Fair lending exception monitoring, explained.

A rate sheet sets the price and a person changes it: why fair lending reviews start with exceptions

The life of one exception, in five steps

1. The price is setThe rate sheet prices the loan on score, LTV, product and lock date. Nobody has discretion yet.
2. Someone moves itA concession, a waived fee, a lender credit or an override, granted by an officer or approved by a manager.
3. The reason is written and checkedThe officer records why, with the evidence the policy asks for. This is the moment most reviews miss.
4. The exception is comparedAgainst loans like it, by branch, officer and product, and by borrower group in compliance's hands only.
5. Gaps are explained or fixedRemediation for every borrower affected, limits on discretion, a policy change, a report to the board.

Steps 3 and 4 are the subject of this guide.

Four surprises about pricing exceptions and the law

Four terms behind every exception review

Fourteen terms are defined in the one-page summary.

Rate concessions, fee waivers, lender credits and overrides: the four kinds of loan exception

In shortAn exception is the part of a loan's price or approval that a person chose. On the pricing side that means concessions, waived or reduced fees and lender credits with no matching rate increase; on the credit side, exceptions to underwriting criteria and credit score overrides. Examiners treat broad discretion without "clear and objective criteria" as a risk factor in itself.

The clearest definition in a public document comes from the Justice Department's 2022 consent order with a Memphis bank, Evolve Bank & Trust. It defines concessions as "offering a combination of rate and points that yields a below par price", waiving or reducing customary lender charges (origination, underwriting, discount, processing, document and administrative fees), and "offering lender credits without a corresponding rate increase".7 Each one lowers the borrower's price below what the sheet would have charged.

Overrides sit on the credit side. A "high-side" override declines a file the score would approve; a "low-side" override approves one it would decline. Underwriting exceptions approve a debt-to-income ratio, a reserve level or a loan-to-value ratio outside policy, usually on compensating factors such as savings or residual income.

The interagency examination procedures list the warning signs examiners look for. For pricing: "Presence of broad discretion in loan pricing", which "may be present even when institutions provide rate sheets and fees schedules, if loan officers or brokers are permitted to deviate from those rates and fees without clear and objective criteria" (indicator P2).3

For underwriting: "Lack of clear guidance on making exceptions to underwriting criteria, including credit scoring overrides" (U5), "Lack of clear loan file documentation regarding reasons for any exceptions" (U6) and "Relatively high percentages of either exceptions to underwriting criteria or overrides of credit score cutoffs" (U7).3

Who usually grants what

No federal rule sets authority tiers. A typical retail lender lets a loan officer discount a few basis points alone, sends larger concessions to a pricing desk or secondary marketing manager, and sends underwriting exceptions to an underwriting manager. Regulation Z adds one hard edge: a creditor "may change credit terms or pricing to match a competitor", but the loan originator's compensation on that loan "may not be changed for those reasons".8 The lender pays for a concession out of its own margin.

ECOA, the Fair Housing Act and Regulation B after the 2025–2026 disparate impact changes

In shortThe statutes have not changed. What changed is how the federal government reads them: an April 2025 executive order, OCC and FDIC exam changes in mid-2025, and a CFPB rule effective July 21, 2026, now challenged in court, that says ECOA does not support disparate-impact claims. Disparate treatment, including comparative evidence from like files, is untouched. Disparate impact still lives under the Fair Housing Act in the courts, and in New Jersey and, from 2027, Illinois.

What the 2026 Regulation B rule changed, and what it kept

The CFPB's final rule, published April 22, 2026 and effective July 21, 2026, deletes the language saying the "effects test" may apply under ECOA and adds that the Act does not recognize it.4 Its new commentary keeps a line that matters for monitoring: the Act does not reach facially neutral practices "except to the extent that facially neutral criteria function as proxies for protected characteristics designed or applied with the intention of advantaging or disadvantaging individuals based on protected characteristics".4

Commenters asked the Bureau to confirm that lenders may keep using proxy analysis to check their own compliance. It declined to add guidance, calling the question "too nuanced and detailed to be appropriate for inclusion in a regulation", and said courts are best positioned to decide whether evidence makes a prima facie case.4 Nothing in the rule restricts a lender's own testing.

