ReviewDesk guides · By the Ingenious AI team
Guides to ADMT notices, adverse-outcome letters and human review
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. These guides explain, in plain language, the duties ReviewDesk is built around: notice before an automated decision tool is used, a clear explanation after an adverse outcome, a meaningful human review when someone asks, California opt-out handling, and records kept for three years — under Colorado SB 26-189, the California CCPA ADMT regulations and Illinois HB 3773. Every guide ends with five concrete actions and a worked example. ReviewDesk itself is free for one workflow: a hosted pre-use notice page, the adverse-outcome letter generator, the human-review inbox and a timestamped audit trail.
What must a Colorado pre-use notice for an automated decision tool say?
Colorado SB 26-189 requires notice before an automated decision tool is used. What the notice must cover, when it is due, and how a hosted notice page holds it.
What must a California ADMT pre-use notice include?
California's CCPA ADMT regulations require a pre-use notice with opt-out and appeal rights. What belongs in it, and how it differs from Colorado's notice.
What does Illinois HB 3773 require when AI is used in employment decisions?
Illinois notice duties for AI in employment are already in force. Who must be told, what the notice covers, and how it sits beside Colorado and California.
What is automated decision-making technology (ADMT)?
A plain-language definition of ADMT: what counts, what does not, and why hiring, credit, housing and insurance tools sit at the centre of the new rules.
How do you explain an adverse automated decision within Colorado's 30 days?
Colorado expects an explanation of an adverse outcome within 30 days: the decision, the principal reasons, the data, correction and review. How to produce it on time, every time.
What makes a human review of an automated decision 'meaningful'?
A human review is meaningful when a person with authority, information and time can actually change the outcome. The checklist that separates review from rubber-stamping.
How does the California ADMT opt-out work for consumers and businesses?
California gives consumers a right to opt out of automated decision-making for significant decisions. How the opt-out works, what businesses must build, and the exceptions.
California ADMT: should you offer the opt-out, the human appeal, or both?
California lets some businesses rely on a qualifying human appeal instead of an opt-out. How the two routes differ, what each demands operationally, and why many run both.
How long must you keep ADMT decision records — and what should the audit trail hold?
ReviewDesk keeps records for three years. What the audit trail must contain per decision — notice, factors, letters, reviews — and how to keep it usable, not just stored.
What goes into California's annual ADMT risk assessment (due 1 April 2028)?
California's ADMT framework pairs notices with an annual risk assessment — ReviewDesk's pack is due 1 April 2028. What it examines and how the workflow record feeds it.
How does one decision trace become Colorado, California and EU evidence packs?
File one hash-chained decision trace per automated decision and project it into sealed Colorado, California and EU evidence packs. How the ReviewDesk evidence engine works.
What is a hash-chained decision record — and why does immutability matter as evidence?
Each ReviewDesk decision trace links cryptographically to the one before it, so altered history is detectable. A plain-language explanation of hash chains and corrections.
How do you run the scope test: does your system materially influence or substantially replace human decision-making?
The worksheet question at the heart of ADMT scope: materially influence or substantially replace? How to assess acceptance rates, review depth and real human discretion.
How should you prepare EU Annex III evidence for automated decisions — before the duties apply?
ReviewDesk's third pack frames decision traces for the EU Annex III file, labelled preparatory until 2 December 2027. What to gather now and what not to claim yet.
How do you let someone correct wrong data after an adverse automated decision?
Wrong data is the most fixable cause of wrong decisions. How to build the correction route Colorado expects: what to accept, how to re-assess, and how to record it.
How do you write principal reasons for an adverse decision that a person can actually use?
The principal reasons are the heart of every adverse-outcome letter: specific, sourced from the decision record, and written so the reader can verify, correct or accept them.
What must you tell job applicants when a hiring algorithm screens them?
Hiring is the most common ADMT use in the fleet's regimes: Illinois duties are live, Colorado and California follow in 2027. Notice, reasons and review for applicant screening.
How do ADMT notice and review duties apply to automated credit decisions?
Credit banding and lending tools are consequential decisions in every ReviewDesk regime. Notice before scoring, principal reasons after a decline, correction and review.
What must landlords and screening services disclose when screening is automated?
Tenant screening is consequential, data-error-prone and often outsourced. Notice duties, reason quality, correction routes and the landlord-versus-screener split.
How do notice and human-review duties apply to automated insurance underwriting and claims?
