The CX Platform RFP: 42 Questions to Ask Qualtrics, Medallia, and Everyone Else

TL;DR#

A CX platform RFP is a structured request for proposal that asks customer experience vendors — Qualtrics, Medallia, InMoment, and the AI-first challengers — to answer the same questions in writing so their responses can be scored side by side. Most CX RFPs fail not because the questions are wrong but because they are too generic to differentiate: "Do you support NPS surveys?" gets a yes from every vendor in the market. The 42 questions below are written to be unanswerable with a yes, and each one is paired with the answer pattern that should worry you. Gartner found that 56% of organizations reported a high degree of regret over their largest technology purchase in the prior two years, and that 74% of B2B buying teams show unhealthy conflict during the decision process — both are downstream symptoms of an evaluation that never forced vendors to be specific. This question bank covers six sections: listening methods, analysis, closed-loop action, integration and data portability, security and compliance, and commercial terms. Federal procurement rules require that all evaluation factors and their relative importance be stated in the solicitation itself; private-sector buyers should borrow that discipline. Run the RFP only after you have written a requirements document — otherwise you are asking vendors to design your program for you.

---

When a CX Platform RFP Is the Right Instrument (and When It Wastes a Quarter)#

A CX platform RFP is the right instrument when you have more than three credible vendors, a budget large enough to require procurement sign-off, and requirements stable enough to write down — and the wrong instrument in almost every other case.

The RFP earns its cost in three situations. First, when the buying group is large: Gartner's Future of Sales research puts the median enterprise technology buying group at 11 stakeholders for purchases above $100,000 in annual contract value, and a written record is the only way 11 people evaluate the same evidence. Second, when procurement or legal requires competitive documentation before signing. Third, when you genuinely cannot tell the vendors apart from their websites — which, in enterprise CX, is the normal state of affairs, because every suite markets the same eight capability names.

It wastes a quarter in three other situations. If you have not yet written down what you need, an RFP outsources your strategy to sales engineers; start with the customer experience platform requirements checklist instead and come back. If you already know you are buying one specific platform and the RFP is theater to satisfy a policy, skip to a scoped pilot. And if your shortlist is two vendors, a structured bake-off with real data beats 60 pages of prose from each.

There is a fourth failure mode worth naming: the RFP that over-specifies the solution instead of the outcome. Federal acquisition practice is instructive here — the Federal Acquisition Regulation's rules on evaluation factors require that "all factors and significant subfactors that will affect contract award and their relative importance shall be stated clearly in the solicitation," and that proposals be assessed "solely on the factors and subfactors specified." The discipline that imposes is not bureaucratic. It forces you to decide what you actually care about before you read a single glossy response, which is the single highest-leverage thing you can do to avoid purchase regret.

If you are still deciding whether to run a procurement at all, the build vs. buy decision framework for a customer experience platform is the prior question. An RFP assumes you have answered it.

---

How to Structure the 42 Questions#

Structure the RFP as six scored sections that move from what the platform captures to what it costs, because vendors are strongest on capture and weakest on cost — and reading in that order surfaces the gap.

Send every vendor the same file, require answers inline under each question, and cap each answer at 200 words. The word cap is not a courtesy — it is a diagnostic. Vendors who cannot answer a specific question in 200 words are usually answering a different question.

One rule to state in the cover page: marketing collateral attached in lieu of an inline answer scores zero. This single sentence removes roughly a third of the volume from a typical CX procurement and makes the responses comparable.

---

Section 1: Listening Methods and Data Capture (8 Questions)#

This section tests whether the platform can capture what customers actually mean, or only what fits a schema — the distinction that separates a modern CX platform from a survey suite with a dashboard.

Ask these eight:

1. List every listening channel your platform captures natively, and mark which require professional services to configure. Why it matters: "Supported" and "configurable by our team without services hours" are different products at different prices. Worrying answer: A channel list with no services column, or the phrase "our team will handle that for you."

2. When a respondent gives a vague answer — "it depends," "I'm not sure," "the usual issues" — what does the platform do next, in that same session? Why it matters: This is the highest-value moment in customer research and the one static forms cannot handle. Nielsen Norman Group's analysis of the ten most common survey challenges makes the mechanism plain: surveys are self-reported, so respondents filter everything before they share it, and no amount of question-writing fixes that. Worrying answer: "The verbatim is captured for analysis." That means nothing happens in-session.

