Prompt Pilotby Write.law
All prompts

Draft

Draft a RIF rationale memo that survives discovery

Builds the privileged memo defending your reduction-in-force selections: business case in numbers, four-fifths and standard-deviation math run on your own data, and an independent rationale for every sensitive selection.

About 25 minadvancedEmployment, In-house, Litigation

Your prompt5,330 characters

Still to fill in: RIF summary, Selection criteria and weights, Demographics (unit and selected), States and applicable notice law

RoleYou are senior employment counsel who has defended RIF selections in EEOC charges and in federal court. You write on the assumption that a plaintiff's lawyer will read every sentence aloud to a jury: you run the numbers before you write the justification, you never let an HR euphemism into a document you will have to produce, and when the data is bad you say so on page one.What I needDraft the RIF rationale memo for the reduction below. Address it to the General Counsel, mark it Attorney-Client Privileged / Attorney Work Product, and write it as if it will be produced anyway. Apply the law of States and applicable notice law.InputsRIF summary: RIF summary Selection criteria: Selection criteria and weights Demographics - unit and selected: Demographics (unit and selected) States and notice law: States and applicable notice law Flagged individuals: Flagged individualsHow to work this1. State the business justification in numbers: headcount, cost-out target, the deadline driving it, the org afterward. If RIF summary gives you no number, say the memo cannot be written without one. 2. Define the decisional unit and say what it excludes and why. The boundary is the first thing a plaintiff attacks; defend it here, not in a deposition. 3. Restate each criterion in Selection criteria and weights with its weight and a definition a stranger could apply. Say whether the criteria were fixed before names were scored; if the input does not establish that, flag it as the memo's biggest problem. 4. Run four-fifths for every class in Demographics (unit and selected) both ways (on selection-for-termination rates and on retention rates), showing each division with the counts, not the rounded percentages. The direction of the ratio is a choice in a reduction and can flip the headline, so report both and never hand over only the version that helps. Flag every ratio below 0.80 and name which convention produced it. 5. For age, add the standard-deviation calculation: expected selections under neutral selection, actual, SD, resulting Z. Above two, call it a statistically significant disparity, not "a skew." 6. List the alternatives considered and the number that killed each: modeled buyout cost, headcount a freeze would have saved, redeployment openings that existed. 7. For each person in Flagged individuals, give the independent rationale and name the dated document proving it predates the protected activity. If no such document exists, say so. That absence is the finding.Close with these four sections, every time, without being askedAssumptions I made. Which state's law governs each selection, that the criteria predate scoring, that the demographic counts are complete, that disability figures are self-reported. Mark each [verify] or [safe]. Where this is weakest. The two or three sentences a plaintiff's lawyer would quote back. Name the sentence and the section, not "the memo generally." What only you can decide. Options with tradeoffs, not flags. At minimum: proceed on the announced schedule and defend the disparity, or re-open selections to reduce it. Re-opening lowers exposure but creates a second set of documents showing the company knew. And whether to offer enhanced severance to the 40+ group, which buys releases but reads in discovery as consciousness of the problem. What would make this materially better. Ranked: the dated org-design document, the calibration notes, the performance ratings behind criterion two, or the redeployment log.Output formatA privileged memo: header and privilege legend, then numbered sections for business justification, decisional unit, criteria, process and dates, alternatives, adverse-impact analysis with tables and shown arithmetic, per-selection notation, flagged individuals, OWBPA decisional-unit disclosure, and next steps with owners and dates.Never do this- If the justification would fit any company's restructuring, it is too generic and worthless in discovery. Every paragraph carries a number from these inputs. - No HR euphemism and no hedging filler. Never write "rightsizing," "streamlining," or "arguably." Do not tell me to consult employment counsel. I am the counsel writing this. - Never invent a regulation cite, a case, or a statistical threshold. Authority comes from my inputs or carries [UNVERIFIED - confirm before circulating]. - Where the demographic data is incomplete, say what is missing and stop. Do not estimate a rate you do not know or smooth over the gap with confident prose. - Do not pad. If the impact ratios are clean, that section is three sentences. Every extra line is one more thing to be cross-examined on.Before you answer- Did I show the arithmetic for every impact ratio and the standard-deviation figure, run each ratio on both selection and retention rates, and name the convention behind each number? - Could a stranger re-rank this unit using only my definitions and reach a similar list? - Does every flagged individual have a rationale tied to a dated document, or did I assert one without a source? - Is there a sentence here a plaintiff's lawyer would enjoy reading to a jury? Rewrite it.

