ReviewDesk
← Back

ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026

How do you respond to a human-review request, step by step?

Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing

The short answer. Acknowledge with a reference and a timeframe. Assemble the full case before anyone judges it. Assign a competent reviewer with authority. Decide among three real outcomes and write the note that engages the person's own grounds. Communicate from the record, close both halves — outcome recorded, reply sent — and let the audit trail keep the whole sequence. ReviewDesk's inbox runs this sequence as its native shape; this guide walks it as a handling standard for any team.

What is responding to a review request?

Responding to a human-review request is the end-to-end handling of one case in the review inbox: receipt and acknowledgement, case assembly, assignment, review and decision, communication of the outcome, and filing in the decision history. In ReviewDesk each step is a docket event with a timestamp, the case carries its SLA clock on paid plans, and the applicant follows progress through their status link.

Step one: acknowledge like you mean it

Within your stated target — two working days is a sound default, same day is better — send the acknowledgement that does three things: confirms the filing exists (reference, date), states what happens next and by when, and hands over the status link. The acknowledgement is not the answer and must not impersonate one ('we have received your request and it will be reviewed' is complete; anything about merits belongs to the decision). Its real function is to stop the clock in the person's head — the anxiety clock, which starts at filing and generates the calls, duplicates and complaints that unmanaged silence produces.

File the acknowledgement's text to the case. The timeframe promised in it becomes part of the case's fairness record, and the status link delivered in it is the channel through which your later stages become visible without further correspondence. The pattern's full rationale is in the status guide; the handling point here is simpler: no case should ever be older than its acknowledgement target without its person having heard from you. That single rule, kept, transforms a queue's reputation.

Step two: assemble the case before judgement

The reviewer should never gather their own case. Assembly means the docket presents, in one view: the person's request verbatim, the original submission, the decision record with its factors, the notice version, the letter sent, and any corrections already in play. Assembly is where ReviewDesk's single-desk design pays its rent — all five artefacts already share the workflow — and where multi-system organisations must do deliberate work: a named assembler (often the inbox owner in the morning sweep) completes the file before assignment, chasing the missing screening export or bureau record so the reviewer's hour is spent on judgement.

Check assembly against the completeness question: could a competent stranger decide this case from this file alone? If a material artefact is missing, the case is not ready, and assigning it anyway manufactures a mid-review stall — the commonest hidden cause of deadline pressure. Stalls discovered at assembly are cheap; stalls discovered at day twenty-five are escalations. The distinction is a morning habit, not a system feature.

Step three: assign for competence and authority

Match the case to a reviewer who can judge the decision type and who holds the authority its outcomes require — restore, overturn, adjust — per the assignment guide. Note the assignment on the case with its timestamp; the clock's provenance and the assignment together tell the case's whole management story. Where the only competent reviewer made the original decision, record that fact and proceed as a distinct review act, per the checklist — small teams are a reality, undocumented conflicts are a choice.

Brief the reviewer with the case, not with a conclusion. 'This one looks like a namesake error' is a helpful assembly note if it points at evidence; it is contamination if it pre-decides. The line is whether the note cites the file or predicts the outcome. Keep assembly notes factual and sourced, and let the reviewer's fresh reading be genuinely fresh — the entire value of the institution depends on the second look being a look, not a nod.

Step four: decide, and write the note as the decision's public face

Three outcomes, all real: uphold with reasons, overturn, or repair the data and re-assess — with the re-assessment's result itself decided and recorded. The note engages the person's grounds in their order, names the evidence weighed, identifies what proved decisive, and states the outcome without fog. Length follows difficulty; the test is whether the person, reading it, can see their own case in it — the documents they sent, the argument they made — answered rather than processed. Notes that fail this test are the leading indicator of review theatre, and sampling them is leadership's quality control, per the history guide.

Where the decision needs evidence still outstanding — a bureau confirmation, a court record — do not let the case idle silently: escalate per the escalation guide, with the note recording what is awaited, the chase dates and the next update promised to the person. A decided-late case with a managed middle beats a decided-on-time case whose middle was a mystery, in every downstream forum where the two are compared.

