CX platforms need to connect customer feedback to action

Evaluate CX platforms by the action they enable

An AI assistant can explain why customers are unhappy while leaving the organization unable to respond. For a customer experience (CX) platform buyer, the decisive question is whether that explanation reaches an accountable team through a workable process, with controls over customer data and the response. The platform’s architecture and the organization’s readiness must support that whole sequence.

General-purpose AI makes this question more pressing. Forrester describes a “ChatGPT effect”: buyers question the need for a dedicated feedback platform when a general-purpose model can summarize sentiment. A dedicated platform earns its place by connecting analysis and conversation intelligence, the analysis of customer interactions, to operational systems and governance. Those connections depend on interoperability: the ability to exchange and use information across systems.

But making those connections creates work for the teams running the platform. Fragmented systems, limited access to IT or professional services, and modest operating maturity can constrain adoption. Integration helps information move, but the analysis must be reliable, the recommended response must fit the business, and someone must have the authority and capacity to act.

Use rankings to frame the evaluation

For buyers building a shortlist, the current competitive assessment is The Forrester Wave: Customer Feedback Management And Analytics Solutions, Q3 2026, published this week. Forrester analysts Colleen Fazio and Boris Evelson developed the guidance with contributions from Rick Parrish, Angela Lozada, and Shayna Neuburg. The assessment describes vendor capabilities, while CMSWire’s interviews show how vendors explain their positions and priorities.

In that assessment, Medallia is the sole Leader, with a substantial lead. Qualtrics, Sprinklr, and CallMiner are Strong Performers; Alchemer, Cresta, and Salesforce are Contenders. These tiers help buyers identify products to examine, while implementation requirements determine how each would fit their organization.

The assessment’s scope matters when comparing it with earlier rankings. The 2024 Wave for Text Mining and Analytics Platforms Q2, 2024 named Medallia, InMoment, Qualtrics, expert.ai, and Google as Leaders; InMoment is now part of Qualtrics. The 2024 Q4 Customer Feedback Management Solutions Wave named Medallia and Qualtrics as Leaders. The current evaluation combines those previously separate Waves, placing Qualtrics as a Strong Performer after its Leader positions in both predecessor assessments.

The rankings give buyers a starting point for more specific questions. They still need to examine how customer records connect, who owns decisions, and how teams will change their work. Those conditions determine whether a highly rated capability becomes useful in practice.

Establish how customer information connects

Research on CX systems shows why those conditions need scrutiny. A 2026 Oxford Economics study conducted with SAP surveyed 300 senior CX decision-makers and found that 25% described their CX technology environment as fully harmonized and integrated, while 29% reported highly fragmented environments. SAP sells enterprise software, including CX systems, and has a commercial interest in investment that connects these environments.

The study also linked these environments to difficulty using customer information. Among respondents with siloed CX technology, 58% reported an inability to connect customer needs to actionable insight, compared with 47% in harmonized environments. Buyers therefore need to examine both the connections between systems and the work required to turn connected information into a decision.

Fragmentation also appears in Zendesk’s CX Trends research, where 54% of organizations identified fragmented or siloed data as their single biggest obstacle to using data effectively. Zendesk sells customer-service software and benefits when organizations invest in addressing these problems. Its finding concerns barriers to effective data use, while the Oxford Economics study compares reported difficulty across different technology environments.

Forrester’s guidance adds a specific architectural question. Fazio and Evelson identify gaps in the availability of configurable unified data models, which give information from different systems a shared structure that an organization can adapt. Linking customer, profile, and other entity records through that structure helps a CX program interpret feedback across the channels where a customer interacts.

Connected records provide the foundation, but teams need a way to use the resulting interpretation. So evaluation has two parts: how the platform receives, connects, and analyzes information, and how the organization controls and acts on it. The latter brings workflows, decision authority, and organizational change into the purchase decision.

Examine measurement reliability as models change

Once information is connected, buyers need to know whether changes in the analysis reflect changes in customers. Sonal Singhal, Sprinklr’s director of product management, framed the question directly: “Did my customers change, or did my model change?” Sprinklr sells CX software whose value depends partly on reliable measurement, so its position on model controls also supports its commercial offering.

Forrester recommends hybrid, or neurosymbolic, AI to combine flexible interpretation with structured business meaning. This approach joins generative AI and machine learning with explicit business rules, taxonomies that organize feedback into categories, ontologies that define concepts and their relationships, and knowledge graphs that represent those relationships in connected data. Each method offers a different way to interpret customer evidence.

The combination matters because each method has limits. Large language models, or LLMs, can produce inconsistent results and explanations that fall short of what a business needs to scrutinize them, while rules and ontology-based systems can struggle with the volume and complexity of modern feedback. Buyers need to examine how the methods work together on decisions their teams actually make.

Consistency needs scrutiny too, because a taxonomy can repeatedly exclude relevant customer language. As Singhal put it, “If a rules-based taxonomy consistently misses semantically relevant feedback, the trend is perfectly comparable over time and still systematically incomplete.” A “stable trend line” can therefore conceal missing feedback, leaving managers with a dependable-looking measure that omits part of the customer experience.

