ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
How do you explain an adverse automated decision within Colorado's 30 days?
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. When an automated decision goes against a Colorado consumer, they are entitled to an explanation within 30 days: what was decided, the principal reasons, the type of data that drove those reasons, how to correct wrong data, and how to ask a person to review it. ReviewDesk generates that explanation as a letter from the workflow's own record — applicant, decision, reasons — and its paid plans run the 30-day clock so the deadline is a countdown you watch, not a date you discover.
What is Colorado's 30-day adverse-outcome explanation?
Colorado's 30-day explanation is the post-decision half of SB 26-189's transparency pair. The pre-use notice warns the person before the tool runs; the adverse-outcome explanation tells them afterwards, within 30 days of the adverse decision, why the outcome went against them and what they can do next — correct their data, and request meaningful human review. It is owed for adverse consequential decisions, not for approvals.
What exactly does the 30-day clock measure?
Thirty days from the adverse decision — not from the end of the month, not from when someone in operations notices the file. That precision matters because automated pipelines make decisions continuously, including weekends and holidays, while explanation processes often run in weekly batches. A decision made on the first of the month and batched on the twenty-fifth has already consumed most of its window before a human sees it. The operational answer is to treat the letter as part of the decision event itself: when the adverse outcome is recorded, the explanation task is created with its deadline attached.
On paid plans ReviewDesk's inbox runs statutory clocks — countdowns and overdue stamps — precisely because a deadline distributed across inboxes and spreadsheets is a deadline waiting to be missed. The free workflow still timestamps every step in the audit trail, which lets you reconstruct compliance after the fact, but reconstruction is a weaker position than a clock that was visibly running. The wider deadline landscape, including California's assessment date, is mapped in the deadlines calendar.
What must the explanation contain to be worth the paper?
Five elements, in the order a reader needs them. The decision itself, stated plainly — not 'your application was unsuccessful at this time' but what was actually decided. The principal reasons: the factors that counted most against this person, specific enough that they can tell whether the reasons are accurate. The type of data behind those reasons, so an error can be located. The correction route, with a realistic way to submit corrected information. And the human-review offer, with the exact route to request it and the assurance that a person, not the tool, will conduct it.
Generic reason lists fail the specificity test that matters — the reader's. 'Did not meet our criteria' explains nothing and invites the reader to assume the worst. Contrast 'your application showed two missed rent payments in the last twelve months and current income below three times the monthly rent' — a statement the reader can verify, dispute or accept. Writing reasons at that resolution is a skill covered in the principal-reasons guide, and the letter generator enforces the structure by asking you for the reasons per applicant, one per line, every time.
Who writes the reasons — and from what source?
The reasons must come from the decision record, not from a communications template. In ReviewDesk the workflow's trace holds what the tool actually used — the factors recorded for that decision — and the letter is generated from that record plus the reasons the decision owner enters. That chain matters for two reasons. First, accuracy: a letter inventing plausible reasons the tool never used is worse than no letter, because it misleads the reader and corrupts the file. Second, consistency: the notice promised certain factors, the trace recorded certain factors, and the letter should speak about the same factors in the same words.
Where the trace shows a missing element, ReviewDesk's evidence discipline renders it as 'not recorded' rather than inventing a value — the principle examined in the not-recorded guide. Apply the same honesty inside the letter: if a factor cannot be reconstructed for a legacy decision, say the explanation rests on the records held, and let the human review do the deeper work. Readers forgive gaps admitted plainly far more readily than gaps dressed as certainty.
How does correction fit inside the 30 days?
Correction is not an appeal of the decision; it is a repair of the input. The explanation should distinguish the two routes clearly, because they have different outcomes: corrected data leads to a fresh assessment on accurate facts, while human review re-examines the decision as made. Many adverse outcomes are data stories — a transposed digit, a namesake's record, an out-of-date balance — and a fast correction route resolves them without either side entering a dispute posture. Make the route concrete: what to send, where to send it, and what happens next, including a fresh decision where the corrected data changes the picture.
Operationally, corrections should pause the argument, not the clock. Log the correction request against the same docket row as the decision, re-run the assessment on corrected data where that is feasible, and record the new outcome in the audit trail. The worked mechanics live in the correction guide. A business that handles corrections gracefully discovers that many potential complaints convert into accepted, accurate decisions — sometimes approvals.