3. Can the platform ask a follow-up question that was not authored in advance? If yes, who controls the boundaries of that follow-up? Why it matters: Adaptive follow-up is the actual dividing line in 2026. Branching logic is not adaptive follow-up — it is a decision tree someone wrote last quarter. Worrying answer: Conflating skip logic with AI follow-up, or an AI follow-up feature with no governance controls.

4. What is your median completion rate across deployed programs, and how is "completion" defined? Why it matters: Completion definitions vary wildly. Reaching the thank-you page and answering every substantive question are not the same event. Worrying answer: An aggregate figure with no definition, or a refusal on the grounds that "it varies by client."

5. Describe how a non-technical CX manager launches a new listening program end to end, with the number of business days and the roles involved. Why it matters: Self-serve capability determines whether the platform gets used after the implementation team leaves. Worrying answer: Any answer where a vendor employee appears in the critical path.

6. How does the platform handle multilingual capture — translation at capture, translation at analysis, or native-language models? Why it matters: Translate-then-analyze loses exactly the nuance you bought the platform to find. Worrying answer: "We support 100+ languages" without specifying where in the pipeline translation happens.

7. What percentage of a typical deployed program's data volume is structured (scores, scales, multiple choice) versus unstructured (open text, transcript, voice)? Why it matters: This one number tells you whether the vendor's installed base uses them as a survey tool. If 95% of captured data is structured, the AI layer is decoration. Worrying answer: Deflection to "we support both."

8. Show a single anonymized transcript or response record, unedited, from a live program in our industry. Why it matters: Everything else in this section is a claim. This is evidence. Worrying answer: A curated highlight reel, or a demo environment with synthetic data.

For the underlying capability map behind these questions, the 12 capabilities that separate a CXP from a survey tool is the companion piece. And if you want a working artifact to hand vendors as a capture spec, our voice of customer interview template and AI customer experience template both show the shape of an adaptive program rather than a questionnaire.

---

Section 2: Analysis and the "Why" (7 Questions)#

This section tests whether the platform explains why a metric moved or merely reports that it moved, which is the difference between a research capability and a reporting capability.

9. Walk through how the platform arrives at a root cause for a 4-point NPS drop in one segment. Name each processing step. Why it matters: Most "root cause" features are theme frequency counts. Frequency is not causation. Worrying answer: Any answer that ends at "our AI surfaces the top themes."

10. What is your theme-extraction accuracy, how was it measured, and against whose labels? Why it matters: Accuracy claims without a labeled benchmark and an annotator are unfalsifiable. Worrying answer: A percentage with no methodology, or "our models are trained on billions of responses."

11. Can an analyst ask a free-form question of the whole corpus and get a cited answer that links back to source responses? Why it matters: Traceability from claim to quote is what makes an insight defensible in a board meeting. Worrying answer: A generative summary with no citations back to individual records.

12. How does the platform distinguish a signal from 8 customers representing 40% of revenue from a signal from 800 low-value accounts? Why it matters: Unweighted CX analysis systematically over-indexes on whoever answers surveys. Worrying answer: No revenue or segment weighting in the analysis layer, only in the filters.

13. What happens to analysis quality at low sample sizes — 30 responses, 12 responses, 5 responses? Why it matters: B2B programs live at low n. A platform tuned for 10,000 retail responses degrades badly there. Worrying answer: A minimum-sample threshold below which the AI features simply stop working, disclosed only when asked.

14. Which analysis outputs are generated automatically versus configured by services, and what does reconfiguration cost after go-live? Why it matters: This is where year-two budgets disappear. Worrying answer: "Included in your success plan," with no hours or rate card.

15. Provide a sample insight report from a comparable program, with the source data volume stated. Why it matters: A two-page report built on 6,000 responses and a two-page report built on 60 are different products. Worrying answer: A template with placeholder findings.

Score this section against the outputs you already know you need. The seven numbers a CX scorecard should show the board is a useful forcing function: if a vendor's analysis layer cannot produce those seven, the section fails regardless of how the demo looked. For the adjacent question of whether you need a dedicated analytics layer at all, see how the customer analytics software market compares in 2026.

---

Section 3: Action and the Closed Loop (6 Questions)#

This section tests whether the platform moves an insight to a named owner with a due date, because a closed loop that ends in a notification is not closed.