Adds driver's-seat tunes: options instead of answers, questions before work, every citation flagged. Your values come with it.

2

Pressure-test it

Makes the AI switch hats and attack its own answer.

Assigned to the charge filed by the 61-year-old employee, an EEOC investigator now has this memo in the production. Read it as an investigator reads it: for the sentence that does not match the rest. Name the three sentences you would highlight as suspicious, the one document you would request first, and the question you would put to the decision-maker in their interview. Then rewrite the two weakest passages so they survive that reading without changing a single underlying fact.
3

Go deeper

Pushes the work further once the basics are right.

Everything said out loud about this RIF gets produced next to the memo. Draft the communications package that has to match it: the all-hands message to remaining employees, the manager talking points for the notification conversations, a short FAQ for inbound media questions, and the script for the affected-employee one-on-one. Every reason given must already appear in the memo: no new justifications, no softer framing.

Before you run it

What to gather first

  • Headcount before and after, by function and location
  • Cost-out target and the deadline driving it
  • The decisional unit: the org boundary used to make selections
  • Counts by protected class inside the decisional unit and inside the selected group
  • Anyone selected who is on leave, recently complained, or is over 50

Watch for

  • Run the adverse-impact analysis under counsel's direction and label it accordingly. An HR-run analysis sitting in a shared drive is usually discoverable, and it is the first thing plaintiff's counsel asks for.
  • The OWBPA § 626(f)(1)(H) disclosure is mandatory for group separations of employees 40 and older, and a defect voids the ADEA release for everyone who signed.
  • The model will do the four-fifths arithmetic quickly and sometimes wrongly. Recompute every ratio and the standard-deviation figure before the memo goes to the GC.
  • Selecting someone who recently took protected leave or filed a complaint requires contemporaneous documentation that the rationale predates the activity. If that document does not exist, the memo cannot manufacture it.
  • Federal WARN and state analogues run on their own clocks and headcount triggers. Confirm the notice obligation independently. The memo's timeline assumes it has been handled.

What comes back

A privileged memo with a header and privilege legend, then numbered sections: business justification, decisional unit, selection criteria, process and dates, alternatives, adverse-impact analysis with tables and shown arithmetic, per-selection notation, flagged-individual analysis, OWBPA decisional-unit disclosure where applicable, and next steps with owners and dates.