Model updates create another problem: better interpretation can change reported results even when customer behavior stays the same. Singhal said organizations must track model versions and evaluate changes against ground truth, meaning reference examples with an established expected interpretation. Confidence scores alone cannot explain whether a shift comes from customers, model behavior, or both.

Sprinklr’s evaluation also put these controls under scrutiny. Forrester identified gaps in longitudinal tracking, meaning comparisons of results over time, and in confidence scores, while Singhal emphasized the engineering required to address this class of problem. “That’s an architectural commitment, not a field placed next to an LLM output,” she said. Buyers should therefore inspect model lineage, the record of which model version produced a result, alongside evaluations using reference examples.

Match deployed capabilities to the team’s operating capacity

Reliable analysis still needs people who can use and maintain it, so product capability and customer maturity need separate evaluation. Sid Banerjee, Medallia’s chief strategy officer, described a three-phase AI strategy in the CMSWire interviews and said 600 to 700 customers use the company’s AI-accelerated analytics. Medallia sells these capabilities, so its account of deployment and adoption also supports the case for buying its platform.

Beyond that deployed analytics base, Banerjee described a second phase expanding conversational interfaces, including Insights Assistant, with a broader rollout happening “this year.” The third phase develops further agentic capability, in which AI agents perform coordinated tasks. Buyers therefore have different things to inspect: existing use, rollout plans, and capabilities under development.

Insights Assistant already uses a multi-agent process, according to Banerjee. One agent interprets the user’s intent, another produces a precise query, and a third analyzes the returned results. This separates interpretation, data retrieval, and analysis before producing an answer. “This isn’t aspirational,” he said. “That is how Insights Assistant works today.”

That working sequence gives buyers something concrete to test, including the tasks that make the interface usable. Forrester’s customer feedback identified filter complexity and gaps in AI topic tuning, which affects how feedback is grouped into useful subjects. Banerjee responded with relative ratings: “We didn’t get the lowest ratings on anything … that’s a testament to the fact that we were focusing on the right things.” Buyers can weigh Medallia’s response against tests of the filtering and tuning their own teams would perform.

Usability tests also need to reflect customer maturity. Medallia’s CX Evolution Framework runs from one through five, and Banerjee said, “The majority of customers are level one.” At that level, customers use traditional surveys sent after an experience, while he described the level-two group interested in greater AI enablement as “a healthy minority.”

The framework’s higher levels involve a different operating model. Levels four and five involve agent-to-agent interconnection and are “kind of being defined” as Medallia develops those capabilities, Banerjee said. Buyers should examine the team processes those stages require alongside the existing query-and-analysis sequence they can test today.

Implementation adds another constraint, especially where teams have limited support. Alchemer’s CEO, Martin Mrugal, described fragmented data, siloed systems, implementations requiring IT, changes requiring professional services, growing feedback-channel counts, and rising customer expectations. In his account, these problems predate AI, while AI increases the speed at which businesses operate.

Alchemer sells a lower-effort approach to those demands, so Mrugal’s claims about ease and cost form part of its commercial pitch. “There are no professional services or IT resources needed because it’s intuitive. This keeps TCO low and helps teams move fast,” he said. TCO means total cost of ownership; buyers can test the claim by asking their teams to perform the implementation and maintenance changes they expect to need.

Ease of use also needs to be tested alongside integration. Customers in Forrester’s evaluation reported that Alchemer’s newer AI offering, Pulse, felt disconnected from the core platform, while Mrugal noted that Forrester’s qualitative feedback involved two to four customer interviews per vendor. He reported a Pulse renewal rate of more than 90% and cited Malwarebytes as a customer using it to create significant new revenue streams. Those measures of customer value address different questions from the work required to connect Pulse to an existing process.

The planned launch of Alchemer Iris later this year is intended to bring deeper integration. A buyer considering Pulse therefore needs to examine whether its current connection to the platform supports the intended work and when the planned changes would matter. That timing determines how much implementation work the team must carry before deeper integration arrives.

Follow each vendor’s route from feedback to action

Once buyers understand their implementation capacity, they can examine how different products move analysis into business activity. Vendors emphasize different routes: experience history, interaction analysis, cross-functional workflows, or broader conversation coverage. Each route changes the information available for a decision and the systems through which a response can happen.

Qualtrics emphasizes outcome orchestration, which Forrester describes as “using experience context to anticipate needs and improve retention and customer lifetime value.” An unnamed company spokesperson said its XM Data and AI platform also encompasses simulation, prediction, and trusted outcomes. Qualtrics sells that platform, so its argument for experience context also makes the commercial case for its approach.

Access to that context depends on the platform’s current architecture. The semantic layer, which gives data shared business meaning, and the knowledge graph identified in the Wave are currently limited to XM Discover; extending them is part of the roadmap. The spokesperson argued, “A general-purpose model can guess at what should happen next, but it’s guessing without experience context.” Buyers should test which experience information can inform decisions today and which requires the planned expansion.

