Negotiation role-play simulator

A negotiation role-play simulator is an AI buyer you load with your approved objections and fallbacks, then rehearse against until the response is fluent and citable. The first time a rep hears "you are 30 percent above IncumbentCo" should not be on the live call.

Hyperbound and Second Nature sit in this class: a configurable buyer persona, a scored session, and a tape the rep and a manager can replay. They are not ranked here, and they are not a live buyer. Gong still holds the real calls. Salesforce still holds the deal. The sim is rehearsal.

Do not buy this as a win-rate program. The useful outcome is quality of the talk track: can the rep use the approved fallback, with a cite the manager can check, before they sit with a real procurement lead. A high sim score is not a commit, not quota, and not a substitute for the live call.

If you need help during the call, that is a different job: live objection-response prompts or real-time call coaching. Those tools interrupt a live buyer. The sim does not. Use the sim so any live prompt is a reminder, not a script the buyer can hear being read.

The sim is only as hard as the objections you put in it. A generic "skeptical CFO" persona produces generic answers and a bot that eventually says yes.

Source objections from what buyers actually said. Pull pricing, competitive, and process pushback from Gong recordings and lost-deal notes, not from a vendor default library. Write the approved fallback next to each objection: the sentence legal and product already signed off, including what the rep must not say (discount floors, competitor claims, payment terms you do not offer). If the fallback is not written down, the scorer has nothing to cite.

Bind the bot to refuse until those conditions are met. If the persona accepts a discount below your floor, or agrees to net-60 when finance has net-30, the tape is teaching the wrong close. The failure mode is a bot that only says yes: reps finish sessions feeling ready, then stall the first time a real buyer holds the line.

Keep the library small. Three to five objections that actually kill deals in this motion beat a catalog of twenty that nobody hears. When product, pricing, or a competitor change, update the library the same week. Stale fallbacks rehearse last quarter's story.

Paper negotiation is a different queue. Once the talk track is fluent, contract redline AI is how legal triages markup. Do not ask the buyer bot to role-play clause language it was never given.

The following is a made-up but realistic trial design, not a case study and not reported results.

Meridian sells warehouse labor software to mid-market 3PLs. AEs keep losing late-stage deals when procurement says the list is above IncumbentCo and when finance asks for net-60. Enablement has a one-pager. Nobody uses it on the call.

Enablement loads two scenarios into a role-play tool in the Hyperbound and Second Nature class:

1. Price vs IncumbentCo. The bot names the competitor, quotes a round number below Meridian's list, and will not move unless the rep uses the approved value fallback (labor hours recovered, not a matching discount). Going below the discount floor is an automatic fail, with the line cited.
2. Net-60. The bot asks for 60-day terms. The approved fallback is net-30 with a documented exception path through finance. Agreeing on the call is a fail.

In the first pass, the bot agrees after two polite pushes. Every AE "passes." The sales coach listens to three tapes and hears the same pattern: the bot folded. They tighten the persona so it must hear the approved sentence, ask a follow-up, and refuse off-policy concessions.

Scoring v1 rewards talk time and "confidence." AEs who ramble outscore AEs who use the one-pager. The coach throws that rubric out. Scoring v2 is binary on the fallback, with a timestamped cite, plus a flag if the rep invents a competitor claim that is not on the battle card.

The manager does not watch every tape. They review outliers: fails on the fallback, passes with a cite the coach cannot find, and anyone who finished the IncumbentCo scenario without being pressed. Those tapes go into 1:1s. They do not go into the Salesforce opportunity as a stage gate, and they do not change commission.

Before the next live IncumbentCo bake-off, the AE still does a dry-run with the manager. The sim was practice. The manager still decides if the talk track is ready.

A useful score points at a passage. "Used approved IncumbentCo fallback at 04:12" is coachable. "Strong presence" is not.

Align the sim rubric with the same methodology scorecard you already use on live Gong tapes (MEDDPICC, SPIN, Sandler, or whatever you actually train). If practice scores confidence and live scoring scores "economic buyer identified," reps will perform for the sim and still miss the live call. Publish the rubric. Hidden criteria get gamed.

Do not score talk time. Long answers feel thorough and often bury the fallback. Do not average a dozen fuzzy dimensions into a single enablement KPI.

Cites are the audit trail. If the model says the rep hit the fallback and the timestamp plays a different sentence, the rubric or the model is wrong. Treat disagreement as a playbook bug first, a people bug second.

The manager's job is outliers, not a queue of every session. Sample fails, implausible passes, and reps who skip practice before a known hard call. If every tape needs a human grade, you bought a scoring product you cannot operate.

Salesforce is the system of record for the deal. The sim is not.

Do not add a "role-play passed" field that blocks stage movement. Do not rank the team on sim scores in the QBR. Do not put sim completion on the comp plan. The moment the number pays, reps optimize the bot: short sessions, memorized openings, and a buyer that never leaves the happy path.

A passing sim means the rep can deliver the approved fallback under a predictable buyer. Real procurement changes the order, brings a new stakeholder, or asks a question the library does not contain. Treating a passing sim as deal-ready is how you send someone into a bake-off with false confidence.

Use the tape as input to coaching, then use live evidence for the deal. Last-mile deal coaching is the late-stage job: one next action on a named opportunity (missing signature, stalled legal, missing economic buyer), with a cite from the CRM and the live motion, not from a practice bot.

Onboarding and a new SKU or competitor launch are the right times to mandate a small set of scenarios. Ongoing practice should be pulled by the AE before a known hard call, not pushed as weekly hours.

Ready for the live call is a manager judgment after practice and a look at recent live tapes, not a green badge on a dashboard.

Run this loop:

1. Load approved objections and fallbacks from real Gong losses and the current battle card. Bind the bot so it does not say yes to off-policy terms.
2. Each AE runs a scored tape. The score must cite a timestamp. Throw out talk-time and vibe grades.
3. The manager reviews outliers in 1:1s. Enablement updates the library when legal, product, or a competitor changes.
4. Keep sim results out of quota, stage gates, and the Salesforce forecast.

Kill the program if the bot is still agreeable after you tightened it, if scores do not match what managers hear on the tape, or if AEs who pass still freeze on the same objection in Gong. That is a playbook or a scoring problem. Buying another role-play seat will not fix it.

The sim exists so the expensive surprise happens in practice. It does not close the deal. The live buyer still does.