ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
California ADMT: should you offer the opt-out, the human appeal, or both?
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. The opt-out acts before the decision — the consumer stops the technology being used on them at all. The human appeal acts after it — the decision stands unless a person overturns it. California's regulations allow a qualifying appeal route to substitute for the opt-out in defined circumstances, but 'qualifying' is doing heavy lifting: the appeal must reach a human with the file, the time and the authority to reverse. Many businesses run both routes, and ReviewDesk's California workflow supports that combination in one inbox.
What is the opt-out versus appeal choice?
The opt-out-versus-appeal choice is a design decision inside California's ADMT framework. An opt-out is a pre-decision direction not to use ADMT for a significant decision. A human appeal is a post-decision route to a person empowered to overturn the outcome. The regulations permit the appeal route to stand in place of the opt-out only where its conditions are met; the two routes protect the consumer at different moments and demand different machinery from the business.
How do the two routes protect the consumer differently?
The opt-out protects autonomy before the fact: the consumer who distrusts automated processing, or who knows their file contains something a model will misread, can keep the technology out of their decision entirely. The appeal protects accuracy after the fact: the decision is made efficiently, but a person who believes it wrong can put their case to another person. Neither is strictly stronger. An opt-out cannot help the consumer who never read the notice; an appeal cannot help the consumer whose real objection is to being scored at all. A framework offering only one route has a known blind spot, which is why the regulations treat the substitution as conditional rather than free.
For the business the routes differ in where the human cost falls. The opt-out moves human effort to the front — every opted-out file is human-decided, sight unseen by the tool. The appeal concentrates human effort on the contested minority — typically a small fraction of adverse outcomes. That cost shape tempts businesses toward appeal-only, which is legitimate where the conditions hold, but the saving evaporates if the appeal route is so weak that complaints route around it to regulators instead.
What makes an appeal 'qualifying'?
Substance, tested the same way meaningful review is tested anywhere: a human reviewer, competent for the decision, holding the complete file, with adequate time and genuine overturn authority, producing an answer that engages with the person's grounds. An appeal answered by the team that built the screening criteria, from the score alone, inside a day, with a template paragraph, fails in fact whatever the notice claims. The full standard is the subject of the meaningful human-review checklist — California's appeal is that standard wearing a statutory label.
Process visibility is the second condition in practice. The consumer must be told the appeal exists, at the moment it matters — in the adverse-outcome letter, not only in a notice read months earlier — and must be able to follow its progress. ReviewDesk's applicant status link and SLA clocks give the appeal a visible spine: requested, assigned, in review, decided, each stage timestamped. An appeal conducted in silence reads as an appeal ignored, even when the file shows diligent work.
When is opt-out-only or appeal-only the right design?
Opt-out-only fits decisions where automated processing itself is the sensitivity — contexts where consumers reasonably object to being profiled at all, and where the manual lane is cheap because volumes are modest. Appeal-only fits where the substitution conditions are squarely met and the manual lane for every objector would be genuinely disproportionate — high-volume, low-variance decisions with a strong review bench. Both-only fits most mixed pipelines, and it is the design least likely to need rebuilding when a new decision type or a regulator's question arrives.
Whichever design you choose, choose it per workflow and record it. A business running hiring screening on appeal-only and tenant screening on both routes is not inconsistent; it is calibrated — provided each workflow's notice describes its own arrangement accurately and the worksheet records the reasoning. What fails is the accidental design: the notice says opt-out, the inbox is configured for appeals, and nobody notices until a consumer tests the seam.
What happens when a consumer uses the 'wrong' route?
They will. A consumer will opt out after the decision, or appeal a decision they could have opted out of, or file both at once. The correct response is conversion, not correction: an after-the-fact opt-out becomes an appeal with an explanation; an appeal from someone who meant to opt out is honoured as a review with the opt-out applied to future decisions. ReviewDesk's single inbox makes conversion trivial because both routes are docket rows on the same case — the officer reclassifies, the timeline keeps both events, and the consumer receives one coherent answer instead of a lesson in your taxonomy.
Never let taxonomy leak into the consumer's experience. They asked for a person; which internal lane supplies the person is your concern. The letters generated from the case record handle this naturally, since they are built from the decision and the review outcome rather than from route labels.
How do the routes affect your evidence pack?
