ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
How do you write principal reasons for an adverse decision that a person can actually use?
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. Name the factors that counted most, attach the person's own values where you can, and write each reason so the reader can do one of three things with it: verify it, correct it, or accept it. Reasons drawn from the decision record — never from a generic list — are what separate an explanation from a brush-off. ReviewDesk's letter generator asks for the principal reasons per applicant, one per line, and renders them as the letter's core.
What is principal reasons?
Principal reasons are the factors that most materially contributed to an adverse outcome for this person — not every factor the system holds, and not the organisation's general criteria. In the Colorado frame they anchor the 30-day explanation; in every frame they are the part of the letter the reader actually reads. A reason is 'principal' if changing it would plausibly have changed the outcome.
What makes a reason 'principal'?
The counterfactual test: would this factor, had it been different, plausibly have changed the decision? Reasons passing that test are principal; the rest is context. Most decisions have between two and four principal reasons. Fewer than two usually means the letter is hiding behind a single blunt factor; more than five usually means the author listed the entire scorecard rather than judging what mattered. The reader's task is to understand their own case, and a short, ranked set serves that task better than a data dump wearing an explanation's clothes.
Source the reasons from the trace, not from memory or the criteria document. The criteria say what could count; the trace says what did count, for this person, in this run. The letter generator's discipline — reasons entered per applicant from the workflow record — exists because letters written from the criteria document describe a hypothetical decision, and recipients can tell. The evidence chain behind the reasons is the subject of the evidence-pack guide.
How specific is specific enough?
Specific enough to be checkable. 'Insufficient income' fails; 'verified monthly income of $2,850 against a requirement of three times the $1,150 rent' passes — the reader can confirm the figure, spot the error if the figure is wrong, or accept the arithmetic if it is right. Values do double duty: they prove the reason came from the person's real file, and they give the correction route its target. Where a value is genuinely not disclosable, give the band or the gap — 'below the required threshold by a narrow margin' — rather than retreating to the bare factor name.
Beware false precision in the other direction. A reason quoting a model score to two decimal places explains nothing, because the reader cannot act on a score. Translate scores back into the factors that compose them: the score is the tool's internal arithmetic; the reasons are the world's facts — history, income, record, availability — that the arithmetic consumed. Letters that confuse the two are the ones recipients describe as 'the computer said no', however many paragraphs surround the score.
How should reasons be ordered and phrased?
Strongest first — the reason that weighed most leads, because readers reasonably assume order signals weight, and burying the decisive factor third reads as concealment. Phrase each reason as a complete factual sentence about the person, not a label: 'Your tenancy history shows two late-payment records in the last twelve months' outperforms 'Payment history: unsatisfactory' in every dimension — clarity, checkability, dignity. Keep one reason per line, as the generator renders them; a paragraph blending three reasons blurs each one's separate truth value.
Tone is accuracy's companion, covered fully in the letter-tone guide. The phrasing rules that matter here: no blame vocabulary ('failed to demonstrate' becomes 'the record shows'), no legal fog, no apology theatre substituting for information. State what the record shows, what it was tested against, and — in the letter's later sections — what the person can do about each reason: correct it if wrong, or understand what would need to change.
What about reasons the person cannot change?
Some principal reasons are historical facts — a past default, a short tenure — that no correction can alter. Honesty still serves the reader: name the reason, name the lookback window the criteria apply, and where your process allows it, say when the factor ages out of consideration. A person who learns 'this record counts until March next year' has been given something real — a timeline — instead of a wall. Where re-application is possible, the letter can say what a stronger future application looks like without promising an outcome.
Never soften an unchangeable reason into vagueness to spare feelings; vagueness reads as arbitrariness, and arbitrariness invites escalation. The respectful form of a hard fact is a clear fact with its boundaries drawn. Reviewers see this in outcomes data: letters with time-bounded reasons generate fewer repeat contacts than letters with fog, because the reader's next question — 'for how long?' — was already answered.
How do reasons stay consistent across notice, trace and letter?