See an example of what you’ll get
MEMORANDUM - ATTORNEY-CLIENT PRIVILEGED / ATTORNEY WORK PRODUCT To: J. Reyes, General Counsel, Westwind Manufacturing From: Outside Employment Counsel Date: April 27, 2026 Re: NA Engineering Reduction in Force: Selection Rationale and Adverse-Impact Analysis I. Business justification. Following the December 2025 Coyote Controls acquisition, NA Engineering carries 124 FTE supporting two product lines being consolidated onto a single platform by Q3 2026. Engineering leadership identified $4.2M of annual run-rate cost reduction required to fund the platform rebuild and meet the board-approved 2026 EBITDA target. The reduction is 18 FTE (14.5%), effective June 15, 2026, targeting roles subsumed by the consolidation and skill sets mismatched to the unified roadmap. II. Decisional unit. NA Engineering, 124 FTE. Sales, G&A, and International Engineering are excluded because the platform consolidation does not change their scope; International Engineering was separately reviewed and is growing. The boundary is documented in the April 1 org-design memo. III. Selection criteria. Fixed April 8, before any employee was scored: role redundancy post-consolidation (40%), defined as a position whose responsibilities are absorbed by a named retained position on the unified org chart; last two performance ratings (30%), 2024 and 2025 cycles on the 1–5 scale; critical-skill match (20%), scored against a fourteen-technology inventory by the skip-level manager; tenure (10%), tiebreaker only. VI. Adverse-impact analysis. One note before the numbers, because it decides the headline. The four-fifths ratio is direction-sensitive, and in a reduction the direction is a choice: run it on selection-for-termination rates and you get one set of figures; run it on retention rates (the favorable outcome) and you get another. On this data the age result moves from 0.29 to 0.80 depending on which you use. I have run both. Neither version leaves this reduction alone. | Class | In unit | Selected | Selection rate | Retained | Retention rate | |---|---|---|---|---|---| | Male | 71 | 9 | 9 ÷ 71 = 12.7% | 62 | 62 ÷ 71 = 87.3% | | Female | 53 | 9 | 9 ÷ 53 = 17.0% | 44 | 44 ÷ 53 = 83.0% | | Under 40 | 78 | 6 | 6 ÷ 78 = 7.7% | 72 | 72 ÷ 78 = 92.3% | | Age 40+ | 46 | 12 | 12 ÷ 46 = 26.1% | 34 | 34 ÷ 46 = 73.9% | | No disclosed disability | 110 | 14 | 14 ÷ 110 = 12.7% | 96 | 96 ÷ 110 = 87.3% | | Disclosed disability | 14 | 4 | 4 ÷ 14 = 28.6% | 10 | 10 ÷ 14 = 71.4% | *Four-fifths on selection-for-termination rates*, taking the least-selected group as the reference. Sex: (9 ÷ 71) ÷ (9 ÷ 53) = 0.75. Age: (6 ÷ 78) ÷ (12 ÷ 46) = 0.29. Disability: (14 ÷ 110) ÷ (4 ÷ 14) = 0.45. All three fall below 0.80. *Four-fifths on retention rates*, lower-retained group over higher. Sex: (44 ÷ 53) ÷ (62 ÷ 71) = 0.95. Age: (34 ÷ 46) ÷ (72 ÷ 78) = 0.80. Disability: (10 ÷ 14) ÷ (96 ÷ 110) = 0.82. None falls below 0.80, and age lands exactly on the line. Age is the exposure on either convention: the worst ratio under one, sitting on the threshold under the other. Assume plaintiff's counsel runs the version that produces 0.29, and be ready to explain why you also ran the other. *Age standard deviations.* Expected 40+ selections if selection were neutral: 18 × (46 ÷ 124) = 6.68. Actual: 12. SD (binomial) = √(18 × 0.371 × 0.629) = 2.05. Z = (12 − 6.68) ÷ 2.05 = 2.60. Courts commonly treat two standard deviations as the line for statistical significance [UNVERIFIED - confirm how your circuit frames that threshold before the memo repeats it]. On that convention this is a statistically significant age disparity and should be described that way internally, not as "a skew." VIII. Flagged individuals. - *Employee A (FLSA complaint February 8).* I cannot complete this one. Your input gives me the complaint date and nothing else: no criteria scores, no scoring manager, and no dated document showing the redundancy determination predates February 8. That absence is the finding. Treat this as the most exposed selection on the list and issue no notice on it until the document exists or the selection is reconsidered. - *Employee C (age 61, 22 years).* Role redundancy 5/5: the position is absorbed by the retained Platform Architect role held by Nakamura. Performance 2.0 average across the 2024 and 2025 cycles. Skill match 2.5/5. Tenure operated in her favor as a tiebreaker and did not overcome the other three scores. The redundancy determination first appears in the December 18, 2025 org-design draft, three months before any RIF list existed; that draft should be preserved and indexed now. - *Employee B (returned from FMLA April 1).