16. Describe the full path from a detractor response to a resolved case, naming every system the record touches. Why it matters: Most closed-loop features stop at the alert. The remaining 80% of the loop lives in your CRM or ticketing system. Worrying answer: A path that ends inside the CX platform's own inbox.

17. What is the median time from response capture to case creation in deployed programs? Why it matters: Batch processing overnight makes real-time service recovery impossible. Worrying answer: "Near real time," undefined.

18. Can a frontline manager see and act on their own team's feedback without a full platform seat? Why it matters: Per-seat licensing that excludes frontline managers structurally prevents the loop from closing. Worrying answer: A viewer seat that costs the same as an analyst seat.

19. How does the platform track whether an action taken actually changed the metric it was meant to change? Why it matters: Without action-to-outcome attribution, the program cannot prove ROI at renewal. Worrying answer: Case-closure counts presented as outcomes.

20. What governance exists to prevent 400 alerts a day from being ignored within a month? Why it matters: Alert fatigue is the most common cause of CX program abandonment, and it is a design problem, not a training problem. Worrying answer: "Users can configure their own notification preferences."

21. Give one named or anonymized reference where a closed-loop program measurably moved retention, with the time frame. Why it matters: Six years of category marketing has produced very few of these. Worrying answer: A logo slide.

Loop design is where CX programs live or die operationally, which is why the operations teams and customer success teams pages exist as separate money pages — the workflows genuinely differ. If churn is the loop you care most about, the churn interview template shows what a depth-first version of the same loop looks like.

---

Section 4: Integration and Data Portability (7 Questions)#

This section tests whether you can get your data out as easily as the vendor gets it in, which is the only real check on renewal leverage.

22. List your pre-built integrations by system, and state for each whether it is bidirectional and who maintains it. Why it matters: One-way push into a data warehouse is not an integration with your CRM. Worrying answer: A logo wall with no direction or ownership column.

23. Provide your API rate limits, pagination behavior, and historical uptime for the last 12 months. Why it matters: "Open API" plus a 100-calls-per-minute cap is a closed API for anyone with real volume. Worrying answer: Rate limits described as "generous."

24. On termination, in what format do we receive our data, over what window, and at what cost? Why it matters: Under Article 20 of the GDPR, the right to data portability entitles data subjects to receive personal data in a "structured, commonly used and machine-readable format" — but your organization's right to bulk export of derived data is contractual, not statutory. If it is not in the MSA, it does not exist. Worrying answer: PDF or dashboard export only, or an export fee quoted as "professional services time."

25. Do we receive derived data — theme labels, sentiment scores, embeddings, model outputs — or only raw responses? Why it matters: Raw-only export means every year of enrichment stays with the vendor and you restart analysis from zero elsewhere. Worrying answer: Raw-only, framed as a security feature.

26. Which of your integrations require middleware or a third-party iPaaS, and is that license included? Why it matters: An undisclosed integration platform dependency is a five-figure surprise. Worrying answer: Silence on middleware until the statement of work.

27. What is the typical integration effort in person-days for our named stack, and who performs it? Why it matters: In enterprise CX programs, integration consistently consumes more budget than insight generation. Worrying answer: An estimate with no assumptions listed.

28. Can we host the platform in a specified region, and does that change the feature set? Why it matters: Regional deployments frequently lag on AI features by two or three releases. Worrying answer: "Yes" with no feature-parity statement.

Portability is the section vendors most reliably underanswer, and the one that most affects your position at renewal. The deeper treatment is in connecting CX data to the rest of the stack, which covers the middleware trap in detail.

---

Section 5: Security, Privacy, and Compliance (6 Questions)#

This section tests whether the vendor's AI features are governed to the same standard as its database, which in 2026 is a materially different question from a standard security review.

29. Provide your current SOC 2 report, stating type, period covered, auditor, and any exceptions noted. Why it matters: Type 1 assesses control design at a single point in time; Type 2 assesses design and operating effectiveness over a period, typically 6 to 12 months. The AICPA's SOC suite of services defines both. Vendors sometimes present Type 1 as equivalent. Worrying answer: "SOC 2 certified" with no type, no period, and no exceptions disclosed.

30. Is customer data used to train models that serve other customers? Answer yes or no, then explain. Why it matters: This is the question with the highest rate of ambiguous answers in CX procurement. Worrying answer: Any answer containing "aggregated and anonymized" without a technical description of the boundary.

