The next tech directive: Service excellence

Moving CIOs from infrastructure maintenance to proactive, experience-led service

As CIOs shift from managing infrastructure to shaping experiences, service excellence is becoming a strategic priority.

In this Food for Thought discussion, in partnership with Zendesk, senior IT leaders explored how AI-powered service can help CIOs improve productivity while navigating the challenges of governance, organisational readiness and the changing relationship between humans and machines.

Keeping the lights on was once the defining responsibility of the technology leader. Today, however, the CIO’s remit extends considerably further, encompassing employee activity, customer experience, organisational transformation and, increasingly, the responsible deployment of AI.

But as AI capabilities accelerate, are organisations genuinely prepared to translate technological potential into meaningful service excellence? This question underpinned a recent HotTopics Food for Thought, held in partnership with Zendesk.

Moderated by HotTopics Editorial and Strategy Director, Doug Drinkwater, the discussion brought together senior IT leaders to examine the evolving CIO mandate, the practicalities of agentic AI adoption and the organisational foundations to deliver experience-led service.

Setting the scene, Doug referenced Zendesk research revealing that technology leaders now spend 72 percent of their time on non-traditional responsibilities including strategic planning and cross-functional initiatives spanning HR, finance and legal.

With expectations expanding and resources remaining finite, the conversation quickly established that service excellence requires considerably more than introducing another intelligent tool.

From keeping the lights on to expanding business value

The opening discussion explored a familiar leadership dilemma: how can CIOs balance business-as-usual operations with the growing expectation to deliver transformation?

One perspective advocated fundamentally reversing the traditional operating model. Rather than beginning with existing resources, defining tasks and subsequently identifying opportunities for automation, organisations should first consider their desired outcomes and where AI might contribute. Human involvement, governance and resource allocation could then be designed around those decisions.

Other participants challenged the practicality of this approach, particularly for organisations still managing complex legacy environments. For these leaders, transformation begins with simplification, reducing unnecessary architectural complexity, understanding the proportion of resources dedicated to running, growing and transforming the business, and gradually redirecting capacity towards value-generating activities.

The distinction was this: while AI presents opportunities to redesign operations, introducing it into an inefficient environment does not automatically resolve underlying organisational problems.

Customer expectations also need to inform those decisions. One attendee described discovering, through customer journey mapping, that their customers actively preferred speaking to a person when experiencing service difficulties, rather than interacting with a chatbot.

In this instance, the more appropriate application of AI was potentially behind the scenes, improving operational processes without compromising the human relationships that customers valued.

The broader question, therefore, was not simply where AI could be deployed, but where it could improve the experience being delivered.

The bionic organisation: Finding the human-machine balance

For Sarah Al-Hussaini, Vice President AI at Zendesk, the emergence of agentic AI introduces a shift in how organisations should think about work.

Where traditional automation largely followed predefined instructions, increasingly autonomous AI systems can take on more complex activities, changing the relationship between human employees and technology.

“Work in general, in every industry, has now split into human work and machine work,” Sarah explained. However, identifying the appropriate division is not necessarily something organisations can accomplish entirely before implementation. Sarah suggested that experimentation and adoption are essential to understanding where machines can contribute and where human capabilities are most valuable.

This prompted discussion around the emergence of the bionic organisation: an operating model in which repetitive, rules-based activities are increasingly handled by machines, allowing employees to concentrate on judgement, critical thinking, communication and relationship-building.

Nevertheless, several participants cautioned against treating human interaction as an inefficiency to be eliminated. As AI becomes more widely available, the quality of personal relationships could become an increasingly important differentiator, particularly in service environments where empathy, reassurance and trust influence customer satisfaction.

AI readiness: Building the foundations before scaling

If the ambition for AI-powered service excellence was broadly understood, the practicalities of implementation generated more debate.

Several technology leaders around the table described a familiar challenge: employees are enthusiastically experimenting with AI, yet moving successful proofs of concept into production remains difficult.

Security requirements, fragmented data, inadequate governance and limited internal resources can prevent promising initiatives from progressing beyond experimentation.

Sarah distinguished between two forms of organisational pressure: pressure to deliver when a clear implementation pathway exists, and pressure to adopt AI when the business is not sufficiently prepared.

“The pressure gets really uncomfortable when you need to adopt AI, but you're not ready for it,” she explained.