* Skill match 1.5/5: proficient in two of the fourteen retained technologies, per the March 2025 inventory, which predates the leave request by eleven months. Director Patel, who scored him, had no role in approving the leave. X. Recommended next steps. (1) Second-look review of the 40+ selections before any notice issues, given Z = 2.60. (2) Confirm 45-day consideration, 7-day revocation, and the decisional-unit disclosure for every 40+ separation agreement. (3) Verify Cal-WARN applicability for the California site count. (4) Preserve the December 18 org-design draft and the April 18 calibration notes. --- Assumptions I made. The demographic counts are complete for both the unit and the selected group [verify - they reconcile arithmetically to 124 and 18, which is not the same as being right]. Disability figures are self-reported and therefore understate the class [safe - treat 14 as a floor, not a count]. California law governs the California selections and Texas law the Texas ones, with no selection straddling both [verify - this memo does not know where each of the 18 sits]. The criteria were fixed April 8 before any employee was scored [verify - the memo says so because your input says so; the April 8 document is the proof and I have not seen it]. Race and national origin are absent from the data entirely, so no ratio was run for them [verify - that is a gap, not a clean result]. Where this is weakest. Two sentences. Section III's "Fixed April 8, before any employee was scored" is the load-bearing sentence in the memo and it rests on your say-so; if any scoring spreadsheet carries a modified date before April 8, this document becomes the exhibit. And Section VIII's rationale for Employee B (a March 2025 skill inventory that "predates the leave request by eleven months") is true of the inventory but not of the score. Patel applied it during selection in April 2026, after the leave. State when he scored, or opposing counsel will. What only you can decide. Proceed on the June 15 schedule and defend the age numbers, or re-open the 40+ selections before notices issue. Proceeding keeps the cost-out on the board's timetable and leaves you defending a Z of 2.60 with an individualized rationale for each selection. Re-opening lowers the disparity and creates a second set of documents showing the company looked at demographics and adjusted, which is its own exhibit. Separately: enhanced severance for the 40+ group buys releases and shrinks the plaintiff pool, and in discovery it reads as consciousness of the disparity; uniform severance across the unit is cleaner in the record and gets you fewer releases exactly where you need them most. And whether the impact numbers go to the business before the decision is final is the GC's call, not counsel's. What would make this materially better. Ranked by impact. (1) The April 1 org-design memo and the April 8 criteria document with their file metadata. That is what proves the criteria predate the names. (2) Race and national origin counts for the unit and the selected group; two protected classes currently have no analysis at all. (3) The April 18 calibration notes, which show whether the subjective skill-match scores were applied consistently or reflect one manager's view. (4) The performance ratings behind criterion two, so a stranger could re-rank the unit from the definitions alone. (5) The redeployment log, which is the only thing that turns "alternatives considered" from an assertion into a record.
Why this prompt is built the way it is
## Framework 1. **Business justification in numbers.** Not "restructuring." Why this reduction, why now, the cost-out target, and what the org looks like afterward. The legitimate non-discriminatory reason starts here or it does not exist. 2. **Criteria written before names.** Objective where possible: role redundancy, performance rating, skill match, tenure. Subjective criteria are permitted but must be defined in writing and applied consistently. 3. **Adverse-impact analysis is the core.** Four-fifths rule by race, sex, age 40+, disability, and any other class visible in the data. Show the division. Flag every ratio under 80%. 4. **Statistical significance for age.** Courts look at standard deviations in ADEA pattern cases. Compute expected selections, actual, and the resulting Z. Above two, call it what it is. 5. **Alternatives considered, with the number that killed each one.** Voluntary buyouts, hiring freeze, reduced hours, redeployment. 6. **Selection-by-selection notation.** Criteria scores, decision-maker, date, reviewer, individualized factors. 7. **OWBPA disclosure for group programs.** Job titles and ages of everyone in the decisional unit, selected and not selected, with 45-day consideration and 7-day revocation. 8. **Heightened scrutiny for flagged individuals.** Independent rationale plus a dated document proving the rationale predates the protected activity. 9. **Privileged in form, producible in substance.** Address it to the GC and mark it, but write every line as if it will be produced anyway.