Introduction — Why 'what‑if' answers matter now
When a lender denies a loan or offers less favorable terms, you normally get an adverse‑action notice listing one or more reasons. Increasingly, the underwriting behind those decisions uses machine learning or other complex algorithms — which can make the reasons sound vague or generic. A counterfactual explanation (a concise what‑if statement) tells you the smallest, concrete change that would flip the decision in your favor: for example, “If your reported credit utilization were below 30% or your on‑time payments increased by three in the past 12 months, you would qualify.” This form of explanation is designed to be actionable for consumers and avoids exposing proprietary model internals.
The Consumer Financial Protection Bureau (CFPB) has warned lenders that they must still provide accurate, specific reasons for adverse actions even when AI or complex models are used — meaning adversarially vague checklist reasons may not suffice. That regulatory context makes counterfactual, factor‑level explanations especially relevant for applicants who want to know whether and how they can realistically change an outcome.
What exactly is a counterfactual explanation?
Counterfactual explanations answer a simple consumer question: "What minimal change to my information would have produced a favorable decision?" Instead of exposing an algorithm’s internal weights or proprietary features, counterfactuals report a realistic, minimal change in observable inputs (e.g., debt level, utilization, income documentation) that would alter the model’s output. The approach was formalized in the literature as a privacy‑friendly way to give meaningful recourse to affected individuals.
- Example (simple): “If your revolving balances were reduced by $1,200, your application would have met this lender’s minimum underwriting threshold.”
- Example (multi‑factor): “If your bank‑verified monthly income were $500 higher and one recent late payment on Account X were removed, you would qualify for the advertised rate.”
Recent work on counterfactuals for tabular credit‑style data emphasizes creating feasible, realistic changes (called feasible recourse) versus impossible or illegal suggestions (for example, telling someone to change their age or fabricate income). Practical implementations use optimization and domain constraints to keep recommendations actionable.
What lenders should — and often must — provide
Under U.S. fair‑lending law (ECOA and Regulation B) and the CFPB’s guidance to creditors, adverse‑action explanations must be specific and accurately reflect the principal reason(s) for the decision; reliance on generic checkbox language is insufficient when it doesn’t describe the actual algorithmic factors considered. In practical terms, regulators expect disclosures that let applicants know what data or behaviors drove the outcome and give them a usable path to fix problems.
What to expect and ask for from a lender:
- Factor‑level description: Which data categories mattered (for example, revolving balances, recent delinquencies, verified income). This goes beyond a one‑word checkbox like “credit history.”
- Concrete counterfactual(s): Clear “what‑if” statements showing minimal, realistic changes that would likely change the outcome (e.g., reduce credit card balances by $X, provide a paystub showing $Y additional monthly income).
- Data provenance note: Whether the decision relied in whole or in part on third‑party data (bank feeds, alternative data, bureau records) and which sources so you can verify or dispute them.
- Limits and trade secrets: Lenders may refuse to disclose proprietary model details or trade secrets, but the CFPB makes clear they still must provide accurate reasons tied to the individual consumer’s data.
Sample phrasing you can request: “Please provide a written, itemized ‘what‑if’ explanation stating the smallest change(s) to my verifiable inputs that would have produced a different underwriting decision, and identify any bureau or third‑party source used.” (Keep a copy for disputes.)
How to use a counterfactual explanation to actually improve your odds
1) Read the recommendation as an operational checklist. If the counterfactual says “reduce revolving balances by $1,200,” that is an explicit target you can try to hit before reapplying — or document (if balances were reported in error). If it suggests ‘verify income’ you can gather paystubs or a letter from your employer. The CFPB explicitly notes that consumers should receive reasons that let them identify incorrect data and take corrective steps.
2) Verify the underlying data sources. If the explanation references bureau data or a bank feed, pull your credit reports, bank statements, or alternative‑data records to confirm accuracy. If you find errors, follow the lender’s dispute process and, when applicable, dispute with the credit bureau and the data provider. Document all communications, dates, and evidence.
3) Ask about timing and thresholds. Counterfactuals are model‑ and lender‑specific. Ask whether the lender’s thresholds are hard cutoffs or probabilistic ranges (e.g., is $1,200 an approximate target or a precise cutoff?) and whether partial improvements materially increase approval probability. That helps prioritize what to change first.
4) Use feasible recourse principles. Not every suggested change is realistic (e.g., you can’t lawfully change age or employment history overnight). Favor counterfactuals that propose feasible, documentable changes — paying down balances, correcting a misreported delinquency, or providing verified income documentation. Recent technical work stresses that recourse must be feasible and lawful to be useful.
5) Plan reapplication strategically. If the lender’s counterfactual gives a clear path (e.g., reduce balances by $1,200 and re‑verify income), set a target timeline, gather proof, and ask whether the lender will accept supplementary documents for an expedited manual review. If the lender won’t provide recourse, consider applying elsewhere or using starter products (secured cards, credit‑builder loans) to change the underlying profile first.
Practical tips, sample scripts and a short checklist
Quick consumer checklist:
- Request a written counterfactual explanation when you get an adverse action (keep dates and the name of the representative).
- Collect the specific documents the explanation implies (paystubs, bank statements, settlement letters, account statements showing corrected balances).
- Dispute any incorrect data with the lender and the bureaus promptly; use the counterfactual wording to show how an error affected the decision.
- Track time: many bureau disputes and lender responses have statutory or contractual timelines; follow up if deadlines pass.
Sample script to request a counterfactual explanation (phone or secure message):
"I recently received an adverse‑action notice for Application ID [#]. Under CFPB guidance and Reg B, I respectfully request a written counterfactual explanation that identifies the specific, minimal changes to my verifiable inputs (for example, a target balance reduction, verified income amount, or removal of a reported delinquency) that would have made my application qualify for the offered product. Please identify any third‑party data sources used so I can verify their accuracy."
Keep this note short, factual, and attach any supporting documents that already address the counterfactual points (for instance, a current bank statement showing you’ve paid down the stated balance).
Remember: lenders can lawfully refuse to disclose trade secrets, but current CFPB guidance requires that reasons still accurately reflect the factors relied on for the individual decision. If you believe a lender’s notice is non‑specific or misleading, you can escalate: file a complaint with the lender, copy your complaint to the CFPB, and preserve all correspondence.
Limitations and realistic expectations
Counterfactuals are powerful but not magical. They are typically model‑dependent and may indicate what would likely change the model’s output, not guarantee approval. Lenders differ in how they define thresholds; a counterfactual from Lender A won’t necessarily transfer to Lender B. And in some cases, post‑hoc counterfactuals generated by approximation techniques can be inaccurate if the model is not amenable to faithful approximation — which is why the CFPB urges lenders to ensure explanations are accurate and supported by validation.
If you follow the counterfactual advice and still get the same result, collect evidence of the steps you took and ask the lender for a manual review or a more detailed explanation of the persistent constraints. If necessary, consider alternative financing paths that explicitly help build the recommended items (secured cards, credit‑builder loans, rent reporting services).
Bottom line: A clear, feasible counterfactual explanation turns an opaque denial into an operational plan: it tells you what to verify, what to fix, and whether it’s worth reapplying — and regulators now expect lenders to provide explanations that do exactly that.