Customer feedback management as a named discipline is younger than the practice. The practice is as old as commerce. The discipline was assembled in pieces — first in industrial production, then in marketing science, then in software.
This is the history of customer feedback management as it shows up for software product teams. The industrial roots matter — they're where the vocabulary comes from — but the practical lineage that shaped the tools and methodologies we use today runs through software, not factories.
The industrial roots (1931–1984)
Walter Shewhart at Bell Labs (1931) connected consistency to customer trust. W. Edwards Deming carried statistical process control to postwar Japan, where the Japanese Union of Scientists and Engineers built it into Total Quality Management — the customer at the centre of every production decision. Noriaki Kano's 1984 paper Attractive Quality and Must-Be Quality classified customer requirements into five categories (must-be, one-dimensional, attractive, indifferent, reverse) and showed that customer satisfaction is non-linear. Must-be features customers never thank you for; absent them, you lose them. Attractive features no customer asks for; present them, and you win them. Kano gave product teams the first language for the difference between raising the floor and raising the ceiling.
1993: "Voice of the Customer" formally named
Abbie Griffin and John Hauser's The Voice of the Customer (Marketing Science, 1993, MIT Sloan) codified the term and the method — structured interviews, affinity diagramming, and systematic translation of customer needs into product requirements. VoC became the organising name for the discipline in marketing and new product development.
1990 / 1996: the retention economics
Frederick Reichheld and W. Earl Sasser's Zero Defections (HBR, September 1990) produced the finding that anchors most modern customer-retention work: a 5% improvement in retention produces 25% to 95% in profit growth. Reichheld's Loyalty Effect (Bain, 1996) extended the argument across customers, employees and investors. The cost of losing the customer — and by extension the cost of not listening to them — moved from intuition to a calculable number.
1999–2002: survey software democratises
SurveyMonkey (1999), Medallia (2000) and Qualtrics (2002) turned survey distribution from a research-team speciality into something a marketing manager or a PM could run themselves. The output was the same as before — a survey result — but the friction of getting it collapsed. The next decade of feedback management was shaped by the consequences of that collapse.
2003: NPS
Reichheld's The One Number You Need to Grow (HBR, December 2003) proposed Net Promoter Score — one question, 0–10, promoters minus detractors. NPS became the most widely adopted customer-feedback metric in history. It was also the most contested. Keiningham, Cooil, Andreassen and Aksoy's 2007 Journal of Marketing replication across 21 industries found NPS no better than alternative satisfaction metrics at predicting growth. The number stayed. The debate did too.
2008–2017: product-feedback software emerges
UserVoice launched in 2008 as the first dedicated product-feedback tool — Joel Spolsky's voting model from Stack Overflow applied to customer requests. Aha! shipped in 2013 from Brian de Haaff and Chris Waters — roadmap planning and product strategy first, feedback capture second. Pendo (also 2013) came at the same problem from the other side: in-app analytics, NPS surveys triggered at the moment of friction, behaviour data feeding feature decisions. Productboard launched in 2014 with Hubert Palan's bet that feedback collection and roadmap synthesis belonged in the same tool — themes extracted from support tickets, sales calls and feedback portals, then organised against a strategic plan. Canny followed in 2017 with a lighter, SMB-friendly bet: a public voting board and a changelog, nothing else. Dovetail launched the same year out of Sydney, focused on qualitative research repositories.
2021–2026: the AI shift
Large language models changed what the analysis layer could do. Manual tagging — the brittle, drift-prone backbone of every pre-AI feedback tool — was always going to fail at scale. Embedding-based semantic clustering let teams surface themes from raw text without a predefined taxonomy. Goal-driven LLM clustering — described in the EMNLP 2023 ClusterLLM paper from Zhang et al. — moved the state of the art further: describe the clustering goal in natural language, the LLM returns themes with descriptions. The question stopped being "can we extract themes?" and became "what should we do with them?"
The history of customer feedback management is the history of each step being absorbed into infrastructure, then the next step becoming the bottleneck.
Collection was the bottleneck. Survey software solved it. Analysis was the bottleneck. LLMs are solving it. Decisions are the bottleneck now — and the discipline is being rebuilt around what to do when collection and analysis happen continuously in the background while the team is building.
Why customer obsession fades at scale
Almost no team decides to stop being customer-obsessed. The fade happens five specific ways.
Each one is structural. None of them are a failure of intent. They are failure modes of a system that is asked to scale linearly when customer volume scales exponentially.
The loudest voice
Recency and volume beat truth.
The agency on your fintech who Slacks daily about a colour gets four roadmap items. The enterprise paying ten times as much churns because the SAML they raised once in a sales call never made the backlog.
The black hole
Feedback in. Feature ships. Customer never told.
A data analyst at a marketplace startup requests scheduled exports in March. The feature ships in October. They never find out it was theirs and tweet that the product never listens.
The planning gap
Decisions made on a picture that has already moved.
The quarterly planning prioritises a feature twelve customers asked for last quarter. By the time the session ends, three of them have churned and two have built the workaround.
The sequencing trap
Right things, wrong order.
Stripe Connect support ships two months after the three biggest enterprise accounts had already moved their payment workflow elsewhere. Everyone agreed Connect mattered. Nobody had read the signal in time.
The vision vacuum
Strategy debate replaces customer signal.
The product team spends six weeks debating AI agents vs vector search. Both arguments are internally coherent. Neither references a single customer call from the last sixty days.
The fade isn't a failure of intent. It is a failure of system.
Each of these modes is solvable. None of them are solvable by trying harder. They are solvable by infrastructure — the kind of infrastructure that processes customer signal continuously, scores it against the floor and the ceiling, surfaces the priority that needs attention now, and closes the customer feedback loop when the work ships.