Prepare to Mediate Conflicting Requirements — Steel-Man Both Sides First

Workflow area: Stakeholder Communication · Tier: Foundation · Hands-on time: ~25 minutes

Requirements conflicts are the moment a business analyst earns the title. Handled badly, they become escalations where the louder stakeholder wins and the quieter one sabotages adoption later. Handled well, they surface the interests under the positions — the thing each party is actually protecting — and usually reveal options nobody had proposed because everyone was defending a corner. The preparation for that conversation is structured thinking work AI accelerates enormously: steel-manning (arguing each side better than its own advocate did), interest analysis, and option generation with trade-offs. What AI must never do is attend the meeting for you or pre-decide the outcome. Mediation is a human act; this playbook is about walking in prepared.

When to use — and when not to

• Use it when two legitimate requirements collidespeed versus control, standardization versus local flexibility, one team's output being another's rework.
• Use it before any alignment or escalation meeting you facilitatewalking in with steel-mans and options changes the meeting from combat to selection.
• Use it when you suspect you are biasedif you privately agree with one side, steel-manning the other is precisely the corrective; the model has no stake.
• Skip it when the "conflict" is a factual questionif data can settle it ("does the approval actually catch errors?"), run pb-07 first; mediating what should be measured wastes goodwill.
• Skip it when the conflict is really about authority or headcounta turf dispute wearing a requirements costume needs a sponsor conversation, not a facilitated workshop; naming that honestly is the analysis.

The recipe

1. 1.

   Assemble the conflict file. Each party's stated requirement verbatim (from pb-01 citations or the pb-11 decision log), the context of who they are and what they are measured on, and any relevant constraints (policy, regulation, system limits). Get the positions in the stakeholders' own words — paraphrases smuggle your bias in at the foundation.

2. 2.

   Steel-man both sides and surface interests.

   Copy

   You are helping a business analyst prepare to mediate a
   requirements conflict. Be scrupulously even-handed.
   
   Party A: [ROLE, WHAT THEY ARE MEASURED ON]
   Position A: [VERBATIM REQUIREMENT/QUOTE + CITATION]
   Party B: [ROLE, WHAT THEY ARE MEASURED ON]
   Position B: [VERBATIM REQUIREMENT/QUOTE + CITATION]
   Constraints: [POLICY / REGULATORY / TECHNICAL FACTS]
   
   Produce:
   1. STEEL-MAN A — the strongest honest case for A's position,
      argued better than A argued it. Then the same for B. Equal
      length, equal effort.
   2. INTERESTS — for each party: what they are actually protecting
      (risk exposure, workload, accountability, past incidents).
      Distinguish stated position from underlying interest. Mark
      inferred interests (INFERRED) — I will verify them.
   3. COMMON GROUND — what both parties already agree on, stated
      explicitly.
   4. THE REAL DISAGREEMENT — the narrowest possible statement of
      what is genuinely in tension once interests are separated
      from positions.
   
   Rules: no verdicts. Do not rank the parties' cases. If the
   conflict dissolves once interests are separated, say so.
   

3. 3.

   Generate resolution options with honest trade-offs.

   Copy

   Using the analysis above, draft 3-5 resolution options. For each:
   - THE OPTION — one sentence, concrete.
   - WHAT A GETS / GIVES UP — specific, not diplomatic.
   - WHAT B GETS / GIVES UP — same.
   - RESIDUAL RISK — what could still go wrong, and who carries it.
   - TEST — a cheap way to trial or validate the option before
     committing (pilot, time-box, data check).
   
   Include at least one option that reframes the problem (different
   process design, different control point) rather than splitting
   the difference. Do not include a recommended option — ordering
   is the mediator's call.
   

4. 4.

   Read a sample of good output.

   Copy

   THE REAL DISAGREEMENT: not whether a control exists, but whether
   it must be a pre-approval (blocking, 4.2-day median delay) or can
   be a post-transaction audit (non-blocking, catches errors after
   the fact). A needs cycle time; B needs demonstrable control for
   the auditors.
   
   OPTION 3 (reframe): risk-tiered approval — vendors under $25k
   with clean data auto-approve and enter a 100% post-audit sample;
   above threshold keeps pre-approval. A gets: ~70% of volume
   unblocked. B gets: audit trail arguably stronger than today.
   Residual risk: threshold gaming (split purchases) — carried by B.
   TEST: run last quarter's data through the tiers; count what
   would have auto-approved and what the post-audit would have caught.
   

5. 5.

   Verify the inferred interests before the meeting. Every (INFERRED) interest gets checked in a short one-on-one with each party — "it sounds like the audit finding from 2023 is what this protects; is that right?" Two things happen: your analysis gets corrected by reality, and each party arrives at the meeting already feeling heard, which is half of mediation.

6. 6.

   Run the meeting as the human you are. Open with common ground, present both steel-mans (watch each party relax when their case is stated well), name the real disagreement narrowly, then put options on the table — including the invitation to improve or combine them. You facilitate; the parties choose; the sponsor breaks genuine ties. Log the outcome with its rationale in the pb-11 decision log, including the options considered and set aside.

Make it yours

• Claude/ChatGPT as sparring partner: after drafting, ask "as Party A's most skeptical ally, attack this steel-man of A — what did it miss or soften?" per side; the critique pass catches subtle imbalance.
• The pre-read variant: for mature teams, turn the steel-mans and common ground (not the options) into a one-page pre-read so the meeting starts past the positional stage — options land better when generated partly in the room.
• Multi-party scale: with three or more factions, add an explicit interest-matrix table (party × interest) and ask which coalitions exist — multi-party conflicts are usually two real disagreements braided together, and separating them makes each tractable.
• Escalation prep: when mediation legitimately fails, the same file becomes the escalation brief — both cases fairly stated, options with trade-offs, and the specific question the sponsor must decide. That is the difference between escalating a decision and escalating a fight.

Quality check

• Both steel-mans pass the advocate testeach party would say "yes, that is my case, stated well or better"; if one reads stronger, rebalance before anyone sees it.
• Positions came in verbatimthe inputs were quotes with citations, not your summary of what each side "basically wants".
• Inferred interests were verified with the humansevery (INFERRED) tag was confirmed, corrected, or dropped in the one-on-ones; unverified psychology is not analysis.
• Trade-offs are honestevery option names what each party gives up; an option with no losers is usually an option with hidden losers.
• No AI verdict leakedthe pack contains no recommendation ordering; the moment stakeholders sense the fix was in, the mediation is dead.
• The outcome landed in the decision logwith rationale and roads-not-taken, so this conflict is not re-fought from scratch in six months.