From Zendesk’s perspective, readiness begins with knowledge accessibility. AI systems become more useful when they can securely access relevant, accurate and regularly maintained information.

For service organisations, this might include internal knowledge bases, help centres, product catalogues and customer information. However, access must be accompanied by appropriate controls governing what an agent can retrieve, process or modify.

The conversation also examined the importance of MCPs, integration frameworks, auditability and clearly defined permissions.

One attendee described the challenge of enabling business teams to develop their own AI solutions without allowing unverified applications into production. Another participant highlighted the risk of new AI functionality exposing information that existing data-access arrangements were never designed to accommodate.

These examples reinforced a recurring theme: organisations cannot separate AI transformation from data modernisation and operational resilience.

Innovation needs a safe place to play

An interesting analogy emerged around the responsibility of central IT functions to create environments resembling children’s soft-play areas: spaces that encourage exploration and experimentation while incorporating safety into their design.

For CIOs, the equivalent involves establishing clear governance boundaries within which employees can innovate without requiring technology teams to approve every individual automation. Yet those boundaries must continuously evolve.

As AI becomes embedded across enterprise applications, organisations face a growing visibility challenge. Organisations may no longer be introducing every AI capability themselves, making it harder to maintain oversight of emerging risks.

The discussion highlighted the importance of combining employee empowerment with monitoring, education and controls that adapt alongside technological change.

Build, buy or automate? The economics of AI

Beyond governance, the speakers questioned whether organisations are always choosing the most commercially appropriate technology.

The traditional build vs buy debate has acquired new significance in the agentic AI era, particularly as businesses weigh internal development against increasingly sophisticated SaaS platforms.

Building proprietary solutions can provide greater control over architecture, sensitive information, and specific business requirements. However, it also introduced responsibility for maintaining the wider infrastructure necessary for secure deployment, including data minimisation, anonymisation, monitoring and integration.

Purchasing established solutions may reduce some of that complexity, although suitability remains dependent on organisational requirements and risk appetite. Sarah also highlighted an emerging commercial consideration: outcome-based pricing. Rather than charging exclusively for individual AI interactions or model calls, some service providers are increasingly linking costs to completed resolutions.

Within customer service, this means considering whether an AI agent genuinely resolved an enquiry, whether an escalation was required, and whether the customer was satisfied with the way this was handled. Value-based pricing models like this have become increasingly popular, eliminating the cost variability risk of LLMs and offering customers a guaranteed ROI on a unit of outcome.

The discussion raised another important distinction: not every automation opportunity requires generative or agentic AI. One participant observed that AI can be valuable for discovering and demonstrating use cases, even when the eventual production solution would be more efficiently delivered through deterministic automation.

Without this consideration, organisations risk accumulating a new generation of technical debt, alongside unnecessary token consumption and operating costs. The commercial imperative here is to select technology according to the problem, rather than redesign every problem around AI.

Who is setting the pace of AI transformation?

As the conversation turned towards the relationship between technology buyers and vendors, the participants questioned whether software providers were accelerating innovation faster than enterprise customers could really absorb it.

The distinction between the agentic enterprise being promoted across the technology market and the relatively cautious experimentation occurring within many organisations was apparent.

Sarah offered an alternative interpretation of this dynamic. From Zendesk’s perspective, the pressure to innovate is increasingly originating within the customers themselves, particularly early adopters who are already experimenting with AI across their personal and professional lives.

“We’re seeing a pull,” she explained, describing customers approaching providers with more sophisticated expectations of what should be technically possible. Nevertheless, participants agreed that different industries and organisations operate at different levels of maturity.

From individual productivity to enterprise-wide impact

Attention shifted to the workplace and the growing disparity between enthusiastic AI adopters and employees experiencing transformation fatigue.

For some, digital literacy remains a substantial barrier while certain employees are actively requesting access to sophisticated AI tools, others are still adapting to relatively fundamental changes in their working environment.

This creates a difficult balancing act for CIOs wanting to accelerate adoption without alienating experienced employees or overlooking valuable institutional knowledge.

One important distinction emerged between personal productivity and enterprise productivity. An employee saving time through AI does not necessarily translate into measurable organisational value.

Participants also discussed the importance of distributing responsibility for adoption across business functions, including adopting data and automation champions based on enthusiasm and willingness to drive change, rather than hierarchy alone.

Can you trust AI Agents? Join Zendesks' digital event on building trustworthy AI agents: