A QA software RFP is the request for proposal you send to quality assurance vendors so their answers can be compared on the same questions. For a contact center it should cover scoring coverage and accuracy, scorecards, channels and integrations, coaching, AI agents, security, implementation and price, and come with a weighted scoring sheet agreed before the first answer arrives.
In short
- Write your requirements before the RFP: how much you want scored, on which channels, by whom, and which answers rule a vendor out.
- Use the 40 questions below, grouped into nine sections, and ask for evidence with every answer rather than a yes.
- Score each answer 0 to 3 against written definitions, weight the sections, and treat security and data terms as a pass or fail gate.
- The worked example shows why a close total needs a tie-break rule and a scripted demo, not a longer spreadsheet.
- Avoid the mistakes that make answers impossible to compare: open questions, no weights, and price scored as a feature.
This article is the buying checklist. If you want to see how named tools compare, read our comparison of customer service QA software. If you are shortlisting QA software for a large support team, the questions below are the ones we would want a buyer to ask us.
What should a contact center QA software RFP include?
A contact center QA software RFP should include your context, your requirements, the vendor questions, the scoring method and the timeline. Vendors answer better when they know your volume, channels and helpdesk, and you compare better when the scoring is fixed in advance.
| RFP part | What goes in it | Why it matters |
|---|---|---|
| Context | Seats, monthly conversation volume, channels, languages, helpdesk, current QA process | Vendors size the answer to your operation instead of sending a generic deck |
| Requirements | Must-haves, wanted extras, knockouts | Separates what you need from what sounds good in a demo |
| Questions | The 40 questions below, with “show us” evidence requests | Makes answers comparable line by line |
| Scoring | Scale, section weights, knockout rules, who scores | Stops the loudest stakeholder deciding after the fact |
| Commercials | Pricing table to fill in, contract term, renewal terms | Lets finance compare like with like |
| Timeline | Deadline for questions, answers, demos, decision | Keeps the process to weeks, not quarters |
Keep the document short enough that a vendor’s product team answers it, not only their sales team. Forty well-chosen questions get better answers than two hundred copied from a generic software template.
Decide your requirements before you send it
Your requirements come from four decisions you make internally first: how much you will score, who scores, which channels count and how the tool must connect. Each one turns a vague wish into a question a vendor can fail.
Step 1: Decide the coverage you need
A manual program usually reviews a small sample. If agent-level decisions, compliance checks or AI agent oversight depend on QA, a sample is rarely enough, as we explain in our post on why a 3% QA sample cannot support agent-level decisions. Write down whether you need every conversation scored, or a larger sample reviewed faster.
Step 2: Decide who has the final word on a score
Some teams want fully automated scoring; others want AI to draft and a reviewer to approve. Many want both, on different queues. Your RFP should ask how each mode works and how a reviewer overrules the AI.
Step 3: List your channels and systems
List every channel you want on the scorecard (calls, chat, email, messaging) and every system the conversations live in. A vendor that scores chat well but treats voice as a separate product will look fine in a chat demo and fail your phone queue.
Step 4: Write your knockouts
A knockout is an answer that removes a vendor whatever its score. Typical knockouts are no signed data processing agreement, no support for your helpdesk, or no way to see why a score was given. Keep the list to five or fewer, or nothing survives.
The 40 RFP questions, by section
These 40 questions cover the nine areas where QA tools differ most for contact centers. Copy them into your RFP, keep the numbering, and ask vendors to answer each one with a short description and the evidence you request.
Section A: Coverage and scoring (6 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| A1 | What share of our conversations can you score, and when is each one scored? | A specific share and trigger, such as every conversation once it is resolved |
| A2 | Which parts of scoring are fully automated and which need a reviewer? | A clear split, with both modes available |
| A3 | How does a reviewer change an automated score or comment? | An edit and approval flow, with the change recorded |
| A4 | How do you show why a score was given? | The evidence in the transcript next to each criterion |
| A5 | How do you score auto-fail criteria, such as sharing card data? | Auto-fail treated separately from weighted criteria |
| A6 | How do you handle conversations the AI is unsure about? | Routing to a human queue, not a silent guess |
Section B: Scorecards and your standard (5 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| B1 | Can we import our current scorecard as it is? | Your criteria in your words, not a forced template |
| B2 | How many automated criteria can we have, and are there limits per plan? | A number, per plan |
| B3 | How do we weight criteria and run different scorecards per team, queue or client? | Weights per criterion and scorecards scoped by team |
| B4 | When we change a criterion, what happens to future and past scores? | A documented rule, ideally with version history |
| B5 | How do you use our knowledge base or policies when scoring? | Your own content as context, not generic judgement |
For how to build the scorecard itself, see our guide to the QA scorecard.