One vocabulary, inherited. The notice promised the factor groups; the trace recorded the factors in those groups' names; the letter's reasons should use the same terms, so the person can follow one thread from warning to outcome. Divergence — notice says 'payment reliability', letter says 'obligation performance index' — breaks the thread and suggests, unfairly or not, that the goalposts moved between the promise and the decision. Generate the letter from the workflow record, as ReviewDesk does, and consistency is structural rather than editorial.
Consistency also disciplines criteria changes. When the factors or thresholds genuinely change, the notice updates first, and letters after the change will honestly differ from letters before it — with the audit trail showing when the change took effect, per the records guide. What a person must never receive is a letter citing factors their notice never mentioned, dated before the notice caught up.
How do you quality-check reasons before letters go out?
Sample and read cold. Each week, a reviewer who did not write them reads a handful of generated letters against three questions. Could I verify each reason against my own facts if this were my letter? Could I correct the correctable ones using only the routes this letter gives? Does the first reason match what the trace says weighed most? Any 'no' is a process fix — a factor description to sharpen, a threshold to disclose, a generator prompt to tighten — not a one-off edit.
Track the downstream signals too: correction submissions reveal which reasons readers find checkable; repeat contacts asking 'but why?' reveal which letters failed as explanations. The generators' structure — decision, disclosure, numbered principal reasons, jurisdiction note, review path — is fixed precisely so that quality work concentrates on the reasons themselves. You can generate a sample free at /try/letter and judge the shape against your current letters; the gap, if any, will be in the reasons, and now you know how to close it.
Your five actions for principal reasons
- Apply the counterfactual test: a reason is principal only if changing it would plausibly have changed the outcome.
- Draw every reason from this person's decision record — the trace — never from the general criteria document.
- Attach the person's own values or the gap to each reason, so it can be verified, corrected or accepted.
- Order strongest first, one reason per line, in the same vocabulary the pre-use notice used.
- Sample sent letters weekly against the three questions: verifiable? correctable? consistent with the trace?
Worked example: two letters about the same decline — only one explains
A fictional lender declines a sole trader's application. Letter A says: 'Your application did not meet our lending criteria, including credit history, affordability and business profile factors.' The applicant learns nothing, can correct nothing, and phones the lender twice before filing a complaint that consumes a week of staff time. Letter B, generated from the trace, says: '1. Trading history of 11 months against a 24-month expectation for this facility. 2. A county court judgment recorded in May last year, shown as unsatisfied in the file we received. 3. Monthly drawings averaging above the level the repayment schedule supports.'
The applicant's response to Letter B is a document, not a phone call: evidence that the judgment was satisfied in November and the file is stale. Correction verified, re-assessment run, facility approved at a reduced amount matched to the drawings — reason 3 honestly standing, reason 2 repaired, reason 1 unchanged and understood. The difference between the letters is not courtesy. It is whether the reasons carried enough of the person's own facts to be acted on.
Frequently asked questions
How many principal reasons should a letter list?
Usually two to four — the factors that pass the counterfactual test for this person. One reason often hides the real story; more than five usually means the whole scorecard was pasted instead of the decisive factors judged.
Should reasons include the person's actual values?
Where disclosable, yes — values let the reader verify, correct or accept each reason. Where a value cannot be shared, give the band or the size of the gap rather than retreating to the factor's name alone.
What if the main reason is a model score?
Translate it. The score is the tool's arithmetic; the reasons are the real-world factors the arithmetic consumed — history, income, record, availability. A score quoted to a reader explains nothing they can act on.
Where do the reasons come from in ReviewDesk?
From the workflow's decision record. The letter generator asks for the applicant, the decision and the principal reasons — entered per applicant from what the decision actually used — and renders them as a numbered list inside the letter, filed in the audit trail.
Is this guide legal advice?
No. ReviewDesk is templates and workflow tooling, not legal advice. Reason-writing is described as ReviewDesk's generator structures it; consult counsel for your situation.
Keep reading: What must you tell job applicants when a hiring algorithm screens them? · How do ADMT notice and review duties apply to automated credit decisions? · What must landlords and screening services disclose when screening is automated? · How do notice and human-review duties apply to automated insurance underwriting and claims? · What notice and review duties arise when software helps make healthcare decisions about patients? · 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.