Could Natural Language Understanding transform trust in utility CX?

Utility providers are fighting a dual battle. Alongside cost pressures, there is an ongoing crisis in customer trust. Just one in ten utility consumers trust their suppliers, according to latest research. Compounding this issue, earlier this year, Ofgem issued a tough-to-read report on the state of customer service across 17 of the biggest domestic energy suppliers in the UK. From weak policies and unclear pathways for customer service journeys to inconsistent scripts for staff handling complex calls, all the suppliers reviewed as part of the Market Compliance Review had areas for improvement with customer service.

Loyalty, trust and customer experience (CX) are, of course, inextricably intertwined. In a complex and competitive customer environment, utility providers should deliver clear, consistent and customer-focused experiences. Many have already turned to AI to assist in this mission - chatbots powered by Natural Language Processing (NLP) are commonplace. Now, a subset of NLP - Natural Language Understanding (NLU) - is taking AI-empowered CX even further, transforming customer interactions and how utility companies can integrate next-gen AI to ensure trustworthy CX.

NLU: A deep dive into customer intent

As CX leaders know, great customer service is a complex recipe of speed, ease, convenience and personalisation. In utility sectors, customers often only get in touch when things go wrong or when basic needs are compromised, leading to heightened emotions. This is where NLP CX technologies can sometimes fall down.

NLP is instrumental in enabling machines to interact with users through text or speech interfaces. It powers tasks like generating responses in chatbots and processing voice commands, so is primarily concerned with processing language data, transforming it into structured formats that machines can work with.

NLU goes beyond NLP. While NLP deals with surface-level language processing, NLU delves deeper into the nuances of comprehension. It delivers the ability to not only recognise and process language but also understand the context, intent, and meaning behind it. It aims to capture the nuances, intentions, and emotions conveyed by language in a way that mirrors human understanding.

NLU’s ability to understand language idiosyncrasies, including colloquialisms and dialects, is essential for utility companies seeking to accurately identify why a customer has contacted them (their ‘intent’). Not only that, but NLU can tailor responses and recommendations based on the individual customer's usage patterns and preferences while also identifying situations that require human intervention, and escalating them appropriately.

With rising levels of customer vulnerability and increasing negative sentiment towards energy companies triggered by the cost-of-living crisis, it’s more critical than ever that utilities understand customer intent. Technology that helps provide rapid, nuanced support increases self-service options while enabling human agents, ultimately ensuring all pathways are embedded with personalised, empathic communication.

Implementing NLU for enhanced self-service

NLU technology’s ability to identify intent has numerous applications, not least to support greater self-service options. However, embedding NLU technology, even with the most basic functionality, requires attention to detail beyond NLP, with specific AI model training.

Customer communications within utilities can be complex, nuanced and niche. The energy sector has a rich set of industry-specific technical terms - basic examples include ‘smart meter’, ‘price cap’ and ‘fixed tariff’.  In specialised sectors like energy, NLU models may struggle to recognise and interpret industry-specific jargon, acronyms and terms. Typical AI training models for generic customer service use will lack this detail. Utility NLU technologies will only succeed when thoroughly and expertly trained using custom large language models fine-tuned for specific industry terminology.

Training NLU for use in utility CX involves a multi-step process that combines data collection, annotation, model selection, and continuous improvement. For example, training utility NLU models involves gathering a diverse and representative dataset of utility-specific customer interactions, including phone calls, emails, and chat transcripts. Without training models with this specific ‘vocab’, self-service functionality will likely fail. The resulting frustration could put customers off from future use or lead to negative sentiment towards subsequent agent-led interactions. 

As with all implementations, it is also vital to continuously monitor the model's performance in real-world customer interactions. Using this feedback, future iterations have enhanced accuracy and the ability to handle a wider range of queries, especially as industry trends evolve.

For example, it is only in the last two years that ‘cost of living crisis’ has become a part of everyday conversation. This short phrase can encompass much sentiment and nuance - especially when customers are speaking in the context of utility services. It’s essential, therefore, that the NLU model is regularly updated and retrained to adapt to evolving customer needs, industry trends, and regulations.

With 71% of European utilities and telecoms providers having invested in AI for CX, NLU is the next obvious step for enhanced integration. However, with nearly four in ten European business executives identifying that training AI models takes too much time, they increasingly turn to expert partners to manage this continuous process.

Strategic NLU will redefine CX

Evolving AI for CX is not a mere tactic – it’s now a core strategy. In the next two years, over nine in ten European executives plan on investing more in these technologies, with the majority saying it will undoubtedly contribute to performance.

This focus on overarching strategy is vital. In a complex, high-volume CX environment, simply automating or adding NLU to singular processes and expecting a successful outcome is unlikely to deliver ROI and could even introduce pain points. CX leaders must strategically target real customer pain points, set clear objectives, define project ownership and standardise ongoing supervision as key to success. 

For example, excessive reliance on NLU-driven automation can erode the human touch in customer service, leaving customers feeling undervalued. As with all AI implementations, some issues may require human intervention, and automation should not replace the need for human support entirely. Combining AI with human oversight and intervention can help balance automation and personalised, effective customer support.