A Five-Year Decline That Should Not Still Be Happening

Since Ofwat introduced the Customer Measure of Experience in 2020-21, customer satisfaction across the water sector has fallen every single year. By 2024-25, every water company in England and Wales was performing worse on C-MeX than it did in the first year of measurement. Not some companies. All of them. Five consecutive years of decline across an entire regulated sector, in a period when capital investment has been at record levels and customer experience has been a stated priority for almost every utility on the list.

The question that should be keeping customer directors and regulatory leads awake is not whether C-MeX performance needs to improve. It is why, after five years and billions of pounds of investment in customer service operations, the trajectory is still pointing in the wrong direction.

Part of the answer - a significant part, is that most water utilities are not actually reading what their customers are telling them. Not because the feedback does not exist. It does, in extraordinary volume. But because the analytical capacity to process it intelligently, consistently, and at speed does not.

The Feedback Is There. The Intelligence Is Not.

C-MeX is built on two survey streams. The Customer Service Survey captures the experience of customers who have recently contacted the company - after a supply interruption, a billing query, a planned works notification, a complaint. The Customer Experience Survey gauges broader satisfaction across the entire customer base. Together, under the PR24 framework running from 2025 to 2030, these surveys account for a financial incentive mechanism that Ofwat expects to represent up to 18% of a company's annual allowed revenue - a penalty or reward calculated on the distance between a company's score and the sector benchmark.

Both surveys include open-text fields. This is where customers stop ticking boxes and start talking. They describe, in their own words, exactly what happened, how it made them feel, and what the company could have done differently. A large water utility will collect tens of thousands of these open-text responses every year across C-MeX surveys alone - before counting transactional surveys after every engineer visit, contact centre interaction, and service event.

This qualitative data is the richest customer intelligence a utility holds. It contains the specific, granular reasons behind the scores - the communication failures, the repeat visits, the unexplained delays, the moments where a customer's experience tipped from neutral to negative. It tells you not just that satisfaction has fallen, but exactly why, in whose experience, and in what circumstances.

Most of it goes unread.

Not entirely - customer insight teams do analyse survey responses. But they do so manually, working through samples, applying subjective coding frameworks, and producing reports that are weeks old by the time they reach the people who could act on them. In an organisation receiving fifty thousand open-text responses a year, a team of three analysts working manually is reading a fraction of the available signal and interpreting it inconsistently enough that trend analysis over time is only partially reliable.

The Consistency Problem Nobody Talks About

Manual survey coding has a structural flaw that most customer insight teams are aware of but rarely discuss openly: it is not consistent.

Two analysts reading the same response will frequently code it differently. One sees a communication complaint. The other sees a service delivery complaint. Both are correct - the response contains both themes. But if coding decisions vary by analyst, by day, by workload, and by the evolving interpretation of what a theme means, then the trend data produced over time is not measuring what the customer said. It is measuring the variation in how the team coded what the customer said.

This matters enormously for C-MeX management. If a utility is tracking the proportion of responses that mention communication failures quarter on quarter, and the coding of that theme has drifted over time as analysts have turned over and interpretations have evolved, the trend line is not a reliable picture of whether communication performance has actually improved or deteriorated. It is a picture of how consistently the coding was applied.

AI-powered survey response theming eliminates this problem entirely. The same model, applied consistently to every response across every time period, produces coding that is comparable over time by design. Trend analysis becomes genuinely meaningful. Improvements in customer experience - or deteriorations - show up in the data as changes in what customers are actually saying, not as artefacts of analytical inconsistency.

What AI Theming Actually Does?

The mechanics are straightforward, even if the underlying technology is not. Natural language processing models are trained on the specific language and context of water utility customer feedback - the vocabulary customers use to describe supply interruptions, engineer visits, billing disputes, and planned works. Once trained, the model processes every open-text response in the dataset, identifying the themes present in each response, the sentiment associated with each theme, and the relationships between themes and other data points - survey type, customer segment, service region, contact reason.

The output is not a replacement for human insight. It is an amplifier of it. The customer insight analyst who previously spent three weeks manually coding a quarterly C-MeX dataset now receives a fully themed, fully coded dataset in hours - and can spend their time on interpretation, pattern recognition, and the strategic recommendations that require human judgement, rather than on the mechanical processing that does not.

The specific capabilities that matter most for C-MeX performance management are:

Whole-dataset analysis - every response is analysed, not just the sample that time allowed. The themes driving the lowest-scoring responses, the sub-segments with the sharpest declines, and the emerging issues that are not yet prominent enough to appear in a manual sample all become visible.

Emerging theme detection - AI identifies themes that are beginning to appear in the data before they are significant enough for a human analyst to notice. For C-MeX management, this is early warning of customer experience deterioration before it becomes a score problem.

Cross-survey comparison - theming applied consistently across C-MeX, transactional, and complaints surveys simultaneously gives a unified picture of customer sentiment rather than three separate reports that cannot be reliably compared.

Speed - a dataset that previously took weeks to code is processed in hours, making it possible to act on customer insight within the same operational period rather than the next one.

Closing the Loop Between Customer Intelligence and Operational Action

The ultimate value of AI survey theming for water utilities is not analytical - it is operational. The insight it generates is only commercially valuable if it reaches the people who can act on it quickly enough to change outcomes.

A utility that can identify, within days of a major planned works programme completing, that 23% of customer responses mention inadequate advance notice - and that this theme correlates strongly with low satisfaction scores in that service region - can act on that finding in the next planned works programme. A utility that identifies through AI theming that repeat visits are the single strongest predictor of low C-MeX scores in its contact-based survey can direct operational investment at first-time fix rates rather than contact centre staffing.

This is the connection that manual reporting cycles break. By the time a manually coded quarterly report lands, the operational window to respond to the insight has often closed. AI theming keeps the intelligence current enough to be actionable.

For a sector in which every company has delivered five consecutive years of C-MeX decline - in a regulatory period where the financial consequences of continued underperformance are explicit and material - the ability to read customer feedback faster, more completely, and more consistently than the competition is not a marginal advantage. It is the difference between understanding why satisfaction is falling and continuing to invest in the wrong places while the scores keep moving in the wrong direction.

The feedback exists. The question is whether the intelligence does too.

Our AI-powered customer intelligence solutions are designed to help utilities turn survey volume into actionable insight - faster, more consistently, and at a scale that manual analysis cannot match.

To discuss how AI survey response theming could support your C-MeX performance programme, speak to our utilities practice. Visit our AI solutions for more information