What would have changed the outcome?
Surfaces when clients need to tell affected users not just why a decision was made, but what a realistic path to a different outcome looks like: common in financial services, benefits, and hiring contexts.
Attribution tells users what mattered in a decision. Counterfactual explanation tells users what could have changed the outcome. These are related but different, and both are needed for a decision to feel genuinely explainable.
A counterfactual says: "If your income had been €500 higher per month, this application would have been approved." Or: "If the employment gap had been under 6 months, the risk score would have fallen below the referral threshold." It answers the question that users naturally ask after receiving an adverse outcome: what would have been different if...?
This is not the same as Actionable Recourse, though the two are closely related. Counterfactuals are explanations: they describe the decision boundary. Recourse is practical guidance: it tells users what they can actually do. A counterfactual might describe a change that is technically possible but practically out of reach for this user. Recourse filters for what is realistically controllable.
Used together, counterfactual explanation provides the understanding and recourse provides the path forward.
Show users the minimum or most relevant changes to their situation that would have produced a different outcome: making the decision boundary visible without requiring users to understand the model.
A plain-language sentence describing the smallest change that would have altered the outcome: "If your debt ratio had been below 35%, this application would have been approved." Simple, direct, and actionable.
A side-by-side card showing the current profile vs. the counterfactual profile: with the changed factors highlighted. Users can immediately see what's different.
A simple visual showing where the current score sits relative to a decision boundary, and how far it is from a different outcome. A number line, progress bar, or gauge can make the proximity to the threshold visible without revealing the exact threshold number.
Some decisions require changes to multiple factors simultaneously. A counterfactual explanation that shows "changing factor A and factor B together would change the outcome" is more honest than implying a single-factor change is sufficient.
A specific version of the counterfactual for borderline cases: "This application came close: 3 of 5 key criteria were met. The two factors that were not met are: [X] and [Y]." Particularly useful for motivating reapplication.
A visual distinction within the counterfactual between factors the user can actually change and factors that are fixed (age, historical records). This is the bridge between counterfactual and recourse.
The mathematically minimal change is not always the most useful counterfactual. A change that is theoretically minimal but practically impossible is not helpful. Prioritize counterfactuals that are actionable.
"Had your income-to-debt ratio been 0.43 instead of 0.51" is less useful than "if you were paying €200 less per month in existing debt."
If changing one factor alone would not have changed the outcome: if two or three factors would need to change together, say so. A counterfactual that implies a single change would suffice when it wouldn't is misleading.
A counterfactual describes the model's decision boundary at the time of the decision. It is not a promise that making that change will produce a different result in a future application (model versions change, other factors shift). Use appropriate qualifying language.
The natural user response to a counterfactual is "so what can I do?" Don't let counterfactuals be a dead end: connect directly to Actionable Recourse.
Transparency vs. gaming
Counterfactuals reveal the decision boundary. In contexts where users could manipulate their inputs to cross a threshold without genuine improvement (e.g. staging financial documents), this is a real risk. Consider showing direction without precise thresholds, or scoping counterfactuals to user-facing recourse channels rather than raw system access.
Helpfulness vs. false promise
A counterfactual implies: "if you change this, the outcome will be different." But models are not deterministic across applications, and policies change. Use qualified language and a clear disclaimer about what the counterfactual does and doesn't guarantee.
Simplicity vs. completeness
The simplest counterfactual is a single-factor change. But many real decisions have complex multi-factor boundaries. Showing only the simplest counterfactual when multiple changes are required is misleading. Acknowledge complexity.
Attribution and counterfactual explanation are complementary. Attribution answers "what mattered." Counterfactuals answer "what would have changed it." Together they form a complete explanation.
Counterfactuals describe the decision boundary. Recourse translates that boundary into practical steps. Use counterfactual as the explanation, recourse as the action plan.
"What if I had X instead of Y?" is a counterfactual question phrased as an inspection dialogue query. The two patterns overlap and can be combined.
The "near miss" counterfactual case is also a form of example: a real or illustrative case that was similar but different in outcome. The patterns reinforce each other.
Wachter et al. (2017) — Counterfactual Explanations without Opening the Black Box
introduces counterfactual explanations as a legally grounded approach to explaining algorithmic decisions. The paper directly motivates the GDPR-aligned use of counterfactuals for adverse decisions
Karimi et al. (2021) — Algorithmic Recourse: Beyond Counterfactual Explanations
makes the critical distinction between counterfactuals (what would change the outcome) and recourse (what the person can actually do). Essential for designing honest counterfactual interfaces
Ustun et al. (2019) — Actionable Recourse in Linear Classification
proposes recourse as a measurable property of classification systems and develops methods for identifying actionable changes. Provides the technical basis for distinguishing controllable from non-controllable factors in the counterfactual display