The same rule also narrowed what counts as discouraging a prospective applicant and added conditions for special purpose credit programs, and it struck the words "including a prospective applicant" from the self-test privilege (section 10).4 National Fair Housing Alliance v. CFPB challenges the rule; no preliminary injunction was sought, so the rule is in effect, and the cross-motions for summary judgment will not be fully briefed until January 12, 2027.15

Fair lending exams, as of October 2026

Executive Order 14281 directed agencies that enforce ECOA and the Fair Housing Act to "evaluate all pending proceedings that rely on theories of disparate-impact liability" within 45 days, and all agencies to review existing consent judgments within 90 days.9 The OCC told examiners on July 14, 2025 to stop examining for disparate impact while keeping fair lending risk assessments, HMDA analysis for disparate treatment and risk-based exams.10

The FDIC followed on August 29, 2025: it evaluates discrimination "only through evidence of disparate treatment".11 Its manual now names two routes, "Overt evidence of disparate treatment" and "Comparative evidence of disparate treatment".3

Comparative evidence is exactly what an exception review produces. The 2023 CFPB findings on pricing exceptions were findings of discrimination "in the incidence of granting pricing exceptions" between similarly situated borrowers,5 and they would read the same under the 2026 rule.

Disparate treatment, disparate impact and redlining: which theory reaches a pricing exception now

In shortDisparate treatment means a borrower was treated differently on a prohibited basis; it can be shown by overt statements or by comparing like files. Disparate impact means a neutral rule fell harder on a group without enough justification. A gap in discretionary exceptions between like borrowers is comparative evidence of treatment. Redlining is treatment too. Impact claims survive with private plaintiffs under the Fair Housing Act and with some states.

The procedures describe the comparative route in one line: disparate treatment may be shown by "differences in treatment that are not fully explained by legitimate nondiscriminatory factors".3 They add that it is most likely in the middle of the credit range, where "there is more room and need for lender discretion" and outcomes "may depend on the level of assistance the lender provides".3 That is the territory of competitor matches and compensating factors.

Redlining, giving "unequal access to credit, or unequal terms of credit" because of the make-up of an area, is described in the same procedures as "a form of illegal disparate treatment", so it also survives the 2025–2026 changes.3

The Supreme Court held in 2015 that disparate-impact claims are cognizable under the Fair Housing Act,20 and HUD's proposal to remove its own rule was still open in October 2026.13

New Jersey's rules put the burden on the covered entity in housing finance to show a practice is necessary and that no less discriminatory alternative exists.12 Illinois's law, effective January 1, 2027, puts the same burden on "financial institutions" for "criteria or methods" with discriminatory effects; the act does not define the term, so its reach to nonbank lenders is unsettled.14 A lender in those states still needs impact analysis; everywhere, it needs the treatment comparison.

The Massachusetts attorney general settled with a student lender for $2.5 million in July 2025, alleging among other things an underwriting variable with disparate outcomes under state law.21 Much of the recent enforcement comes from states. The law firm Goodwin counted 12 publicly announced enforcement actions about mortgage origination and servicing in 2025, on every topic with fair lending among them, and states brought nine of them.22

The CFPB's own report on 2025 says it has focused its fair lending enforcement "on matters with direct evidence of intentional racial discrimination and identified victims", and that in 2025 it, the FDIC and the Federal Reserve referred no ECOA matters to the Justice Department.23 The statutes' limitation periods are long: five years for an ECOA suit, two years after the last occurrence for a private Fair Housing Act suit.24,25

How a fair lending exam turns a pricing gap into file-by-file questions

In shortExaminers scope the lender's discretion, pick a prohibited-basis group and a control group, compare terms where discretion lives, and ask the lender to explain each difference. Statistics are used where volume is large; file comparison everywhere. Small groups are not ignored: a pattern among four borrowers can be enough to look.

