AI is increasingly used to automate communication, but in the context of employee engagement, empathy, nuance, and trust must remain central. As enterprises scale these tools across distributed and frontline teams, the ability to preserve human understanding within machine-generated insight will define success.
The idea of listening to employees is not new. For decades, organisations have deployed surveys, suggestion boxes, focus groups and intranet message boards in the hope of accessing a more honest view from the front line. However, few of these legacy methods were designed with action in mind, and most failed to scale or resonate beyond the corporate office.
What AI introduces is the potential to replicate one-to-one dialogue at an enterprise scale. This is not about replacing managers or flattening sentiment into dashboards. It is about delivering insight in a way that respects the complexity of the workplace, adapts to different audiences, and gives everyone, especially those furthest from head office. This voice is heard and understood.
“Employee engagement is fundamentally about giving your people a voice,” Lewis North, Chief Technology Officer at WorkBuzz, explains. “They are the ones driving your business forward, the ones facing customers every day, and the ones who know whether your strategy is working. However, most organisations lack the time or structure to listen to them properly. The aim of our AI is to change that, to replicate the experience of sitting down and having a meaningful conversation at scale.”
Data does not equal dialogue
The tools for collecting feedback are not the problem. Most large organisations are already overwhelmed with survey responses, suggestion forms, sentiment metrics, and open-text data. The challenge is turning this raw information into something coherent, balanced, and genuinely helpful.
“Every HR team we work with tells us the same thing, they have more data than they can act on,” North says. “People leaders are overwhelmed. They may gain access to a dashboard or report, but they are rarely provided with context or direction. On the other hand, HR teams are under immense pressure to deliver insights to the business but cannot possibly coach every single manager on what matters and why. That is the gap we are solving.”
Much of the noise comes from predictable flashpoints. Pay, for example, almost always receives low scores. But as North explains, low scoring does not necessarily mean high importance. “What drives engagement tends to be much more relational, whether employees feel heard, whether they are proud to work where they do, whether they feel they have a voice. The AI we have developed knows how to separate signal from noise and elevate what truly matters.”
Context, confidentiality and control
Designing a system that delivers relevant and trusted insight to every manager in an organisation means handling complexity, not removing it. One of the most significant design challenges has been adapting tone, language and output based on the audience while also respecting privacy and regulatory boundaries.
“We cannot treat everyone the same. An HR director will need a different level of detail and complexity compared to a line manager on the factory floor,” North explains. “We have built our AI to write different summaries depending on who is reading it. That means adjusting the structure, tone, focus and even vocabulary based on the role and context.”
Confidentiality is equally non-negotiable. “There are things we will never include in a summary, mentions of returning from maternity leave, for example, or data that could identify an individual based on demographics,” North continues. “Our evaluation frameworks screen everything, checking for names, unique phrases, and any detail that could compromise anonymity. And if something does slip through, a human still checks it before it goes live.”
This mix of automated and human control remains critical. “We have gradually improved the AI to the point where most content meets our internal standards automatically,” North says. “But the human-in-the-loop is still essential, not just for safety, but to ensure we maintain empathy, balance and integrity in the language. We do not want the system issuing directives or exaggerating urgency. It must reflect reality, not distort it.”
Empathy as infrastructure
One of the most innovative features North and his team are introducing is a ‘voice of employee’ narrative mode, which synthesises first-person accounts from anonymised open comments. “It presents the findings as if spoken by an imaginary employee,” he explains. “This is what we are feeling, this is what we care about. It is incredibly powerful when presented to leadership teams. You are no longer just looking at scores; you are hearing your organisation speak to you. And because it is built from thousands of real voices, it carries weight and emotional intelligence that a graph never can.”
Behind this emotional resonance lies robust infrastructure. “Everything we do runs securely within our AWS environment,” North explains. “Nothing leaves that cloud. We utilise Claude 3.5 from Anthropic as our language model, and we have strict guardrails in place to prevent any personal or identifiable information from leaving our systems. Even something as basic as regional access to LLMs has caused issues; in one case, we had to work with AWS to resolve a power supply constraint just to enable access to the model in the right region.”
These hidden details matter. The rush to integrate generative AI into workplace tools has left many IT teams struggling to assess the associated risks. “A lot of organisations want to use AI but are not ready to build it themselves,” North says. “Some have been pasting data into ChatGPT just to generate summaries, which is a huge data privacy issue. Our approach gives them a secure, compliant, enterprise-grade solution that integrates easily with their systems, but more importantly, gives them confidence.”
From dashboards to decisions
This ultimately leads to a new model of employee listening, one that transitions from generic survey analysis to actionable behavioural insights. “We are about to roll out behaviour analysis,” North explains. “We already ask employees about the behaviours they see in their leaders. When you correlate that with engagement scores, you get a very clear view of what good leadership looks like. You can even start to model which actions drive retention, performance or loyalty.”
The next phase of development is focused on making those behavioural insights predictive rather than reactive. By correlating feedback data with business metrics, such as productivity, absenteeism or even theft, organisations are beginning to quantify the real impact of line management. “We found in one study that if staff had not had a recent conversation with their manager, theft of stock was significantly higher. That is a direct causal link between engagement and cost.”
North believes these connections will become the foundation for a more adaptive workplace. “You cannot rely on a single annual survey,” he says. “You need systems that learn over time, that understand how different signals relate to each other, and that evolve as your organisation changes. That is what AI enables, but only if you design it with purpose.”
The architecture of trust
As enterprises continue to adopt AI across every function, the temptation is to standardise, optimise, and centralise. However, employee experience does not benefit from uniformity. It thrives on nuance, empathy and understanding. North believes that AI’s true value lies not in replacing the human element but in amplifying it.
“This is not about efficiency,” he concludes. “It is about trust. It is about giving every employee the sense that they are being heard, that their concerns matter and that their organisation is willing to act. That trust cannot be mass-produced. It must be earned. What AI enables when built correctly, is the ability to scale that trust, one conversation at a time.”