31. Where do respondents learn they are interacting with an AI system, and can that disclosure be configured off? Why it matters: Transparency obligations under Article 50 of the EU AI Act became applicable on 2 August 2026 and require that people be informed when they are interacting with an AI system, with penalties reaching €15 million or 3% of worldwide turnover. A platform that lets you switch disclosure off is selling you a compliance liability. Worrying answer: Disclosure presented as an optional branding setting.

32. Which AI risk framework do you map controls to, and can you share the mapping? Why it matters: The NIST AI Risk Management Framework is the common reference point; a vendor with no mapping has no systematic view of model risk. Worrying answer: "We follow industry best practices."

33. Describe your data-retention defaults and whether retention is configurable per program. Why it matters: Indefinite retention by default becomes a discovery problem in litigation and a DSAR problem at scale. Worrying answer: A single global retention setting.

34. Who at your company can access our raw response data, under what approval, and is access logged and reviewable by us? Why it matters: Support-team access to verbatims is normal; unlogged, unreviewable access is not. Worrying answer: No customer-visible access log.

Regulated buyers should extend this section rather than trim it — the version for regulated industries is covered in Qualtrics alternatives for financial services and banking, where the compliance questions carry more weight than the feature questions.

---

Section 6: Commercial Terms and Total Cost of Ownership (8 Questions)#

This section tests the five-year number rather than the year-one number, because CX suite pricing is structured to make year one look reasonable.

35. Provide a five-year TCO model including license, implementation, integration, training, and estimated overages. Why it matters: The year-one quote routinely represents well under half of five-year spend once services and response volume are included. Worrying answer: A one-year quote with "we can discuss out-years later."

36. State your annual uplift cap in writing, and what happens if we decline it. Why it matters: Software is the fastest-growing segment of IT spend — Gartner's April 2026 forecast put worldwide IT spending at $6.31 trillion for 2026, growing 13.5%, with software growing faster than the total at roughly 15%. Uncapped uplift compounds against you. Worrying answer: No cap, or a cap tied to an index without a ceiling.

37. Define every metered unit — responses, seats, contacts, API calls, transcript minutes — and the overage rate for each. Why it matters: Multi-axis metering is where CX bills surprise people. You need every axis named. Worrying answer: "Usage-based, priced to your program."

38. Which capabilities in your proposal require an additional module or SKU not included in the quoted price? Why it matters: Demos are routinely assembled from modules across several price tiers. Worrying answer: A feature list that does not map line-by-line to the quote.

39. What is the minimum committed term, and what are the exit rights for non-performance? Why it matters: Without a performance-linked exit, a three-year term converts an evaluation error into a three-year cost. Worrying answer: Termination for convenience with 100% of remaining fees payable.

40. Break out implementation into fixed-fee and time-and-materials components, with the assumptions behind each. Why it matters: Unbounded time-and-materials implementation is the single largest source of CX budget overrun. Worrying answer: An all-T&M statement of work with an "estimate" that carries no cap.

41. What does year two cost if our response volume grows 50% and our seat count stays flat? Why it matters: A concrete scenario forces the pricing model into the open in a way a rate card never does. Worrying answer: A refusal to model the scenario.

42. Name three reference customers of our size who left your platform, and say why. Why it matters: Every vendor gives you happy references. The churn question tells you what kind of buyer this platform disappoints. Worrying answer: "We don't share that." (Note this as a scored deficiency, not a courtesy.)

Do the arithmetic yourself before you read the responses. Our CX platform total cost of ownership model is the calculator; the vendor-specific inputs for the two reference incumbents are in what verified buyers actually pay for Qualtrics, what Medallia costs and why buyers are rethinking the bill, Qualtrics implementation services, seats, and overages, and Medallia implementation cost and timeline.

---

How to Score the Responses#

Score the six sections with published weights, decided and circulated before responses arrive, and treat any unanswered question as a zero rather than a follow-up item.

A defensible default weighting for a modern CX platform evaluation:

Three scoring rules that do most of the work:

1. Score answers, not attachments. An answer that points to a PDF scores as unanswered.
2. Score specificity. A numeric answer with a stated methodology beats a superlative. If two vendors both say yes, the one who says yes with a number wins the point.
3. Score the evasions. Track which questions each vendor dodged. Dodge patterns cluster — a vendor who dodges questions 7, 13, and 25 is telling you their installed base runs structured surveys at high n and their data is not portable. That is a finding, not a gap.

