For about two decades, the answer to “what do our customers think” was simple. Send a survey. The Net Promoter Score, introduced in 2003, became the common language of customer satisfaction, a single number that executives could put on a slide and track from one quarter to the next. That number is not going away. What is breaking down is the assumption behind it: that one periodic questionnaire, answered by whoever bothers to respond, can stand in for the full voice of the customer.

A group of AI companies is reshaping what listening looks like, and what they are building is a more complete picture of the customer than any survey could provide on its own.

The pressure on the survey

The pressure on traditional surveys is well documented. Response rates have fallen for years as people get asked for feedback constantly, after a purchase, after a support chat, after an app download, and learn to ignore most of the requests. The exact numbers vary by source and channel, but the direction is not in dispute, and it creates a quieter problem than low volume alone. As fewer people respond, the ones who do tend to sit at the extremes, the delighted and the angry, which leaves a score drawn from an increasingly narrow and self-selected group.

A metric built on a shrinking sample starts to drift away from what the broad middle of customers actually thinks. Gartner even predicted in 2021 that most organizations would drop NPS as a customer service and support metric by 2025. That prediction was only partly borne out. NPS did lose ground: by 2025, a TELUS Digital and Statista survey found just 23% of enterprise customer-experience leaders still using it to measure performance. But it was downgraded within the stack rather than abandoned, and it remains widely used. It captured a real loss of confidence in any single survey score as the last word on customer sentiment.

Smarter surveys and a new source alongside them

The response to that pressure has not been to abandon surveys. It has been to change them, and to surround them with other sources.

On the survey side, the instrument itself is getting smarter. Instead of a static form with a fixed list of questions, AI-driven surveys can adapt as they go, ask intelligent follow-ups when an answer is vague, and keep the exchange short enough to avoid fatigue. The aim is fewer, better questions that pull richer answers from the people who do engage and turn open-ended replies into something a team can actually quantify. Unwrap, a customer-intelligence startup founded in 2022, sells exactly this kind of conversational survey product alongside its core analytics, built to ask adaptive follow-ups and convert free-text answers into structured insight. The survey, then, is part of where the category is heading.

The larger change is what now sits next to the survey. Customers produce an enormous volume of feedback without ever being asked: product reviews, support tickets, chat transcripts, app store comments, social posts, and messages in community forums. That unsolicited feedback has always existed, but reading it at scale was impractical. Advances in natural-language processing and large language models changed that, enabling sorting of hundreds of thousands of free-text comments, scoring their sentiment, and surfacing issues as they emerge. The work has a name now: customer intelligence, a descendant of what used to be called voice-of-the-customer analytics, and a wave of companies is building tools for it.

Unwrap is among the more visible. Ryan Millner and Ashwin Singhania, two former Amazon Alexa product leaders, founded it after building the data company Graphiq, which Amazon acquired in 2017. It spun out of the Allen Institute for AI’s startup incubator before settling in Santa Barbara, California. In January 2025, it raised a $12 million Series A led by Scale Venture Partners, a round Fortune reported as an exclusive. The company positions itself around analyzing unstructured feedback at scale and serving product teams, not only the research and customer-experience groups that have historically owned survey programs. It lists customers including Microsoft, GitHub, DoorDash, Lyft, JetBlue and Perplexity. Its lineup spans both sides of this shift: a Customer Intelligence platform that reads unstructured feedback, a Surveys product, and a support-quality tool. The pitch is about gathering all channels, including surveys, into a single view of the customer.

The argument running through the category is that unsolicited feedback is both more plentiful and more candid than the kind collected in surveys. Someone writing a review or filing a ticket is describing what genuinely bothers them, including problems no survey thought to ask about. The catch is that unstructured data is harder to trust. Sentiment models misread sarcasm and context, and free text lacks the clean comparability of a score you can chart over time. That is why most serious practitioners, Unwrap included, describe surveys and unstructured analysis as complementary rather than a straight swap. Surveys give you a stable benchmark and a comparable trend line. Unstructured analysis tells you what is driving the trend and catches issues a fixed questionnaire would never surface. The strongest programs run both, and weigh each for what it is good at.

A competitive field

Unwrap is not working in open water. Qualtrics and Medallia have spent years building large enterprise footprints and are now layering generative AI features into their suites. A set of AI-native rivals, among them Enterpret and Thematic, is chasing the same unstructured-feedback problem, several of them with a head start. Software-comparison sites routinely list all of these companies together, which is a fair measure of how crowded the positioning has become. The breadth of the field also reflects how unsettled the category still is: there is no obvious default, as there once was with NPS for measurement.

That sets up the question at the center of Unwrap’s prospects and those of its peers. Can a focused, AI-native tool hold its own against entrenched suites that can fold feedback analytics into a much bigger contract, and against a field of startups making a similar pitch? The incumbents have distribution and the pricing leverage that comes with it. The newer companies argue they were built from the start for this specific problem, rather than bolting a feature onto a platform designed for something else. Which advantage proves more durable is still an open question, and it is the kind of question that usually gets answered through consolidation.

Where this lands

What is clearer is the direction of travel. Companies across industries are steadily pointing language models at internal text that used to be too voluminous to use, and customer feedback is one of the most obvious targets. The near-term result is a change in the survey’s role, from the system of record for customer sentiment to one instrument in a wider kit, sitting alongside the unstructured stream and increasingly read by the same software.

For companies like Unwrap, the opportunity is to be the layer that ties those instruments together, the smart survey and the firehose of unsolicited feedback, into a single, current read of what customers want. The risk is the familiar one for any focused tool: that the capability becomes a standard feature inside a larger platform before an independent player gets big enough to stand on its own. Either way, the survey now shares the job with everything else customers say. The companies that win this category will be the ones that can turn all of it, asked and unasked, into something a team can act on quickly.