The procedures tell examiners to review "all loan terms and conditions" with special attention to those "left, in whole or in part, to the discretion of loan officers or underwriters", to "obtain explanations from the appropriate loan officer or other employee for any differences that exist and reanalyze the sample".3 Where volume is large, agency policy may "call for the use of regression analysis or other statistical methods".3

Small numbers do not end the inquiry. The procedures give the example that there is "strong reason to examine a pattern in which almost all of 19 male borrowers received low rates but almost all of four female borrowers received high rates", even below the minimum sample sizes.3 A branch or one officer is often where a pattern starts.

Where a lender has run self-tests or self-evaluations, examiners may use them to streamline scoping. The procedures warn that "voluntarily sharing the results of self-tests will result in a loss of confidential status".3

Measuring a concession from the rate sheet, matching peers and sizing a gap in dollars

In shortMeasure each concession from the price the rate sheet gave on the lock day. Match it to loans alike on product, score, LTV, market and lock window, chosen without looking at the borrower's group. Express the difference in basis points, then in dollars. Test whether a group's gap is larger than chance would explain, knowing that small groups give weak tests. A sample of forty files holds three loans of a cluster of eleven only about 4 times in 100.

Step 1: the concession, from the sheet

The final rate alone hides the exception. A borrower who bought the rate down with points and one who received a concession can end with the same rate. The unit that isolates the choice is the concession: the sheet's price on the lock day minus the price charged, in basis points of the loan amount. The Evolve complaint used it: "A basis point is one hundredth of one percent of the total principal amount of the loan."26

Step 2: true peers

Peers must be alike on what legitimately sets price, and the borrower's group must play no part in picking them. A common definition: same product, credit score within 20 points, LTV within 5 points, the same market, a lock within 30 days on the same sheet. Each band is a policy choice and belongs in writing before the review runs; the Evolve order required analysis "methods, thresholds and controls approved in advance" by the Justice Department.7

Step 3: the gap in dollars, using the example

The fictional example. The 11 borrowers' credits, each rounded to the dollar, add to $13,060.

Step 4: is it chance?

A significance test asks how likely a gap this size would be if group made no difference, after controls for score, LTV, amount and product. Two cautions. Groups of eight to fifteen loans give tests with little power, so one odd loan can swing the result either way. And a test run on estimated groups inherits the estimate's error (section 07). A significant gap is a reason to read the files; the files decide whether there is an explanation.

Why a forty-file sample misses a small cluster

A reviewer needs more than one loan from a cluster to see a pattern. Reading every exception against its peers removes the sampling problem for this part of the review, and leaves the reviewer's time for the files that matter.

HMDA-reported race and BISG estimates: how fair lending reviews know borrower groups, and where estimates fail

In shortFor home purchase and refinance loans, lenders ask borrowers for ethnicity, race and sex, and borrowers may decline. About one originated mortgage in six in 2025 had no reported race. Reviewers fill gaps with a statistical estimate, usually Bayesian Improved Surname Geocoding (BISG), which combines surname and the make-up of the neighborhood where the borrower lives into a probability. It ranks well for some groups and poorly for others, and threshold choices can bias results. Estimates belong to compliance alone and must never reach pricing.

Regulation B requires a lender taking an application to buy or refinance a principal residence to ask for ethnicity, race, sex, marital status and age; the applicant is "asked but not required" to answer. When the applicant declines, the lender notes, "to the extent possible", ethnicity, race and sex "on the basis of visual observation or surname".16 Those answers become HMDA data. Outside mortgages, lenders generally may not ask, so reviews rely on estimates.