Underwriting banding and claims triage are automated decisions about people with real consequences. Notice, factors, review by someone who can re-underwrite or re-handle.
What notice and review duties arise when software helps make healthcare decisions about patients?
Triage, eligibility and scheduling tools shape care. Handling the ADMT pattern — notice, reasons, human review — with the sensitivity healthcare demands.
What must schools and universities disclose when software screens applicants or students?
Admissions screening, scholarship allocation and student-progress flags are consequential decisions. Notice for applicants, fairness records, and review that respects academic calendars.
Do you have to tell employees when monitoring or productivity software scores them?
Productivity scoring, shift allocation and performance analytics decide things about employees. Applying the employment notice pattern to the existing workforce.
How can a tenant applicant request human review of an automated screening decision?
Written for applicants: what to ask for, what to send, and what a proper review of a screening decision should involve — and for landlords, how to receive it well.
How should a human-review inbox be designed so no request is missed?
One docket for every request — review, correction, opt-out, appeal — with timestamps, assignment and outcomes. The design principles behind the ReviewDesk inbox.
How do SLA clocks keep human reviews inside their deadlines?
Statutory windows like Colorado's 30 days fail quietly without machinery. How ReviewDesk's SLA clocks — deadlines with provenance, escalation on read — keep cases on time.
What should happen when a human-review request goes overdue?
Escalation computed on read, emitted once, with a human note: what overdue stamps mean, how reassignment works, and how to talk to the applicant whose case is late.
How does an applicant status link change the experience of waiting for a review?
A live status page for a review case — requested, assigned, in review, decided — removes the silence that generates complaints. How ReviewDesk's status link works.
How should human-review cases be assigned across a team?
Named owners, matched competence, visible queues and clean reassignment: the assignment discipline that keeps a review bench fast, consistent and auditable.
What should a review case's decision history record?
Who reviewed, when, with what file and what outcome — the decision history is how a review programme proves itself. What belongs in it and how ReviewDesk keeps it.
How does the 3-year evidence vault export a complete decision file?
One export per decision file: notice version, trace, letters, requests, outcomes — sealed and readable. How ReviewDesk's paid-plan vault works and when to use it.
What should you ask a software vendor before you rely on their tool for consequential decisions?
Your vendor's tool, your duty. The questions that reveal whether a screening or scoring product can support notice, explanation, correction and review — before you sign.
ADMT duties explained for software vendors and the businesses that deploy their tools
Two sides of one decision: vendors build the tools, deployers make the decisions. Who owes what — notice, factors, records, review — when software decides about people.
How do you set up your first ADMT workflow, step by step?
From inventory to live notice in one pass: define the decision, record the scope call, draft the notice, wire the inbox, test the letter, and start the audit trail.
What makes a hosted pre-use notice page work better than a PDF or a policy link?
One live address, one current version, a working review form, and a version history: the practices that make a hosted notice page the strongest form of the disclosure.
How should the review-request form on a notice page be designed?
The form is where a right becomes a request. Fields that route without interrogating, expectations set at filing, and requests born classified into the inbox.
How do you write an ADMT notice in plain language that people actually understand?
Notices fail when they read like legal defences. The writing discipline — reader first, factors in world-words, rights as actions — behind notices and letters people can act on.
How do you write an adverse-outcome letter that de-escalates instead of inflaming?
The letter is read at a bad moment. Tone choices — lead with the decision, own the reasons, show the working, keep the door open — that lower temperature without lowering honesty.
What does review coverage mean — and what breaks when requests go unanswered?
Coverage is the share of review requests actually answered with a recorded outcome. What unanswered requests break — legally, operationally and reputationally — and how to keep coverage at whole.
How do you answer every human-review request — without sounding like a bot?
The operating method behind full coverage: route to one docket, decide on the record, generate the letter from it, and keep the audit trail — the ReviewDesk way, end to end.
How do you run ADMT workflows across Colorado, California and Illinois at once?
One tool, three regimes, no copy-paste: shared factor sources, jurisdiction-specific notices and letters, and one desk where the whole estate stays visible.
How can an agency run ADMT desks for multiple client businesses?
Team plans never expire for agencies and resellers, with multiple businesses under one desk. Keeping each client's notices, inboxes and records separate — and provable.
How can HR-tech resellers offer ADMT notice and review tooling to their clients?
Resellers sit between screening software and employers. Packaging ReviewDesk workflows — notices, letters, inboxes, evidence — as the compliance layer your platform lacks.
How can a small business meet ADMT notice and review duties without a compliance department?