Step five: communicate from the record and close both halves

Generate the outcome communication from the case record — the note becomes its substance, the letter machinery carries the jurisdiction frame and the remaining routes — and send it through the channel the person used, with the status page updated to decided with the full outcome. Then close the case in the docket only when both halves exist: outcome recorded, communication sent and logged. The two-half rule from the inbox guide sounds clerical and prevents the two ugliest failure shapes in this field: the decided case the person never heard about, and the promised outcome the file cannot produce.

Offer the remaining doors accurately in the closing letter: correction of anything still wrong, the position on re-application where relevant, and how the person can obtain their file. A closed case with its doors correctly described generates almost no further correspondence; a closed case with vague doors generates the correspondence anyway, now with grievance added. Precision at closure is the cheapest correspondence management there is.

Step six: file, count and learn

The completed case joins the decision history and the programme's aggregates: its type, its days, its outcome, its correction source if any. Nothing further is required of the handler — the filing is the by-product of the docket's use — but the learning loop is a leadership act: monthly, the patterns are read (which factors, which data sources, which letter paragraphs generate cases) and one fix is commissioned. A response operation that never changes upstream is a cost centre; one whose cases keep repairing the pipeline becomes, within a year, the organisation's best decision-quality instrument, as the worked quarters in the answering-every-request guide show.

That closes the sequence. Six steps, none heroic, all recorded — the distance between a request and an answer that a person can believe in. Run it inside one workflow for one month and the sequence becomes the team's habit; habits, unlike projects, do not end.

Your five actions for responding to a review request

  1. Acknowledge within your stated target with a reference, the timeframe and the status link — and file that text to the case.
  2. Assemble the complete file before assignment: request verbatim, submission, decision record, notice version, letter, corrections.
  3. Assign for competence and specified authority; keep assembly notes factual and free of predicted outcomes.
  4. Decide among the three real outcomes and write the note that answers the person's grounds in their order.
  5. Close only with both halves present — outcome recorded, communication sent from the record — then read the month's patterns and fix one thing.

Worked example: one request, six steps, nine days

A fictional tenant's review request lands on a Tuesday morning sweep. Acknowledgement out by ten o'clock with reference, ten-working-day target and status link (step one). Assembly completes the same day: the screening record shows a filing without disposition; the applicant's own text names the court and date (step two). Assigned Wednesday to the lettings manager with portfolio authority (step three). She verifies the disposition Thursday — dismissed — appends the correction, and the re-assessment clears the remaining factors; her note addresses the filing, the verification and the income arithmetic the applicant had also questioned (step four).

The outcome letter generates from the case Friday — approval, with the specific home held over the weekend by agreement — and the case closes with both halves filed (step five). The following month's pattern read counts it among four disposition-less filings, and the pre-decline verification step from the landlord guide is born (step six). Nine days end to end, most of them waiting on a court record — and the applicant, watching her status page, never once needed to phone. The sequence is not fast because anyone hurried. It is fast because nobody had to reconstruct anything.

Frequently asked questions

How quickly should a review request be acknowledged?

Within your stated target — two working days is a sound default — with a reference, the timeframe you work to and the applicant status link. The acknowledgement confirms and schedules; it must not pre-judge the merits.

What should the reviewer receive?

The assembled case: the request verbatim, the original submission, the decision record with factors, the notice version, the letter sent and any corrections — complete enough for a competent stranger to decide from the file alone.

What are the possible outcomes of a review?

Uphold with reasons, overturn, or repair the data and re-assess (with the re-assessment's outcome itself recorded). All three must be genuinely available to the reviewer, with the authority to enact them.

When is a case properly closed?

When both halves exist: the outcome recorded on the docket with the reviewer's note, and the outcome communication generated from the record and sent. One without the other is a failure the metrics should not call success.

Is this guide legal advice?

No. ReviewDesk is templates and workflow tooling, not legal advice. The sequence is ReviewDesk's handling standard; consult counsel for your obligations.

Keep reading: What happens after a human review decides — for the applicant and for the business? · What should an annual ADMT programme review actually examine? · How do you train hiring managers to work inside an ADMT notice and review process? · How do you choose ADMT compliance software? (A buyer's checklist) · How should you handle a formal complaint about an automated decision? · 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.