Quick answer
By Vozah Editorial·Last updated
AI Sales Training RFP Checklist: 25 Questions to Ask Vendors Before You Sign
An AI sales training RFP should test four things: whether the AI scoring holds up on your own calls (technical), whether you can roll it out (operational), what it really costs over the contract (commercial), and what happens to your data (risk). This page gives you 25 questions, a 1-5 scoring scale, proposed weights and a worked example you can copy into a spreadsheet.
It is not a vendor ranking (use the best AI sales training comparison or the AI sales roleplay tools comparison for that). Disclosure: Vozah publishes this checklist and sells an AI sales roleplay product, so we have marked clearly what Vozah does and does not offer in its own section.
Quick answer: Shortlist about three vendors, send the same 25 questions, and score every answer 1-5 against written anchors. Weight technical 35%, operational 25%, commercial 25% and risk 15% (proposed weights; adjust before responses arrive). The test that separates vendors fastest is asking each one to score a call or practice session you supply, live, and then comparing its scores with your own managers' scores.
How do you use this RFP template?
Fill in the template in this order: set weights, send questions, score against the anchors, then run the live tests. Four rules (our suggested operating model, not an industry standard):
- Fix the weights before responses arrive. Weights chosen after a favorite demo tend to drift toward that favorite.
- Shortlist about three vendors. Enough for comparison and leverage, few enough to run live tests and reference calls properly.
- Score every answer 1-5 against the anchors below. A number forces a decision where prose hides ambiguity. Two people should score independently, then reconcile.
- Run technical questions first. They eliminate weak fits fastest; there is no point negotiating price with a vendor whose scoring you do not trust.
Scoring anchors: what does a 1 through 5 mean?
Use the same anchors for every question so scores are comparable across vendors and reviewers.
| Score | Anchor | What it looks like in practice |
|---|---|---|
| 5 | Demonstrated | Shown live on your material, or provided in writing with documentation you can verify (contract clause, report, API docs) |
| 4 | Documented | Clear written answer with documentation, but not shown on your material |
| 3 | Credible claim | Specific answer from the vendor, no documentation yet; verify before signing |
| 2 | Vague | Generic answer ("we score holistically", "it depends") with no specifics after a follow-up |
| 1 | Missing or red flag | No answer, refusal, or an answer that matches the red flag listed for that question |
Score "not applicable" questions as blank and drop them from the denominator rather than scoring them 3.
Category 1: Technical questions (8)
Technical questions show whether the AI is reliable on your conversations, not just on curated demos.
| # | Question | A strong answer includes | Red flag |
|---|---|---|---|
| 1 | Will you score a call or practice session we supply, live? | Scores and written feedback produced in the meeting on material you chose | Only scores its own demo calls, or needs weeks of "calibration" first |
| 2 | What does your AI score, and can you show the rubric? | The named skills or behaviors it evaluates, with an example of the feedback a rep sees | "We score the whole call holistically" with no breakdown |
| 3 | How does the AI buyer handle interruptions, objections and off-script answers? | A live roleplay where your evaluator interrupts, objects and goes off-script | Voice cuts the rep off, loses the persona or ignores what the rep said |
| 4 | Which model families and providers power transcription, the buyer persona and scoring? | Named model families or providers, and how they are versioned | Cannot or will not say what the system is built on |
| 5 | How do you keep scores consistent when models are upgraded? | Versioned rubrics, regression testing on a fixed set of calls, notice before scoring changes | "We always use the latest model" with no versioning or notice |
| 6 | Which integrations are native, which are partner-built, and which need services work? | A written list split three ways (native / partner / services), plus SSO support if you need it | Logo wall in marketing, but few native integrations on the sales call |
| 7 | Is there a documented API, and what are the rate limits? | Public API documentation URL and stated limits, or a clear "no API" | API "on request" with no documentation |
| 8 | How do you measure scoring and transcription accuracy? | Their method, the test set it was measured on, and willingness to compare AI scores with your managers' scores on the same calls | "100% accurate", or no metric and no method |
How to test accuracy yourself (question 8). We do not quote a benchmark accuracy range here because there is no public, independent benchmark for AI sales-call scoring, and vendor-reported transcription accuracy (often expressed as word error rate) depends heavily on the audio, accents and vocabulary it was measured on. Instead, pick 5-10 recorded calls or practice sessions, have two managers score them independently with your own rubric, have each vendor score the same set, and compare where the AI and your managers agree and disagree. If a vendor quotes a word error rate or agreement figure, ask for the test set, the date and the population it was measured on. The NIST AI Risk Management Framework (AI RMF 1.0, January 2023, voluntary) is a useful neutral reference for the kind of measurement and documentation questions to ask.