Section C: Accuracy and validation (5 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| C1 | How can we test automated scores against our own reviewers’ ratings before go-live? | A test against historical ratings, criterion by criterion |
| C2 | How do you report agreement between automated and human scores? | Agreement per criterion, not one headline accuracy figure |
| C3 | What do you do when a criterion keeps disagreeing with our reviewers? | A fix to the criterion wording, then a re-test |
| C4 | How do you keep reviewers consistent with each other? | Calibration sessions that feed back into the scorecard |
| C5 | How does an agent dispute a score? | A dispute flow with an audit trail |
Section D: Channels and integrations (5 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| D1 | Which of our channels can be scored on one scorecard: calls, chat, email, messaging? | One standard across channels |
| D2 | How do you connect to our helpdesk, and is the integration native? | Named, native integrations for your systems |
| D3 | How do you handle calls: recording import, transcription, languages? | A clear voice path, not “on the roadmap” |
| D4 | How do conversations from an in-house system reach you? | A connector, API or file import, with formats |
| D5 | How do scores and data get out to our BI tools? | Scheduled exports in standard formats |
Section E: Coaching and reporting (4 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| E1 | How do QA results turn into coaching for an agent? | Feedback linked to the conversations it came from |
| E2 | What can a team lead see without asking an analyst? | Dashboards by agent, team and criterion |
| E3 | How do you surface trends and root causes across all conversations? | Analysis on full coverage, not a sample |
| E4 | How do you distribute review work across reviewers? | Assignments that spread work evenly |
Section F: AI agents and bots (3 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| F1 | Can you score conversations our AI agent handled on the same scorecard as human agents? | One standard for human and AI agents |
| F2 | How do you score a handoff from the AI agent to a person? | Checks on both sides of the handoff |
| F3 | How do findings feed back into the bot’s content or rules? | A route from failure to fix |
If AI agents already answer part of your queue, our guide on how to QA AI agents lists what to score.
Section G: Security, privacy and legal (5 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| G1 | Which independent audits and certifications do you hold, and can we see the reports? | Current reports available under NDA |
| G2 | Where is our data stored and processed, and by which sub-processors? | A published sub-processor list with locations |
| G3 | Do you use our data to train or fine-tune AI models? | A clear written answer, in the contract |
| G4 | What happens to our data at the end of the contract? | A return window and a deletion rule |
| G5 | Will you sign a data processing agreement, and what does it cover? | A DPA that covers the AI features too |
Section H: Implementation and support (4 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| H1 | What does onboarding include, and who does what? | A plan with your tasks and theirs |
| H2 | How long until the first conversations are scored? | A realistic time, with what it depends on |
| H3 | Who supports us after go-live, and on which terms? | A named contact and support terms |
| H4 | How do you help our reviewers and agents adopt the tool? | Training and a rollout plan, see our QA tool rollout plan |
Section I: Commercials (3 questions)
| # | Question | What a strong answer shows |
|---|---|---|
| I1 | What is the pricing unit (seat, reviewer, conversation, minute) and who counts as a seat? | A definition you can apply to your org chart |
| I2 | What is included in each plan, and what is an add-on? | A table, not a range |
| I3 | What changes at renewal or when our team grows? | Written renewal and growth terms |
How to score vendor answers
Score every answer from 0 to 3 against written definitions, average the answers within each section, then apply the section weights. Run security and data terms as a gate first, so a high feature score cannot carry a vendor who fails it.
| Score | Definition |
|---|---|
| 0 | Not offered, or no answer |
| 1 | Offered with major limits, or only promised for later |
| 2 | Offered and meets the requirement, evidence partly shown |
| 3 | Meets the requirement with evidence shown in the answer or demo |
Here is a weighting that suits a contact center whose main goal is fuller, fairer coverage. Change the weights to your priorities before the answers arrive, and do not change them after.
| Section | Weight | Gate |
|---|---|---|
| A. Coverage and scoring | 25 | |
| B. Scorecards and your standard | 15 | |
| C. Accuracy and validation | 15 | |
| D. Channels and integrations | 15 | Knockout if your helpdesk or main channel is not supported |
| E. Coaching and reporting | 10 | |
| F. AI agents and bots | 5 | |
| G. Security, privacy and legal | Gate | Knockout if any of G1 to G5 scores 0 |
| H. Implementation and support | 10 | |
| I. Commercials | 5 | |
| Total | 100 |
The weighted score for a vendor is the sum of each section’s weight times its average score, divided by 3. A vendor scoring 3 everywhere gets 100. Have at least two people score independently, then compare: where they are more than one point apart on a question, the question or the answer was unclear.
