Practical guide
How to create a negotiation plan with AI
A negotiation plan should make the team's authority, evidence, alternatives, and trade boundaries usable before the conversation becomes pressured. AI can help organise research, test scenarios, and expose inconsistencies, but it should not invent leverage, set a hidden walk-away point, or commit the organisation. The finished artifact must let an authorised person see what can be proposed, what.
A negotiation plan should make the team's authority, evidence, alternatives, and trade boundaries usable before the conversation becomes pressured. AI can help organise research, test scenarios, and expose inconsistencies, but it should not invent leverage, set a hidden walk-away point, or commit the organisation. The finished artifact must let an authorised person see what can be proposed, what requires referral, and what evidence supports each move.
Ground the plan in current guidance
The World Bank's procurement guidance starts negotiation planning by setting objectives, then defining an ideal position, fallback position, and walk-away position before prioritising issues. The UK Small Business Commissioner's guidance similarly emphasises research, explicit terms, restraint on concessions, and putting the agreement in writing. Local Partnerships' supplier-negotiation guidance adds red lines, reciprocal concessions, scenario planning, the other party's likely constraints, and identification of the actual decision-maker. Together, these sources support disciplined preparation; they cannot establish a favourable outcome or replace legal and commercial advice. World Bank: Negotiation and Best and Final Offer UK Small Business Commissioner: How to negotiate terms and conditions Local Partnerships: Supplier negotiations
Sources: [1] [2] [3]
Build the plan with RANGE
Use RANGE: record the mandate, assemble the evidence, name the parameters, generate conditional packages, and execute with a decision log. Each stage produces a reviewable part of the plan rather than a persuasive narrative.
### R: Record the mandate
Write one sentence for the business decision and one for the negotiation's permitted scope. Name the sponsor, lead negotiator, technical and legal reviewers, note-taker, and final approver. Record who may agree each term and when the team must pause for referral.
Separate outcomes from positions. “Keep annual cost within the approved envelope while preserving service continuity and exit rights” is an outcome. “Reject any increase” is a position. List the interests behind each outcome, the affected users, the decision deadline, and the cost of delay.
Ask AI to turn approved notes into an authority matrix or find contradictions between the stated objective and the proposed terms. Do not give it confidential material unless the tool and use are approved. Treat every inferred objective or authority as unverified until an owner confirms it.
### A: Assemble the evidence
Create an evidence register for current contract terms, usage, performance, incidents, service dependencies, alternative suppliers, transition effort, market comparisons, budget, and relevant policy or legal advice. Give every item an owner, date, source link, confidentiality label, and limitation.
Distinguish observed facts from estimates and assumptions. A retained invoice is not evidence of future market price. A vendor statement about uptime is not the same as independently retained service data. An AI summary is a navigation aid, not a substitute for the source.
Use AI to compare clauses, group issues, or generate questions against the retained evidence. Require citations back to the exact clause or record. Reject claims that cannot be traced or that compress materially different terms into one label.
### N: Name the parameters
For each issue, record five positions: desired, acceptable, fallback, walk-away, and referral-required. Add the rationale and approval source. Do this separately for price, term, service levels, liability, data use, security, intellectual property, audit rights, transition support, and exit.
Then state the best available alternative if no agreement is reached. Describe what the alternative actually requires, who approved it, how long it takes, and which risks remain. A hypothetical supplier name is not a usable alternative without availability, cost, and transition evidence.
Keep red lines out of prompts or systems that are not approved for sensitive commercial information. AI may help test whether parameters conflict—for example, an aggressive price ceiling paired with a non-negotiable service expansion—but authorised people set and approve the parameters.
### G: Generate conditional packages
Build packages rather than a list of isolated concessions. Every proposed movement should be conditional: “If the term extends to 24 months, then the annual increase is capped and the exit-assistance schedule is attached.” Record the value, cost, operational effect, dependencies, and approval level of each exchange.