What does a good process look like week to week?
Decisions flow in continuously; explanations flow out in a rhythm. A workable rhythm for a small team: adverse outcomes generate letters the same day from the workflow record, a named reviewer checks reasons against the trace before sending, and the inbox board shows every open explanation with its remaining days. Weekly, someone reviews the overdue and near-due rows — on paid plans the overdue stamps make that a glance rather than a query. Monthly, sample five sent letters and read them as a recipient: could you, from this letter alone, understand what happened and what to do next?
That sampling habit catches drift early — reasons getting vaguer, correction routes going stale, review offers shrinking to a footnote. It also builds the evidence that the process is real: dated letters, logged reviews, recorded outcomes. When the three-year record is exported from the evidence vault, it should read as a system that ran every week, not a folder assembled the week an enquiry arrived. The vault mechanics are covered in the evidence-vault guide.
What goes wrong when teams improvise the explanation?
The classic failure is the merged letter: one template attempting to be a rejection, an explanation, a marketing message and a legal disclaimer simultaneously, succeeding only as a rejection. The second failure is delay disguised as care — the letter held for legal review until day twenty-nine, then sent in a form so hedged that the reasons disappear. The third is channel chaos: explanations sent from personal inboxes, unlogged, so the audit trail cannot prove which version a person received or when.
Every one of those failures is a workflow failure, not a writing failure, which is why ReviewDesk treats the letter as a generated artefact of the decision record: dated, numbered, stored, downloadable as a .txt, and filed in the audit trail at generation. You can generate one free right now — enter an applicant, a decision and the reasons, and see the complete letter shape before you ever owe one for real.
Your five actions for Colorado's 30-day adverse-outcome explanation
- Create the explanation task at the moment the adverse decision is recorded — the 30-day clock is already running.
- State the decision plainly, then the principal reasons, specific to this person and drawn from the decision record.
- Name the type of data behind the reasons and give a concrete correction route with a next step.
- Offer human review in the same letter, with the exact request route and the promise that a person conducts it.
- File the sent letter — numbered, dated, logged — and watch open explanations on one board with their remaining days.
Worked example: a lender's 30-day window, run properly
Imagine a fictional community lender — Solstice Credit Union — whose tool bands small-business loan applications. On 3 March the tool places an application from a food-truck operator in the decline band, principally for two reasons recorded in the trace: twelve months of trading history against a twenty-four-month expectation, and a county filing the parser attributed to the business that belongs to a namesake firm. The explanation letter is generated the same day from that record: decision, both reasons, the data types behind them, the correction route and the review offer — comfortably inside a window that closes on 2 April.
The operator uses the correction route, supplying the filing record showing the namesake error. The corrected data is logged against the same docket, the application is re-assessed, and on 19 March — day sixteen — the outcome changes to a smaller approved facility. The review request he had also filed is closed as resolved by correction, with the whole sequence in the audit trail. The 30-day duty produced, in this case, a customer instead of a complaint — because the explanation arrived fast enough, and specifically enough, for the error to be found.
Frequently asked questions
When does Colorado's 30-day period start?
From the adverse decision itself. Because automated pipelines decide continuously, the safe practice is to generate the explanation as part of the decision event, with the deadline attached — ReviewDesk's paid plans show the countdown and stamp overdue rows.
What reasons are specific enough?
Reasons the reader can verify against their own facts: the factor, and where possible the value or threshold involved. 'Insufficient history' is weak; 'twelve months of trading history against a twenty-four-month expectation' lets the reader confirm, correct or accept it.
Is correction the same as human review?
No. Correction repairs the input data and can lead to a fresh assessment; human review re-examines the decision as made. A good explanation letter offers both routes and explains the difference in one sentence each.
Does the free tier generate these letters?
Yes. The letter generator is part of the free workflow and also runs with no account at all. Paid plans add the statutory clocks, multiple workflows and the evidence vault.
Is this guide legal advice?
No. ReviewDesk is templates and workflow tooling, not legal advice. Duties are summarised from Colorado SB 26-189 as reported in our research; consult counsel for your situation.
Keep reading: What makes a human review of an automated decision 'meaningful'? · How does the California ADMT opt-out work for consumers and businesses? · California ADMT: should you offer the opt-out, the human appeal, or both? · 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)? · 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.