteria or overrides".22 The procedures explain why: disparate treatment is likelier for applicants who are "neither clearly well-qualified nor clearly unqualified", because "there is more room and need for lender discretion".22

Leveraged credit carries most of the criticized exposure

The Shared National Credit review covers loans of $100 million and more shared among banks, larger than most regional acquisition loans. Criticized credits are those rated special mention, with "potential weaknesses that deserve management's close attention", or classified, with weaknesses that put repayment at risk.25 The review shows where risk gathers: in borrowers with high debt for their earnings. With the leveraged lending guidance rescinded by two agencies, the bank's own leverage limits, its rules on synergies, and its record of exceptions are the controls left.3

Why the answers sit with a few senior officers as commercial credit training programs close

In shortFormal credit training at large banks has mostly closed, analysts move on within a few years, and the lenders who learned the old way are reaching retirement. The bank's practice is written in its memos and reviewers' comments, but finding it still runs through people. Research on loan officers shows what that knowledge is worth when the person is away.

What the training programs used to do

An industry research firm wrote in 2022 that large-bank credit training programs of the 1980s and 1990s "have, for the most part, been shutting down since the beginning of the 2000s", that the lenders and underwriters they trained are "reaching retirement age", and that replacing them "has proven difficult".2

A 2018 sponsored article in Bank Director listed "the evaporation of commercial credit training" among the reasons analysts are hard to hire, and said banks are "lucky if they can keep an analyst in the role for two or three years".29 ProSight, the association formed in 2024 from the merger of RMA and BAI, sells its Commercial Lending School to banks because "few are able to commit the resources needed for a traditional credit training program".30 The OCC still expects training to cover "policies and procedures" and "financial analysis".4

What the knowledge is worth

Economists call the part of lending that never reaches a form soft information. Federal Reserve economists Allen Berger and Gregory Udell wrote that "the loan officer is the repository of this soft information".31 When that person is away, the cost shows. Studying loan officers on leave, Alejandro Drexler and Antoinette Schoar found their borrowers "less likely to receive new loans from the bank, are more likely to apply for credit from other banks, and are more likely to miss payments or go into default".32 The effect was smaller where officers had reasons to pass the information on.

A senior's reviewer comments, attached to the deals they were written on and shown to the next analyst who faces the same question, are one way to make that hand-over permanent.

AI answers about credit policy: what they get right, how they go wrong, and what the 2026 guidance leaves to you

In shortSoftware can find the section, rank past memos, quote reviewers and do the arithmetic in seconds. Generative AI can also produce a confident answer that cites the wrong version or a passage that does not support it. No federal model guidance covers generative AI since April 2026, so testing and controls are the bank's own design. Decisions stay with the people the policy names.

Where supervisors stand on AI answers to policy questions

The OCC's Spring 2026 risk report found banks "taking a measured approach to the adoption" of generative AI and AI that takes actions on its own, "with usage generally limited to specific use cases with guardrails and human-in-the-loop accountability to manage risk", mostly "productivity and customer experience enhancement tools".33 The Federal Reserve's Vice Chair for Supervision said in July 2026 that AI used "in a way that has more direct impact on credit decisions affecting individual customers" presents "more substantial legal compliance challenges than other AI use cases".34

The revised model risk guidance places generative AI, and AI that acts on its own, outside its scope because such models are "novel and rapidly evolving", and says that a bank's own risk management "should guide the determination of appropriate governance and controls" for such tools.6 Buying the tool does not move the responsibility.18

How a cited answer can still be wrong

NIST's generative AI profile calls the failure confabulation: "The production of confidently stated but erroneous or false content". It warns that outputs "may also include confabulated logic or citations that purport to justify or explain the system's answer, which may further mislead humans into inappropriately trusting the system's output".35 The Treasury's 2024 report uses the plain term, "AI hallucinations".36

The nearest published test is in law, where citing the wrong authority is the same problem. A Stanford team ran 202 preregistered questions through three retrieval-based legal research tools and a general chatbot, and counted an answer as wrong if it was false or "misgrounded": cited, but misreading the source or citing "an inapplicable source".37

Two lessons carry over to credit. Ask any vendor for all three numbers on a test set you wrote, because a tool can keep its error rate down by refusing to answer. And decide in advance which failure you prefer: in credit, "the policy does not say" is a smaller risk than a confident answer an analyst pastes into a memo.