Each route generates its own evidence shape. Opt-outs evidence restraint: the request, the unscored file, the human decision. Appeals evidence responsiveness: the grounds, the review, the outcome. The three-regime evidence engine projects both into the California pack as recorded — and, per ReviewDesk's discipline, an element that was not recorded renders as 'not recorded', never reconstructed. That honesty is protective: a pack showing a genuine gap invites a fix; a pack polished into fiction invites a catastrophe when the underlying inbox tells a different story.
The annual risk assessment consumes the same data. Route volumes, overturn rates and decision-time comparisons between the manual and automated lanes are exactly the material an assessment wants — see the risk-assessment guide. Businesses that run both routes with records find the assessment largely assembles itself; businesses that ran the routes from memory find themselves writing fiction at a deadline.
How do you explain the choice in the pre-use notice?
Briefly, concretely and without legal choreography. If both routes exist: 'You can ask us not to use the automated tool for your application — a person will decide instead. If a decision goes against you, you can ask a person to review it.' If appeal-only under the substitution: say that the tool will be used, and that every adverse decision can be appealed to a person who can change it, with the route named. The notice fails if a reader cannot tell which of those two worlds they are in. Draft the block in the notice builder and read it aloud; if it survives being read aloud to a non-lawyer, it is close.
Then keep the promise mechanically aligned: the notice's wording, the inbox's configuration and the workflow's recorded route must be three copies of one fact. ReviewDesk generates the notice from the workflow for that reason. When the arrangement changes — a new decision type, a route added — change the workflow first and let the notice follow, rather than editing the notice and hoping operations noticed.
Your five actions for the opt-out versus appeal choice
- Decide per workflow which routes you offer — opt-out, qualifying appeal, or both — and record the determination with its reasoning.
- If you rely on the appeal substitution, verify the appeal meets the substance test: full file, competent reviewer, time, overturn authority.
- Write the notice so a reader can tell in ten seconds which world they are in, with both routes named if both exist.
- Configure conversion, not correction: a misdirected request is reclassified on the same docket, never bounced back to the consumer.
- Align notice wording, inbox configuration and the workflow record as three copies of one fact.
Worked example: a benefits platform runs both routes, deliberately
A fictional employee-benefits platform — ClearPath Benefits — verifies eligibility for employer-funded programmes in California using automated checks against employment and enrolment data. For eligibility screening it offers both routes. The opt-out draws a steady trickle: of 2,300 checks in a quarter, 58 employees opt out and are verified manually from documents. The appeal route handles the sharper end: 190 adverse determinations, 23 appeals, 6 overturned — mostly stale enrolment data the automated check could not have known was stale.
The two routes catch different failures, exactly as designed. The opt-outs cluster among long-tenured employees with complex records — people who correctly predicted the tool would misread them. The appeals cluster around data freshness. ClearPath's risk assessment writes itself from those numbers, and its notice can honestly offer both doors because both doors demonstrably open. A competitor offering an appeal-only route with a two-line template answer is not running the same programme, whatever its notice says.
Frequently asked questions
What is the difference between the opt-out and the appeal?
Timing and effect. The opt-out is exercised before the decision and prevents ADMT being used for that consumer at all. The appeal is exercised after an adverse decision and asks a human to overturn it. Both routes end with a person deciding; they differ in when the person enters.
Can the appeal replace the opt-out entirely?
Only where the regulations' conditions for the substitution are met — centrally, that the appeal is a genuine human review capable of overturning the decision. Businesses relying on the substitution should record the determination per workflow and keep the appeal's substance auditable.
What if a consumer files both an opt-out and an appeal?
Honour both: apply the opt-out to future decisions and treat the appeal as the review of the decision already made. In ReviewDesk both are events on one docket, so the consumer gets one coherent answer.
Which route costs more to run?
The opt-out costs human effort on every opted-out file, decided manually. The appeal costs effort only on contested adverse outcomes, usually a small minority. Many businesses run both because they catch different failure types.
Is this guide legal advice?
No. ReviewDesk is templates and workflow tooling, not legal advice. The routes are summarised from the CPPA ADMT regulations as reported in our research; consult counsel for your situation.
Keep reading: How long must you keep ADMT decision records — and what should the audit trail hold? · What goes into California's annual ADMT risk assessment (due 1 April 2028)? · How does one decision trace become Colorado, California and EU evidence packs? · What is a hash-chained decision record — and why does immutability matter as evidence? · How do you run the scope test: does your system materially influence or substantially replace human decision-making? · All ReviewDesk guides · ReviewDesk home · Try the free desk · Free letter generator · Free notice builder · Pricing
ReviewDesk is built by Ingenious AI. Templates and workflow tools — not legal advice.