Qualtrics also proposes changing how organizations judge the value of those decisions. Its spokesperson advocated moving the primary measure of success from Net Promoter Score, or NPS, to customer lifetime value: “NPS told you how someone felt in a moment. It never told you what that moment would actually be worth to the business. Lifetime value does.” Measuring value across the customer relationship changes how teams prioritize responses, so adopting that approach requires agreement on reporting and accountability.

CallMiner takes a different route, combining detailed interaction analysis with capabilities for taking action. Forrester currently sees it as more likely to supplement traditional feedback platforms than replace them, while Bruce McMahon, CallMiner’s chief product officer, describes supplementation as a transitional position. CallMiner has a commercial interest in expanding its role within customers’ technology environments. Its ambition is eventual replacement.

Deployment requirements help determine whether that expanded role fits a buyer. Forrester criticized CallMiner’s cloud-only deployment, while McMahon defended cloud-first deployment as an enabler of AI at scale: “We do not view cloud-first deployment as a constraint. In many cases, cloud is necessary to deliver powerful AI-modeled capabilities at scale.” He identified digital behavior analytics as the company’s top priority for addressing remaining gaps, giving buyers another question about the completeness of its interaction evidence.

Sprinklr’s route depends on connecting analysis to a broader native platform. Singhal said its contact center and marketing capabilities operate alongside feedback management, allowing data from different sources to connect directly with cross-functional workflows. Activation of autonomous agents is the next development, so buyers should follow an existing workflow first, then examine the additional actions planned for agents.

Cresta broadens the information available for decisions by analyzing existing customer conversations. Devon Mychal, Cresta’s vice president of product marketing, said, “Surveys are no longer the only way to understand the voice of the customer.” Cresta sells conversation-intelligence capabilities, and Mychal claimed AI analysis across 100% of conversations, covering outcomes, root causes, specific behaviors, influencing factors, and recommended action.

Mychal identified predictive customer satisfaction, or predictive CSAT, as one application: estimating satisfaction from conversation evidence. He said business users can access the analysis through natural language, reducing work that previously required data scientists and extensive manual review, and cited United Airlines and CVS Health as enterprises adopting the approach. Buyers can test how that additional evidence improves decisions alongside surveys and follow the recommendations into a workflow where someone can use them.

Test the complete sequence through a buying evaluation

These different routes make a process-based evaluation useful. Start with the feedback the organization receives, follow how the platform interprets it, and continue through to an authorized response. At each stage, examine the buyer’s responsibilities alongside the product’s capabilities.

1. Start with ingestion, meaning bringing information into the platform, and the data model. Establish whether the platform can accept the required information sources at the necessary scale. Then inspect how its unified model accommodates the organization’s data and ask to see configuration changes, since that flexibility varies across vendors.
2. Examine identity and entity linkage. Identity resolution establishes which records refer to the same customer, while entity stitching connects related customer, profile, and other records using unique identifiers. Follow those links across channels and establish which identification decisions remain the buyer’s responsibility.
3. Establish governance before authorizing action. Identify who can access the evidence, how interpretations are reviewed, and who can approve or execute a response. Procurement can assess product controls, while operational teams define responsibilities and acceptable use.
4. Test analytics and AI controls together. Examine how hybrid methods interpret relevant feedback, then inspect model-version records, ground-truth evaluation, and explanations. Use plausible summaries and confidence scores as starting points for testing decision reliability and comparability over time.
5. Follow an insight into the receiving workflow. Establish which team receives it, which system carries it, and what response the connection enables. Test existing integrations and functioning assistants, then assess the timing and dependencies of plans for broader context, deeper integration, or autonomous agents.
6. Make deployment and service dependencies explicit. Identify work requiring IT, professional services, cloud deployment, or future product changes. Evaluate claims about ease, conversation coverage, retention, revenue impact, and eventual platform replacement against the particular requirement each claim addresses.
7. Test the operating change required for adoption. Determine who will maintain identities, govern decisions, and change team processes after deployment. If the business intends to move from NPS toward lifetime value, establish how that change affects prioritization, reporting, and accountability.
8. Assign responsibility for custom development. Forrester’s guidance is concise: “Build only where differentiation justifies the burden.” For bespoke identity logic, analytical methods, or workflow connections, identify the business advantage that warrants the continuing work, then assign ownership for maintenance and future changes.

The bottom line

A CX platform’s value depends on what the organization can do with its analysis. Rankings and AI capabilities help frame the choice, but the buying decision needs to account for connected customer records, reliable measurement, clear decision authority, and the capacity of teams to respond.

For executives, this makes platform selection an operating decision as much as a technology purchase. Before committing, ask teams to demonstrate a complete sequence using a relevant business problem: connect the feedback, explain the finding, route it to an accountable owner, and show how that owner can act. Identify where the sequence depends on manual work, outside support, or capabilities still on the roadmap.

That test also clarifies where investment is needed. A platform may address gaps in analysis while leaving identity management, governance, or workflow ownership to the buyer. Those responsibilities need funding and named owners if the purchase is to deliver value.

The strongest choice is the platform whose deployed capabilities fit the organization’s priorities and capacity, with a credible path to the next stage. More sophisticated analysis becomes useful when the business can turn it into a reliable, authorized response.