The newest workers gain the most from an AI assistant

The support-worker study is outside banking and no study of credit analysts exists. Its authors' explanation is the one a head of credit analysis wants: the tool "disseminates the best practices of more able workers".38 A Census Bureau working paper found banks using more AI lent more to distant borrowers "about whom they have less soft information".39

What a policy search tool cannot decide

• The approval stays with the "designated authority" the policy names.1
• A silent policy stays silent until its owner writes the rule; an answer engine can only report the silence.
• Whether a precedent fits a new deal is a judgment about the borrower, which the analyst signs.
• Each policy release changes the answers, so the bank's test set has to run again.

How credit teams answer policy and precedent questions today, and where the hours go

In shortThe analyst searches the policy PDF, asks a colleague, digs through shared drives for an old memo, and then waits for a senior. The senior answers from memory between memo reviews. In the estimate below the waiting is the largest cost, and it lands on the same people who review every deal.

The analyst, seniors and CCO behind a policy answer

One policy question, three people, then the memo moves

Most of one policy question's hours: waiting for a senior

Only 3.5 of those hours are the analyst's own work; the rest is elapsed time while the question sits with someone else. That matters because the question usually blocks the memo, and the memo blocks the committee date. Banks answer borrowers quickly by their own account: "Three in four banks approve their typical loan within ten business days", and nearly three in four have "at most three levels of approval".40 A day lost to one unanswered question is a tenth of that window.

Binder, training program or policy search tool

One data point on memo time circulates widely: a student project with a large Brazilian bank's credit division put a manual memo on a listed company at 2 to 4 hours.41 It is one bank and one kind of memo. We found no public figure for US commercial credit memos, or for the time an analyst spends on policy questions.

From "Can we go to 80%?" to a cited exception paragraph, and the synergies question after it

In shortPooja Ranganathan, a credit analyst seven months in the role, gets two questions on Monday, October 12, 2026: a relationship manager's 80% loan-to-value request for Kestrel Precision Machining, and her own Hollis & Crane acquisition going to committee on Thursday. Six agents answer each beside the deal: the rule in force, every past decision on one scale, the closest precedents with how they performed, and the test worked on the deal's numbers. Pooja writes; the senior officers approve the exceptions.

1

The rule in force, and how the bank has used it

Relationship manager Gwen Tolliver asks whether the bank can lend 80% on an owner-occupied plant. Pooja types the question beside the Kestrel deal. The answer opens with §6.4.2 of Credit Policy v14, quoted word for word, and the number that matters set large: 75%. Four facts sit under it, each opening at its passage: 75% to 80% is an exception, Renata Silva may approve it, coverage must be at least 1.25x, and nothing goes above the 85% supervisory limit.

On the right, 27 past decisions on owner-occupied industrial loans sit on one scale against the policy line, the exception ceiling and the federal limit, with Kestrel marked at 80%.

2

Three precedents, with how each loan has done since

The three closest past memos slide in as cards. Brightwater Machining got 80% by exception on 1.48x coverage and two guarantees, and has never been past due. Haldane Fabrication got 80% on 1.41x after the bank took a lien on $640,000 of equipment. Corwin Cabinetry asked for 80% at 1.18x and was held to 75% until the owners added $140,000 of equity.

Each card shows the reviewer's own words and a quarter-by-quarter strip of the loan's performance since. Renata Silva wrote on Brightwater: "Five points on owner-occupied is fine when the operating company carries the debt at 1.4x or better. Below that, no."

3

A $170,000 exception, drafted with six citations

A dark strip closes the answer with the test worked on Kestrel: the lower of price and appraisal is $3,400,000, the policy maximum $2,550,000, the request $2,720,000, the exception $170,000, five points, within §9.2.3. Kestrel's 1.45x sits between Brightwater's and Haldane's.

Pooja selects "Draft the exception paragraph". Every clause carries its citation, and she can edit any word before it goes into section 3 of the memo. One more click sends it to Renata Silva with the rule, the precedents and the numbers.

4

Her own deal: do the synergies count?

Pooja asks about Hollis & Crane. The rule in force is §4.3.4: synergies count only when documented by a quality-of-earnings report or signed contracts, and never above 10% of pro forma EBITDA. The chart plots the 12 closest of 41 acquisition loans to manufacturers by leverage with no synergies. Hollis & Crane sits at 3.43x, between the 3.25x limit and the 3.75x a senior credit officer may approve; capped, it is 3.12x.

Two lines summarize the practice: nine of 29 approvals were above the limit with no synergies, all as exceptions; three loans were downgraded within two years, two of them having credited synergies without a report.

5

The leverage paragraph, in the memo, before Walt reads it

On My deal, Pooja adds the cited leverage paragraph to section 6.2: 3.43x with no synergies, 3.12x with $423,000 credited, a policy exception for senior credit officer approval under §9.2, a step-down covenant to 3.00x by fiscal year-end 2027 as on Ashgrove Valve, and Bramwell named as the difference. Six citations go in with it.