Prepare three evidence-backed scenarios: agreement near the desired position, agreement near the fallback, and no agreement. For each, record next actions, communications, service-continuity steps, and the point at which the alternative is activated.
AI can generate counterfactual questions: What if the supplier accepts price but rejects audit rights? What if implementation dates move? What if a proposed concession weakens a red line indirectly? Review these questions for realism. Do not let generated scenarios create false urgency or unsupported claims about the other party's motives.
### E: Execute and retain the decision trail
Use a live issue log with the latest proposal, evidence cited, open question, owner, authority status, and next move. After each session, reconcile notes with the authorised team before sending a written record. Mark proposals as proposed, agreed in principle, approved, or executed; these states are not interchangeable.
Before signature, verify that the final document matches the decision log, every conditional exchange is complete, approvals are retained, and no AI-generated language introduced an unreviewed obligation. Record unresolved assumptions and the operational handoff.
Negotiation-plan artifact
| Issue | Evidence | Desired | Fallback | Walk-away or referral | Conditional exchange | Owner | | --- | --- | --- | --- | --- | --- | --- | | annual price | contract, invoices, usage, market evidence | | | | | | | service level | incident and performance record | | | | | | | data and security | approved risk review | | | | | | | term and exit | transition plan and dependency map | | | | | |
Add an evidence register, authority matrix, package sheet, scenario sheet, question list, and dated decision log. A reviewer should be able to trace every material parameter to an owner and source.
Worked example: synthetic analytics-service renewal
A fictional operations team is renewing an analytics service used by 140 staff. The supplier proposes a 15% annual increase, a three-year term, and weaker exit support. The current service record shows two priority incidents, 91% active-seat usage, and no completed export rehearsal. A credible alternative exists, but migration would take four months and require a verified data export.
RANGE records the procurement director as final approver, the service owner as lead, security and legal as reviewers, and a rule that liability, data-use, or exit changes require referral. The objective is continuity within the approved two-year cost envelope while retaining export and transition rights.
The evidence register marks the alternative supplier quote as 45 days old and conditional on a technical discovery. It also exposes that the internal “99.9% uptime” claim came from the supplier dashboard rather than the team's incident record. The plan therefore treats both as qualified evidence.
The price parameter is desired at no increase, acceptable at 5%, fallback at 8% if unused seats can be removed, and referral above 8%. A three-year commitment is outside the lead negotiator's authority. Data export within 10 business days and 60 days of exit assistance are red lines until legal and security approve a change.
The team prepares a package: if the supplier caps the increase at 6%, permits annual seat adjustment, retains the export clause, and provides a staffed export rehearsal, the customer will accept a two-year term and quarterly service reviews. A second package trades a lower first-year price for a documented implementation milestone rather than surrendering exit rights.
During the meeting, the supplier offers 4% but removes the export rehearsal. The issue log prevents the attractive price from masking the red-line change. The team pauses, refers the package, and later agrees 5%, annual seat adjustment, the existing export clause, one rehearsal, and a two-year term. The signed draft is reconciled against the log before approval.
This example demonstrates a controlled artifact, not an optimal-deal claim. Different facts, authority, markets, and legal duties would change the plan.
Two Prune boundary
Two Prune can assess whether a participant frames a negotiation problem, navigates supplied evidence, uses AI with traceability, checks tradeoffs, and communicates a bounded plan. It reports observed assessment performance and evidence strength. It does not negotiate, provide legal advice, approve commercial terms, predict workplace outcomes, or make employment decisions.
Sources
World Bank: Negotiation and Best and Final Offer
- UK Small Business Commissioner: How to negotiate terms and conditions
- Local Partnerships: Supplier negotiations
Sources: [1] [2] [3]
Sources
- 1.World Bank: Negotiation and Best and Final Offer · pubdocs.worldbank.org
- 2.UK Small Business Commissioner: How to negotiate terms and conditions · www.smallbusinesscommissioner.gov.uk
- 3.Local Partnerships: Supplier negotiations · localpartnerships.gov.uk