Category 2: Operational questions (6)
Operational questions show whether rollout will actually happen and stick.
| # | Question | A strong answer includes | Red flag |
|---|---|---|---|
| 9 | What is a realistic implementation timeline for a team our size, and what does it depend on? | A dated plan with named dependencies (content build, SSO, integrations, manager training) | "It depends" with no specifics after a follow-up |
| 10 | What does the first 90 days look like for our admin and managers? | Week-by-week milestones and who owns each | "We will figure it out together" |
| 11 | Who builds practice scenarios and buyer personas: us or you? | A built-in library plus a way to create custom scenarios, with which plan tier includes it | Every custom scenario requires a paid services engagement you did not budget for |
| 12 | What practice cadence do your customers actually sustain? | A recommended cadence and how they report usage to managers | No usage reporting, or "as much as they want" |
| 13 | What do managers get, not just reps? | Manager dashboard demo, team-level views and a manager onboarding plan | Rep-only product with no manager workflow |
| 14 | What support do we get at our spend level? | Named contacts or support channel, response times in writing | Support model undefined until after signature |
Category 3: Commercial questions (6)
Commercial questions reveal total cost over the full contract, not just year one.
| # | Question | A strong answer includes | Red flag |
|---|---|---|---|
| 15 | What is the price per seat per month in each year of the contract? | Year-by-year line items, including usage limits (for example practice minutes) | Only a blended annual figure, or pricing only "after we talk to our team" |
| 16 | Can we evaluate the product before signing? | Free plan, trial or self-serve monthly plan you can start without a contract | No hands-on access before an MSA |
| 17 | What is included in the base seat versus add-ons? | A written inclusion list (integrations, analytics, custom scenarios, uploads) | Key features you assumed were included turn out to be add-ons |
| 18 | Will you cap renewal price increases in writing? | A specific cap in the order form | No cap, or "standard increases apply" |
| 19 | What are the terms and discounts for annual or multi-year prepayment? | Written discount schedule and payment terms | Annual commitment required with no difference from monthly |
| 20 | What happens if we reduce seats mid-contract? | A flex band or step-down right in writing | Locked seat count for the full term |
We do not quote "typical" renewal caps or prepay discounts because we have no public, primary data for this category; get the numbers in writing from each vendor and compare them side by side.
Category 4: Risk questions (5)
Risk questions surface what is hard to fix after signature.
| # | Question | A strong answer includes | Red flag |
|---|---|---|---|
| 21 | Where are call audio, transcripts and scores stored, and for how long? | Named hosting regions, retention periods you can configure, a deletion process | Vague storage location, no deletion process |
| 22 | Do you or your model providers train on our data? | A clear written answer, and a contractual opt-out if training is the default | No clear answer |
| 23 | Which security attestations do you hold, and can we see the latest report? | Current report (for example an AICPA SOC 2 examination) available under NDA, with its period covered | Claims of compliance with no report available |
| 24 | Can we terminate for convenience, and with what notice? | Exit right with defined notice and data export | No exit option short of the full term |
| 25 | What happens to recordings, scores and dashboards if we leave? | Defined export window and format at no extra cost | Immediate deletion, or export only for a fee |
On question 23, SOC 2 is an attestation framework run by the AICPA covering controls relevant to security, availability, processing integrity, confidentiality or privacy (AICPA & CIMA, SOC suite of services). Ask whether a report is Type 1 (design at a point in time) or Type 2 (operation over a period), and which period it covers.
Proposed weights: how should you weight responses?
These weights are a proposed starting point, not an industry standard or a researched benchmark. Adjust them to your situation before responses arrive.
| Category | Questions | Proposed weight | Rationale |
|---|---|---|---|
| Technical | 1-8 | 35% | If you do not trust the scoring, nothing else matters |
| Operational | 9-14 | 25% | Unused software has no return |
| Commercial | 15-20 | 25% | Multi-year cost and flexibility vary by vendor |
| Risk | 21-25 | 15% | Lower probability, but hard to fix later; treat any 1 on questions 21-23 as a potential veto |
Optional decision-critical override. If one question decides the purchase for you (for many teams that is question 1, the live scoring test), give it up to 30% of the total and scale the rest of its category down. Decide this before the demos.
Formula. For each category: category score = average of its question scores (1-5). Weighted total = sum of (category score x category weight). The maximum is 5.0.