Walt Ostrander, the senior credit officer, still decides the exception. Pooja did not have to wait for him to write it, and the four questions he usually asks are already answered in the memo.

The solution's design target is 14 seconds from a question to a cited answer, against a day or two waiting for a senior today. The screens show 12 and 14 seconds on sample data; measure it on your own policy and memos.

The full answer, the policy, every precedent, the deal, the gaps and the training log

In shortThe search screen is one click from everything behind it: the written answer with each agent's step, any source opened at the cited passage, the policy itself with what changed in v14, a ranked table of comparable memos, the deal's tests worked line by line, the questions the policy leaves open, the questions sent to seniors, the head of credit analysis's training log, and the settings that govern every answer.

Select any screen to open it larger. The bank, borrowers, people and figures are fictional sample data.

Every sentence checked against its cited passage, and exceptions left to the policy's named approvers

In shortEach agent has one job and limits it cannot cross. None approves an exception, assigns a risk grade or speaks for a reviewer. Every sentence of an answer is checked against the passage it cites before anyone reads it, and anything unsupported is removed. When nothing answers, no answer is written.

Question scoper, precedent finder, citation checker and three more

Analysts sign every paragraph; named approvers grant exceptions

Every question, answer and source is kept in the training log and audit trail. Questions the library cannot answer go to the senior named for that topic, with everything searched attached and a reminder after two business days.

The new analyst and the senior officers she asks

Searching the policy, guides, past memos and reviewers' comments, with access following the credit file

In shortIt reads what the bank already keeps, answers beside the deal the analyst is writing, and leaves behind cited paragraphs, likely reviewer questions with drafted answers, policy gaps for the policy owner and a log of who asked what. Access follows the credit file, and the bank's systems of record stay where they are.

Policy archive, origination system and rating history it reads

For each, check whether it has an API, a file feed or only screens, and who owns it. Sources are indexed nightly in the sample bank.

Design targets and the sample month, to be measured on your documents

Rule plus precedent, or no answer at all

What an unanswered policy question costs you in analyst hours, senior hours and committee dates

In shortThe numbers that matter most here are your own: how long analysts wait for answers, how many memos come back, how often exceptions go unlabeled. Two weeks of logging will find most of them. Two published figures frame them, what an analyst hour costs and how fast banks decide.

Logging two weeks of analyst policy questions

Unlabeled exceptions and precedents that went bad

Analyst and senior-officer hours a cited policy search gives back, with a calculator

In shortPut in your own numbers. The calculator starts from a published analyst wage and asks you for the rest: how many analysts, how many questions they ask, the hours each saves, how many questions reach a senior today and after, and what the tool and your testing cost a year. Every default marked illustrative is an assumption to replace.

Answers in seconds, seniors freed for unwritten questions

Questions a month, priced at analyst and senior rates

A team of nine analysts asking 60 questions a month asks 6,480 a year, above every line here. A team of three analysts asking 20 a month asks 720, and should cost the senior hours and the exceptions caught before it buys anything.

Before a policy search goes live: questions, policy versions, precedent files, exceptions and access, scored

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

A low score: settle one current policy version first

Holding a credit policy search tool or analyst knowledge assistant to account on versions, precedent and questions nothing answers

In shortAsk every vendor the same questions, us included. The answers show what the tool reads, how it handles versions and precedent, what it does when nothing answers, who sees what, and how you would test it on your own policy before your analysts rely on it.

We found no published independent test of any credit policy assistant, as of October 2026. A test set your seniors write, scored by them, is the fairest comparison, and the same test describes your current process.

Your senior credit officers write the test: a twelve-week policy search trial built from logged analyst questions

In shortLog the questions first, then let your seniors write the test, score a tool on it, run it beside your analysts, and decide with four numbers. None of these steps commits you to a vendor.

1. Log two weeks of questions, with minutes spent, who answered and how long the wait was. That is your baseline.
2. Settle the documents: the current policy and guides with their versions, a memo archive linked to outcomes, and written access rules.
3. Have seniors write 40 to 60 questions they already answer, including the hard ones: changed sections, exceptions that look like rules, silences. Write the answers first.
4. Run side by side for six weeks: analysts use the answers next to the policy and their senior, and note every difference.
5. Decide with four numbers: answers correct on the test set, share answered without a senior, time to an answer, and memos returned.

Correct answers on the seniors' test set

Loan policy exceptions, leveraged lending, synergies and AI answers: what analysts search for

In shortThe questions credit teams and buyers search for most, each answered on its own.