How BISG works, and how well

The CFPB's published method "combines geography- and surname-based information into a single proxy probability for race and ethnicity".28 It starts from the surname's distribution in Census Bureau data and updates it with the population make-up of the borrower's block group, a small Census Bureau area, falling back to the census tract or ZIP code when the address cannot be placed more precisely. The result is a set of probabilities for each borrower; an illustrative one might read 82% Black and 15% white. For sex, the CFPB uses a separate proxy from first names, based on Social Security Administration data.28

Tested on mortgage applicants whose race was reported, the method's ability to rank people varied by group. The area under the curve (AUC) measures how often a member of a group gets a higher probability than a non-member: 1.0 is perfect, 0.5 is a coin toss.

Counting is a second test. On the same pool, summed BISG probabilities estimated 11,516 Hispanic applicants where 11,073 had reported, about 4% over; an 80% threshold rule that assigns everyone above it to the group counted 8,314, 25% under. The CFPB warns that thresholds "can artificially bias, usually downward" group counts and "potentially attenuate estimates of differences in outcomes between groups".28 Researchers at Cornell found that disparity estimates could swing "wildly" with the threshold chosen.29

Rules for using estimates

The third rule is the one the 2026 commentary makes sharper: a proxy used "with the intention of advantaging or disadvantaging" people is disparate treatment.4 The estimate exists to measure outcomes after the fact; it has no place in any decision about a borrower.

Competitor-match findings, coaching and the pricing discretion cases from Countrywide to Evolve

In shortCFPB examiners found lenders discriminating in who received competitor-match exceptions in 2021 and again in 2023, and in 2023 pricing exceptions were among its two most frequent fair lending issues. The causes were unwritten policies, no evidence of the competing offer, discounts below a threshold nobody tracked, coaching some customers and not others, and monitoring reports nobody acted on. The larger court cases all turned on unguided discretion over part of the price.

The CFPB's pricing exception findings

In fall 2021, examiners reported that mortgage lenders "violated ECOA and Regulation B by discriminating against African American and female borrowers in the granting of pricing exceptions based upon competitive offers". Lenders relied on "a verbal policy that a consumer must initiate or request a competitor price match exception", granted matches to non-Hispanic white and male borrowers "with no evidence of customer initiation", and "failed to retain documentation". Their own fair lending monitoring had raised the concern, and nothing changed.30

In summer 2023 the finding returned "across a range of ECOA-protected characteristics, including race, national origin, sex, or age". Some policies let officers request an exception with no documentation; managers approved with "no guidelines for the bases for approval or denial"; some policies had "no documentation requirements for pricing exceptions below a certain threshold". Lenders did not monitor whether "the loan officer was not coaching certain consumers and not others regarding the competitive match process", and some "failed to take corrective action based on their statistical observations of disparities".5

The CFPB's report for 2023 said "two of the most frequently identified fair lending issues in supervisory communications related to the granting of pricing exceptions and HMDA violations". It referred 18 matters to the Justice Department that year, including "discrimination in pricing exceptions on the basis of race, national origin, sex, and age".31

The cases about pricing discretion

Countrywide (2011). The Justice Department and the Illinois attorney general settled for $335 million over company policies that gave employees and brokers "almost unlimited discretion in the selection and pricing of loans".32 Ally (2013). $80 million in damages and $18 million in penalties; the agencies found "more than 235,000 minority borrowers paid higher interest rates" through dealer markups on auto loans.33 Honda (2015). $24 million in restitution and a pricing system changed "to substantially reduce dealer discretion".34

Evolve Bank & Trust (2022). The complaint alleged that until at least July 2018 the bank "did not require its loan officers to have a reason or to provide any documentation or justification" for concessions. On about 21,670 loans from 2014 to 2019 it alleged discretionary pricing 13 to 19 basis points higher for Black borrowers than for white borrowers, 14 to 21 higher for female than male borrowers, and 18 to 25 higher for Hispanic borrowers through June 2018.26 The case was pleaded as disparate impact.

The consent order set a $1.3 million fund and a $50,000 penalty, required objective pricing criteria, at least annual monitoring of "concession incidences and value of concessions in basis points", and quarterly board updates.7 It was terminated early, on May 29, 2025, after the Justice Department told the court the bank had compensated the borrowers.35

Per borrower, the harm is small enough that nobody notices it on a closing statement. Across thousands of loans and several years it becomes a settlement.