For owners who are also the HR department: the minimum viable programme — one workflow, one inbox owner, real letters — sized to a small business's week.
What should a startup build first when its product or hiring uses AI screening?
Startups meet ADMT twice: as employers screening candidates, and as vendors whose product screens for customers. Building the evidence habits while both are still small.
A landlord's step-by-step plan for compliant automated tenant screening
The housing guide in checklist form: choose the workflow, write the notice from your real criteria, generate decline letters with records named, run corrections at letting speed.
How should a fintech lender build its ADMT workflow — from application flow to evidence pack?
A build playbook for digital lenders: notice placement in the application journey, trace instrumentation at decision time, correction SLAs and the diligence-ready evidence pack.
What are a staffing agency's duties when it screens candidates with automated tools?
Agencies screen for many employers at once — the notice, letter and review duties multiply across clients. Running agency screening on per-client workflows without mixing records.
How do automated-decision duties apply to gig platforms and marketplaces?
Deactivation, allocation and rating systems decide gig workers' livelihoods at scale. Applying the notice, explanation and review pattern to platform work.
Why does ReviewDesk render missing evidence as 'not recorded' — and never invent it?
The honesty rule at the centre of the evidence engine: a missing element is shown as a gap, in every pack and report. Why fabrication is the one unforgivable evidence sin.
How do you integrate decision traces through the ReviewDesk workspace API?
The integration contract for the evidence engine: workspace API key, one trace per decision at decision time, fields, failure handling and daily reconciliation.
Who should own an ADMT programme — owner, reviewers and record-keepers?
Programmes fail by diffusion: everyone involved, nobody accountable. The three roles every ADMT programme needs named — programme owner, reviewers, record-keeper — at any scale.
What are the key ADMT deadlines — Illinois now, Colorado and California 2027, the EU in December 2027?
Every date in ReviewDesk's world on one page: Illinois already live, California regulations in effect 2026 with compliance 2027, Colorado from 1 January 2027, the first risk assessment 1 April 2028, EU duties 2 December 2027.
What changed with Colorado SB 26-189 — and what does it keep from the earlier AI Act approach?
Colorado replaced its first AI Act with SB 26-189, effective 1 January 2027. What the replacement keeps — notice, explanation, review — and where to focus preparation.
What is the California ADMT compliance timeline — from the 2026 regulations to the 2028 assessment?
Regulations in effect 1 January 2026, compliance duties from 1 January 2027, first annual risk assessment 1 April 2028: California's ADMT sequence and what to build in each phase.
What does ReviewDesk's free tier actually cover — and when do you need Pro or Team?
Free does the complete job for one workflow: hosted notice, letter generator, inbox, audit trail. What Pro and Team add — clocks, opt-out handling, vault, traces, multiple businesses.
What does a good adverse-outcome letter actually look like? (Worked examples)
Three annotated examples — hiring, tenancy, credit — showing the five-part structure in action: decision, tool disclosure, principal reasons, jurisdiction note, review path.
How do you respond to a human-review request, step by step?
The handling sequence from filing to closure: acknowledge, assemble, assign, decide, communicate, file — with the clock running and the person informed at each stage.
What happens after a human review decides — for the applicant and for the business?
The decision is the midpoint, not the end: outcomes communicated, corrections propagated, overturns honoured downstream, and the case's lessons fed back into the pipeline.
What should an annual ADMT programme review actually examine?
Beyond the statutory risk assessment: the yearly self-examination of notices, letters, coverage, corrections and scope that keeps a programme honest as it ages.
How do you train hiring managers to work inside an ADMT notice and review process?
Managers improvise where training is absent — and improvisation is where inconsistent explanations are born. The one-hour curriculum that aligns a hiring team with its notice.
How do you choose ADMT compliance software? (A buyer's checklist)
Evaluate any notice-and-review platform against the duties themselves: generation from the record, inbox discipline, clocks, evidence honesty and exit. Written for buyers, not vendors.
How should you handle a formal complaint about an automated decision?
A complaint is the review system's stress test arriving by another door. Triage it into the same docket, answer from the file, and let the record do the defending.
How do you prepare for future automated-decision rules without knowing what they will say?
More jurisdictions and decision types will follow the Colorado-California-Illinois pattern. The preparations that pay under any future text: records, scope discipline and honest notices.
ReviewDesk is built by Ingenious AI. Templates and workflow tools — not legal advice. See ReviewDesk home, try the free desk and pricing.