Act
Contestability & Appeal
How can I challenge this decision?
Gives people a working route to challenge a decision and get a human review with a documented response.
Overview
Why it matters
A right you can’t find is a right you don’t have.
People affected by decisions that AI contributed to may have a legal right to challenge them, and good practice is to offer a way anyway. From 20 November 2026, in consumer credit, people can ask for human intervention and a review where the creditworthiness assessment uses automated processing (Consumer Credit Directive, Art. 18(8)). Any such right only works with a visible path: a button, a person who reviews, and a response.
Unlike Human-in-the-Loop, where a person decides each case inside the process, this pattern is used by the affected person, after the decision. Unlike Actionable Recourse, which helps with a future decision, it challenges the decision that was made.
What does it look like?
Applicant challenging a declined credit application (case
Applicant challenging a declined credit application (case
- Applicant sees: “Declined by a caseworker on 29 September. Reason: employment records were not received by the deadline.”
- Below the decision: “Disagree? You can ask for a review by a different caseworker.”
- Applicant clicks “Ask for a review.”
- Form appears, prefilled with the case summary, with room for the applicant’s own view and “Add your employment records (missing from your file)”.
- On 30 September at 08:02 the applicant submits and sees: “Received. Reference #4821-R1. A different caseworker will review your case within 5 business days. If you send the missing records, the review reassesses your case. You can also apply again.”
- Applicant can track the status in their account: received, under review, decision.
Content moderation challenge
Content moderation challenge
- User’s post is removed by an automated moderation system.
- Notification shows: “Your content was removed for [reason]. If you believe this was a mistake, you can appeal.”
- User submits appeal with context the automated system didn’t have.
- Human moderator reviews within 24 hours.
- Content is restored, or decision is upheld with an explanation.
Escalation
Escalation
- First-level review upholds the original decision.
- User is not satisfied.
- Interface shows: “If you remain dissatisfied, you can contact [ombudsman or conciliation board] at [link] or file a complaint with [financial regulator].”
- User has a documented reference number from the original challenge to use in further escalation.
When does it fit?
Use it when
- Decisions affect rights, access to services or finances
- Law or regulation gives people a right to review or contest
- Decisions may rest on wrong, outdated or incomplete data
Skip it when
- The decision is clearly low-stakes
- There is no real review behind it yet: build the review first. Where the law gives a right to review, people still need a clear way to ask for one.
Design
How do I apply it?
-
Show the path right after the decision.
That’s when people are most likely to act.
-
Never harder than applying.
If applying took ten minutes, challenging shouldn’t take an hour. Prefill what you already know.
-
Guarantee a human review.
A review by the same system, or the same person, is a repetition, not a review.
-
Give a timeline, and keep it.
A challenge that disappears without updates can feel worse than no path at all.
-
Log every challenge.
Each one shows where people believe the system or the process was wrong. See Audit Trail & Logging.
For your operations team
The interface is only as good as the process behind it. Agree review capacity, response times and escalation contacts before launch, and record every challenge as an event in the audit trail.
How can I show it?
Challenge action
“Ask for a review”, visible on the decision screen.
A clearly labelled, persistent action available after any adverse decision: “Disagree? Ask for a review.” Not buried in a footer or help section. Visible on the decision screen.
Human review promise
A clear statement that a different person reviews it.
An explicit statement that a person reviews the challenge: someone other than the person who made the first decision, not the same system running again. This is the core of meaningful contestability.
Status tracker
Received, under review, decision.
After submission, a status view showing where in the process the challenge is: received, under review, decision.
What happens next
Who reviews, how long, possible outcomes.
Immediately after submitting a challenge, a clear explanation of the review process: who reviews it, what the timeline is, what the possible outcomes are. Sets expectations and can reduce anxiety.
Escalation path
Ombudsman or regulator, if the review fails.
If the first-level review is unsatisfactory, information about further escalation, whether an ombudsman or conciliation board, a regulator, or a court, is made visible. People should not have to research this on their own.
Reason form
Prefilled, with room for context and documents.
A structured form with a reason for the challenge (dropdown and free text), room for the person’s own view, the option to upload supporting documents, contact information for follow-up, and a submission confirmation with a reference number. Prefill it with the case summary.
What does it cost?
Access vs. gaming
Some people will use an open path to delay decisions or to try again without new information. Closing the path for everyone fails the people with a real case. Design for proportion, for example with a short reason field, and keep the path open.
Real review vs. compliance theater
A form with no review behind it ticks a box and fails people. It also leaves the organization unable to show that challenges are actually handled. Staff the review before you show the button, and be honest about capacity.
Speed vs. thoroughness
A review that takes six months no longer helps the person. A rushed one is a repetition, not a review. Set a timeline you can keep, such as five business days, and design the process and the interface together.
Connections
Which patterns work with it?
Human-in-the-Loop
Who reviews this recommendation, and can they override it?
In HITL, a person decides each case inside the process. In Contestability, the affected person challenges a past decision. When caseworkers decide, the review should go to a different caseworker.
Actionable Recourse
What can I do to change this outcome?
Recourse helps people work toward a different future outcome. Contestability challenges the specific past decision. Keep them apart in the interface.
Audit Trail & Logging
What is the complete record of this decision?
Every challenge should be logged. The audit trail gives the reviewer the record to work with.
Decision Timeline
What happened and when?
Showing the decision timeline to the person challenging the decision can help them understand what happened and build a substantive case.
Sources (5)
Sources (5)
Lyons et al. (2021) — Conceptualising Contestability: Perspectives on Contesting Algorithmic Decisions
analyzes public submissions to Australia’s proposed AI Ethics Framework to examine what people expect contesting an algorithmic decision to involve, finding contestability is broadly understood as a protection for individuals comparable to challenging a human decision. Published in Proceedings of the ACM on Human-Computer Interaction, 5 (CSCW1)
European Union (2016) — General Data Protection Regulation (GDPR). Article 22
gives a right to human intervention, to express a view and to contest, for decisions based solely on automated processing, with legal or similarly significant effects, that are allowed because they are needed for a contract or based on explicit consent (Art. 22(3)). Where a person decides, it usually does not apply, but a score that decisively drives the decision can count (CJEU, C-634/21, SCHUFA, 2023)
European Union (2024) — EU AI Act, Article 86
people affected by a decision based on the output of a high-risk system listed in Annex III, such as credit scoring, can ask for a clear and meaningful explanation; deployers must also inform them that such a system is used (Art. 26(11)). Applies from 2 December 2027 (systems already on the market only after a significant design change). A right to an explanation, not a right to contest
Kroll et al. (2017) — Accountable Algorithms
argues that transparency alone can’t make automated decisions accountable, and shows how technical tools can prove that each decision followed announced rules, which gives reviewers evidence to work with. Published in University of Pennsylvania Law Review, 165, 633
European Union (2023) — Directive (EU) 2023/2225 on credit agreements for consumers (Consumer Credit Directive), Article 18(8)
where a creditworthiness assessment uses automated processing of personal data, the consumer can request human intervention, a clear explanation of the assessment, the chance to give their own view, and a review. Applies from 20 November 2026.
Your only appeal path is a “Contact us” form.
Or a regulator has asked how affected people can challenge your system’s outcomes. That’s where this pattern starts. Let’s look at your interface together.
Work with me