Mortgage pricing gaps by race in HMDA data, and how much of the raw gap credit explains

In shortAbout 4,768 lenders filed HMDA data for 2025, on 6.8 million originated loans. Raw denial shares differ widely by race. On price, the best studies find gaps of a few basis points of rate after credit controls, worth about $1,600 to $1,700 in present value on a 30-year, $200,000 conventional purchase loan, and larger where people set part of the price. Nobody publishes how many loans carry exceptions or what the reviews cost; measure your own.

Raw gaps start an inquiry; they cannot end one. In the FDIC study, controls explained more than half the Black-white rate gap and uncovered a larger Hispanic-white one.37 The NBER study adds a pointer to where gaps come from: "FinTech algorithms also discriminate, but 40% less than face-to-face lenders".38 In that study, gaps were larger where a person set part of the price.

The public file cannot settle any of it. Under the CFPB's 2019 disclosure policy it excludes "the credit score or scores relied on in making the credit decision", the loan officer's NMLS identifier and the action dates.17 An outside analyst sees a lender's price gaps but not the factor that most explains them. Examiners hold the full register; the lender holds the rate sheets and the reasons.

Paying borrowers back, the self-test privilege and how long a priced loan stays exposed

In shortWhen a review shows a violation is more likely than not, Regulation B's self-test privilege requires corrective action, including relief for each applicant affected. The privilege covers a voluntary self-test's results, never the loan files or data derived from them. A loan priced today can be sued on for five years under ECOA, while Regulation B only requires the file for 25 months.

Regulation B privileges "the report or results of a self-test that a creditor voluntarily conducts", but "only if the creditor has taken or is taking appropriate corrective action". Action is required when the test "shows that it is more likely than not that a violation occurred", it may include "both prospective and remedial relief", and "taking corrective action is not an admission that a violation occurred".6 The privilege does not cover "loan and application files or other business records related to credit transactions, and information derived from such files and records, even if the information has been aggregated, summarized, or reorganized".6

HUD's Fair Housing Act rule has the same carve-out.39

The 2026 Regulation B rule struck the words "including a prospective applicant who alleges a violation of §1002.4(b)" from the privilege's protection against use in court. The CFPB says no substantive change was intended; state attorneys general objected that the results could now be used against a lender by prospective applicants.4,6

Two practical consequences follow. An exception review built from loan files should be written as if an examiner will read it, stating what the evidence shows and no more. And the privilege is lost if the lender discloses the results "as a defense to charges" of a violation.6 How a program is scoped and recorded is a question for counsel.

Three kinds of corrective action

The gap between five years and 25 months is the practical argument for reviewing exceptions while the files, and the officers' memories, still exist.24,40

SR 26-2 and an automated fair lending review: why the race estimate and the significance test count as models

In shortThe April 2026 model risk guidance covers statistical methods that produce estimates, so the significance test and the race estimate are models; peer matching on fixed bands and the dollar arithmetic are not. AI that writes text or acts on its own is outside its scope, which leaves the lender's own governance to set controls for software that reads reasons or drafts reports. Regulators use AI for the same job, finding outliers, and keep decisions with people.

The April 17, 2026 guidance (SR 26-2, OCC Bulletin 2026-13, FDIC FIL-15-2026) defines a model as "a complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates", excludes "simple arithmetic" and "deterministic rule-based processes", and says generative AI models, and AI that acts on its own, "are not within the scope of this guidance".41

The guidance says it is most relevant to banking organizations with over $30 billion in total assets; smaller banks govern their models under practices suited to their size.41

The GAO found that financial regulators "use AI to identify risks, support research, and detect potential legal violations, reporting errors, or outliers", and that its outputs "inform staff decisions but are not used as sole decision-making sources". It also warned that AI "could also lead to lending bias".42 A monitoring tool should be held to the same line: it may find, measure and draft; people judge what a gap means and what the lender owes.