Publish the weights in the RFP itself. This is the practice federal buyers are required to follow, and it works for the same reason there: it prevents the scoring conversation from becoming a negotiation about scoring after someone has a favorite. Gartner's 2025 finding that 74% of B2B buying teams exhibit unhealthy conflict during the decision process describes precisely the meeting you are trying to avoid.

For the full evaluation apparatus around the scorecard — pilot design, reference-call scripts, and demo scoring — use the vendor-neutral scoring framework for evaluating a customer experience platform. The RFP is one input to that framework, not a replacement for it.

---

Frequently Asked Questions#

How many questions should a CX platform RFP contain?#

A CX platform RFP should contain 30 to 50 substantive questions, which is enough to differentiate vendors without producing responses no one reads. The 42 questions here are calibrated to that range. Longer RFPs push vendors toward boilerplate and push your own reviewers toward skimming, which defeats the purpose. If you need to add questions for a regulated environment, add them to the compliance section and remove low-differentiation questions elsewhere.

What is the difference between a CX platform RFP and an RFI?#

An RFP asks vendors to propose a priced solution against stated requirements, while an RFI asks the market what is possible before requirements exist. Run an RFI when you cannot yet name your must-haves, and an RFP once you can. Skipping the RFI and using an RFP to learn the category is the most common sequencing error in CX procurement — it produces responses that educate you and scores that mean nothing.

How long does a CX platform RFP process take?#

A well-run CX platform RFP takes six to ten weeks from issue to shortlist: two weeks for vendor response, two for scoring and clarifications, and two to four for demos and reference calls. Processes that stretch past a quarter almost always started without a requirements document. Building the requirements first shortens the RFP itself, because the questions get sharper and the response volume drops.

Should we include pricing questions in the RFP or negotiate later?#

Include pricing questions in the RFP, but weight them lightly in scoring. You need the pricing structure early — metered units, uplift caps, module boundaries — because structure determines five-year cost and is hard to change later. The specific numbers are negotiable after you have a shortlist, which is why commercial terms carry 10% weight here rather than 30%.

Can we send the same CX RFP questions to survey tools and enterprise CXM suites?#

Yes, and you should, because sending identical CX RFP questions to both tiers is the fastest way to see where each tier genuinely stops. Survey tools tend to fail sections 2 and 3; enterprise CXM suites tend to fail sections 4 and 6. The comparison only works if the questions are identical and the scoring weights are fixed in advance. For orientation on how the two tiers differ before you issue, start with how the Qualtrics and Medallia suites differ in 2026.

What should we do if a vendor refuses to answer a question?#

Record the refusal as a zero for that question and note it in the scoring summary rather than granting an extension. Refusals are information: a vendor who will not state API rate limits, name churned references, or disclose SOC 2 exceptions during a competitive process will be no more forthcoming once you have signed. Consistent refusal across a section is grounds for elimination on its own.

---

Turning a CX Platform RFP Into a Decision#

A CX platform RFP is only worth running if it changes what you would otherwise have bought. The 42 questions above are built to do that: each one is specific enough that a vendor cannot answer it with a yes, and each is paired with the answer pattern that tells you what the platform actually is underneath the category language. Score the six sections with published weights, treat evasions as findings, and give listening methods the heaviest weight — because no analysis layer, dashboard, or closed-loop workflow can recover the "why" that was never captured in the first place.

That last point is the one worth carrying out of the process. Most CX platforms, including both reference incumbents, are architected around a survey: a fixed instrument, filled in, then analyzed. That architecture caps what section 2 can ever return, no matter how good the models on top are. If you want to see the alternative rather than read about it, the honest comparison is in Medallia vs. Qualtrics vs. conversational AI, with the wider market map in the enterprise CXM buyer's guide for 2026, the category history in what enterprise feedback management became, and the definitional groundwork in what a customer experience platform is and why AI is replacing the survey suite.

Perspective AI answers section 1 differently by design. Instead of sending a questionnaire, it runs an AI interview that asks the follow-up a researcher would have asked — probing "it depends," pulling on vague answers, and capturing the reasoning behind a score rather than the score alone. That means sections 2 and 3 start from transcripts with context rather than verbatim fragments, and the first insight arrives in days rather than after a services engagement. If you want a real answer to question 8 — an unedited transcript from a live program — the quickest path is to run one yourself: start an interview study, see what the first 90 days should produce, or review transparent pricing before you issue the RFP. It is built for CX teams and research teams who would rather test a platform than read a proposal about one.