Worked example: scoring two hypothetical vendors
Hypothetical example: a 40-rep team compares two anonymous vendors using the proposed weights. The scores are invented to show the math, not drawn from any real vendor.
| Category | Weight | Vendor A avg | A weighted | Vendor B avg | B weighted |
|---|---|---|---|---|---|
| Technical (Q1-8) | 35% | 4.4 | 1.54 | 3.1 | 1.09 |
| Operational (Q9-14) | 25% | 3.8 | 0.95 | 4.2 | 1.05 |
| Commercial (Q15-20) | 25% | 3.2 | 0.80 | 4.5 | 1.13 |
| Risk (Q21-25) | 15% | 4.0 | 0.60 | 3.6 | 0.54 |
| Weighted total (max 5.0) | 100% | 3.89 | 3.80 |
What the numbers show:
- Vendor A scored 5 on question 1 (it scored three of the team's own practice sessions live, and the managers agreed with most of its feedback) but only 2 on question 18 (no renewal cap offered).
- Vendor B was cheaper and had a cleaner contract but scored 2 on question 1 (it would only score its own demo calls) and 2 on question 5 (no scoring versioning).
- The totals are close, so the tie-breaker is the decision-critical question. If the team had given question 1 a 30% override, Vendor A's lead would widen. The next step is to negotiate a renewal cap with Vendor A, not to switch to Vendor B.
Row values are rounded to two decimals; totals are calculated from unrounded values. Copy the layout above into a spreadsheet with one row per question, one column per vendor and reviewer, and a reconciled score column.
How Vozah answers this checklist
Vozah publishes this page and is a vendor in this category, so here is what Vozah actually offers today and what it does not. Plan details are on the pricing page.
| Checklist area | What Vozah offers | Not included or not offered |
|---|---|---|
| Hands-on evaluation (Q16) | Free account with 10 practice minutes per calendar month, 1 seat, built-in scenarios, no card; self-serve monthly plans | |
| Practice and scoring (Q1-3) | AI voice roleplay against AI buyer personas in built-in scenarios, with scoring and feedback after each session; real call uploads on paid plans (Solo includes 5 uploads) | Vozah does not record live customer calls |
| Scenarios and personas (Q11) | Built-in scenarios on all plans; custom AI buyer personas and an industry-specific practice library on Growth; custom scenarios and shareable scorecards on Business | |
| Manager workflow (Q13) | Team plan: team leaderboard, practice streaks and badges, manager dashboard | No human coaching, certification or LMS |
| Published pricing (Q15) | Solo $29/mo (1 seat), Team $149/mo (5 seats), Growth $399/mo (20 seats), Business $899/mo (50 seats), with per-seat monthly practice minutes on each plan | |
| Integrations and identity (Q6) | SSO, CRM integrations, white-labeling and custom SLAs are not included in self-serve plans; Vozah does not claim CRM or dialer integrations | |
| Security, data and contract terms (Q21-25) | Ask Vozah directly and get answers in writing | This page makes no claims about them |
If SSO or a CRM integration is a hard requirement (questions 6 and 23), score Vozah's self-serve plans accordingly rather than assuming they are included.
Which red-flag patterns cut across questions?
A few patterns no single question catches:
- The "trust us" vendor. Will not score your material, will not put pricing in writing, will not offer a reference who downgraded.
- The pilot trap. A discounted pilot that converts automatically to a multi-year contract at full price. Read every renewal and conversion clause.
- The integration mirage. Many logos in marketing, few native integrations. Question 6 forces the native / partner / services split in writing.
- The moving rubric. Scores that change when the vendor upgrades models, with no notice. Question 5 covers it; ask again at renewal.
What should you ask vendor references?
Ask each reference these five questions verbatim:
- What did you wish you knew before signing?
- What feature in the demo did not match how it works in production?
- How long did implementation actually take versus what was promised?
- What share of your managers actively use it, and how do you know?
- If you were starting over, what would you do differently?
Specific, slightly uncomfortable answers are the most useful. Uniformly positive answers tell you less.
After the RFP: a suggested 30-day decision window
A suggested schedule once responses are in (an operating suggestion, adjust to your procurement process):
- Week 1: Two reviewers score responses independently against the anchors, then reconcile.
- Week 2: Run the live scoring test (question 1) with the same calls or practice sessions on every shortlisted vendor.
- Week 3: Reference calls (two or three per vendor) and procurement review of commercials.
- Week 4: Negotiate the gaps your scoring exposed, then sign.
Long, open-ended evaluations tend to exhaust the internal champion and let the status quo win by default, so set the end date before you start.
Sources
- NIST, AI Risk Management Framework (AI RMF 1.0, January 2023): voluntary framework for measuring and documenting AI system risks; useful for question 8.
- AICPA & CIMA, System and Organization Controls (SOC) suite of services: what SOC 2 reports cover; useful for question 23.
- Vozah pricing: plan inclusions referenced in the Vozah section.
Related reading: AI sales training pricing, AI sales training ROI data, how to measure training effectiveness, AI sales training guide, Vozah pricing.