Vendors do not carry the duty. The privilege, the remediation and the explanation to an examiner all sit with the lender.6,3

The quarterly spreadsheet, the file sample and the reason chase: pricing exception reviews as lenders commonly run them

In shortA fair lending officer or analyst pulls the exception report after quarter end, joins it to loan data, chases officers for missing reasons, reads a sample of matched files, runs tests where volume allows, and reports to the board. Most of the time goes to chasing reasons and building the sample. The work is done in house, by outside counsel or consultants, or with analytics software.

The five people a pricing exception review depends on

From the exception export to the board report, week by week

How a manual quarter's 22 review days split

In house, outside counsel or analytics software

Whichever route, the lender owns the methods, the thresholds and the conclusions.

The Linden Avenue cluster: 11 small concessions, a 38 bps gap and a $13,060 remediation signed twice

In shortOvernight after the quarter, six agents read every exception's reason against its evidence, match each loan to its peers and test branches, officers and products for gaps. On Monday morning Dana Whitlock, the fair lending officer, lassos the cluster at Branch 14, opens the loans and the closest matched pair, and signs a three-part remediation that Harold Bisset, the chief compliance officer, co-signs. Nothing is paid or changed until both sign.

1

Every rate concession of the quarter on one chart

Each dot is one concession. Across: what similar loans received, as the median of the loan's peers. Up: what this loan received. On the diagonal, a borrower got what peers got; the shaded band is the bank's 15 basis point tolerance; below it, the borrower got less. Colors are borrower groups, reported on the HMDA form or estimated, visible only to compliance. A filled dot has a supported reason, a pale one a reason without evidence, a ring no reason at all.

The run log on the right records the night: 549 exceptions collected at 2:01 am, 549 reasons read by 2:04 (446 supported, 42 written but not supported, 61 missing), every exception matched to its peers by 2:06, three clusters flagged by 2:10.

2

She lassos the cluster below the band

Dana picks the Branch 14 cluster and the lasso closes around 11 dots far below the band. The panel reads the selection in one sentence: concessions 38 basis points smaller for one group, 9 of 11 without a documented reason. Black borrowers (estimated) received 12 basis points on average; similar loans received 50. Eight of the 11 were granted by one loan officer, Kyle Fennimore.

3

The loans, the reasons and the test behind the flag

The investigation lists the 11 loans with score, LTV, the concession each got, its peers' median, the gap and the reason as written: "No reason entered" on nine, "Manager approved" without support on one, and one relocation discount with the employer letter on file. Beside them, the 14 comparison loans at the same branch averaged 50 basis points; ten were coded "competitor match", and only four have the competing offer on file. The panel replays how the gap was tested: 38 basis points, p = 0.002, holding after score, LTV, amount and product.

4

The closest match, factor by factor

Dana opens the first loan. 26-14-0871, A. Whitaker: 30-year fixed, $328,000, score 742, LTV 80%, DTI 34%, locked August 14, 10 basis points, no reason. Its closest match, 26-14-0833, G. Holm: $315,000, score 738, LTV 80%, DTI 36%, locked August 11, 50 basis points, "Competitor match — borrower has offer", with no competing offer found. Same branch, same officer, each measured from its lock day's rate sheet. The stronger file got 40 basis points less: $1,312 on $328,000. The borrower group was not used to choose the match.

5

Three actions, two signatures

The draft remediation has three parts. Credit the 11 borrowers 38 basis points of their loan amount by principal reduction, with a letter: $13,060 in all (38 bps × $3,436,400 is $13,058; the 11 credits, each rounded to the dollar, add to $13,060). For 90 days, any concession by Kyle Fennimore above 10 basis points needs Marcus Ostberg, the pricing desk manager, before lock, reviewed again on January 11, 2027. Policy 4.2 is amended so no competitor-match exception is granted without the competing Loan Estimate, bank-wide from November 1, 2026.