A worked scoring example
This is an illustrative example with three made-up vendors, not real results. It shows how the sheet behaves when totals are close and when a knockout applies.
| Section (weight) | Vendor A | Vendor B | Vendor C |
|---|---|---|---|
| A. Coverage and scoring (25) | 3 | 2 | 3 |
| B. Scorecards (15) | 2 | 3 | 3 |
| C. Accuracy and validation (15) | 3 | 2 | 1 |
| D. Channels and integrations (15) | 2 | 3 | 2 |
| E. Coaching and reporting (10) | 2 | 2 | 2 |
| F. AI agents (5) | 1 | 3 | 2 |
| G. Security gate | Pass | Pass | Fail (G3 scored 0) |
| H. Implementation (10) | 2 | 3 | 3 |
| I. Commercials (5) | 2 | 2 | 3 |
| Weighted score | 78.3 | 81.7 | Removed (would be 80.0) |
Vendor A: (25×3 + 15×2 + 15×3 + 15×2 + 10×2 + 5×1 + 10×2 + 5×2) ÷ 3 = 235 ÷ 3 = 78.3.
Vendor B: (25×2 + 15×3 + 15×2 + 15×3 + 10×2 + 5×3 + 10×3 + 5×2) ÷ 3 = 245 ÷ 3 = 81.7.
Vendor C would have scored 80.0, but it could not say in writing whether it trains models on your data, so the gate removes it.
The lesson is in the gap between A and B: 3.4 points is within the noise of two people scoring 40 answers.
Agree a tie-break rule in advance, such as “within 5 points, the higher score in the top-weighted section wins”. Here that is Vendor A, which scored 3 on coverage against 2 for B. Then confirm it with a scripted demo built on your top-weighted questions.
Run a scripted demo, then check references
A scripted demo shows each finalist answering your highest-weighted questions live, in the same order, so you compare the products rather than the presenters. Send the script a week ahead and score the demo on the same 0 to 3 scale.
A good script asks each vendor to show, not tell:
- one conversation scored on each of your channels, with the evidence behind each criterion
- a reviewer changing an automated score, and where that change is recorded
- your scorecard, or a close copy, set up with your weights and one auto-fail criterion
- the dashboard a team lead would open on Monday morning
- how a criterion change applies to future scores
After the demo, ask for one or two references with a team like yours in size and channel mix. Ask them what took longer than expected, not whether they are happy.
Once you have chosen, plan how you will check the scores on your own conversations in the first weeks. Our one-week protocol to validate AI QA scoring gives you the method.
Mistakes that make RFP answers impossible to compare
Most RFPs fail because the questions invite marketing answers and the scoring is decided after the answers arrive. Five mistakes cause most of it.
| Mistake | What happens | Fix |
|---|---|---|
| Yes/no questions | Every vendor answers yes | Ask “how” and “show us” |
| No weights before answers | The favorite wins after the fact | Sign off weights first |
| Price scored as a feature | A cheap tool that fails coverage wins | Keep commercials at 5 to 10 points, compare cost separately |
| Two hundred questions | Sales teams answer from a library | Keep to about 40 that matter |
| No knockouts | A vendor that cannot sign your DPA reaches the final | Write five or fewer knockouts |
Cost deserves its own model: license, onboarding, reviewer time and the time it takes to reach full coverage. Our guide to the ROI of Auto QA shows how to build it.
How to start your QA software RFP this week
Start by filling in the context table and your knockouts, then copy the 40 questions and the scoring sheet into your RFP document. Agree the weights with IT, legal, procurement and the QA lead in one meeting before anything goes out.
If Kaizo is on your list, here is what our answers draw on. AutoPilot scores 100% of conversations against your rubric once they resolve, and Auto QA drafts scores and feedback for a reviewer to edit and approve.
You can test the AI against your historical ratings before it goes live, and our Trust Center lists our SOC 2 Type II report, ISO 27001 certificate and data processing agreement, available on request. Pricing is per seat, set out on our pricing page.
See how Kaizo answers these questions in a live demo on realistic sample data, with your RFP in hand. Let’s talk.
Frequently asked questions
How do you write an RFP for contact center software?
Describe your operation first: seats, volume, channels, helpdesk and current process. Then list must-haves and knockouts, write questions that ask vendors to show evidence, and fix the scoring method and weights before you send it. Give a timeline for questions, answers, demos and the decision, and name one contact person.
What is the difference between an RFI and an RFP?
An RFI (request for information) is a short, early document to learn what the market offers and build a shortlist. An RFP (request for proposal) goes to that shortlist and asks for a specific solution, with answers you score and a price. Many QA software buyers skip the RFI and send an RFP to three to five vendors.
How many vendors should get your QA software RFP?
Three to five is usually enough. Fewer gives you little to compare; more creates scoring work that rarely changes the outcome. Use our comparison of QA software or a short call with each vendor to narrow the list first, and drop anyone who fails a knockout on that call.
How long should a QA software RFP process take?
For most contact centers, four to eight weeks from sending the RFP to a decision is realistic: about two weeks for answers, one to two for scoring and demos, and the rest for references, security review and contract. Security and legal review often take longest, so start them in parallel with demos.
Should price be part of the RFP score?
Price should be collected in the RFP but weighted lightly in the feature score, at around 5 to 10 points. Compare total cost separately, including reviewer time, onboarding and growth terms. A cheaper tool that scores fewer conversations or needs more manual review can cost more over the contract.