Dana reads the per-loan credits and signs; the dialog states exactly what she is approving. Harold Bisset co-signs, and both signatures land on the audit trail beside each step of the night's run. The draft credits all 11, including the one relocation discount whose employer letter was on file; each action can be switched off before signing, and whether that loan belongs in the remedy is her call.

The solution's design targets: a documented reason within 3 business days of each exception, and every exception compared with matched loans, with no sampling. They are targets, to be measured on your own exceptions.

Beyond Branch 14: comparison loans, officer gaps, the live reason feed, the fee waiver cluster, the borrower letter and the board report

In shortThe cluster is the story. These screens show the rest: the comparison loans with their evidence, the officer view by borrower group, the feed that asks for reasons the day an exception is granted, the fee waiver cluster still in review, the signing dialog, the borrower's letter, the quarterly report and the settings.

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

What the night's reading found

Exception collector to report writer: what each agent may do, and the two compliance signatures before any borrower credit

In shortEach agent has one job and limits it does not cross. None changes a price or a loan, writes a reason for an officer, calls a gap discrimination, contacts a borrower or moves money. Borrower group estimates stay with the compliance role and are never used to choose peers or to price a loan.

The three points where a person decides

The three clusters that night

The data an exception review draws on, from lock-day rate sheets to HMDA records, and the evidence trail it produces

In shortIt reads what your systems already hold: closed loans and their exceptions, the rate sheets of each lock day, the documents behind each reason, your pricing and override policy, and HMDA data. It leaves a reason on every exception, a matched comparison for each, findings with their evidence, signed remediation and a quarterly report.

Connects to your origination system, pricing engine, documents and HMDA register

The bill for an unexplained pricing gap grows each quarter: remediation, penalties, exam time and five years of exposure

In shortThe cost of a gap grows with every month it runs: each new loan adds to the remediation, and the exposure lasts five years under ECOA. Published cases give the scale. The number that matters most is your own: the gap, the loans it touches and how long it has run before anyone looked.

Count your own missing reasons, unrecorded discounts and review hours

What a slow exception review costs beyond dollars

Estimating the return: shorter quarterly reviews and smaller remedies when a pricing gap is caught one quarter in

In shortPut in your own numbers. The calculator asks how many exceptions you grant, what a review day costs, how long the review takes now and after, how many exceptions sit in an unexplained gap, and how long such a gap runs before you find it. Every default is the fictional example's or an assumption marked illustrative.

Six places full exception monitoring pays

Estimate the return on your own exception volume

Exception policy, lock-day prices, reason timing, methods and governance: scoring a fair lending program's readiness

In shortPick the answer closest to your program 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.

What your readiness score means for an exception program

Fair lending monitoring software: fourteen vendor questions on peer selection, reason reading, race estimates and audit evidence

In shortAsk every vendor the same questions, us included. The answers show whether the tool measures from the rate sheet, reads the reason and its evidence, keeps race estimates walled off, separates finding from deciding, and leaves a record an examiner can follow.

Proving a monitor on your own files: replay a reviewed quarter against sealed test gaps, then a live quarter beside your team

In shortSet your methods first, then run the monitor over a quarter your team has already reviewed, with gaps planted by someone outside the pilot. If it reproduces what you found, clears what you cleared and finds the plants at the size planted, run it beside your process for one live quarter and decide with five numbers.

1. Write the methods: peer bands, tolerance, minimum group, significance and borrower-group method, approved before anyone sees output.
2. Pick a reviewed quarter and gather locks with sheet prices, exception reasons, evidence documents and HMDA data; records are kept 25 months, so last year is complete.
3. Plant gaps: one above the band in a large group, one just below it, one large gap in a group under the minimum, one spread across an officer. Seal the list.
4. Run and compare with your own findings and the sealed plants; agree with counsel how a real finding will be handled.
5. Run one live quarter in parallel, with reasons asked the day exceptions are granted, then decide.

The five numbers that decide the pilot

What compliance teams ask about pricing exceptions, disparate impact in 2026, BISG accuracy and the self-test privilege

In shortThe questions compliance teams search for most, each answered on its own.