As access to AI expands, employees also need the confidence and practical understanding to use it well. They need opportunities to see where it can help in their own work, where its limits lie, and how it fits within the organisation’s expectations.
That helps explain why some organisations are extracting more value from AI than others. Too many businesses are still treating it as something that can simply be dropped into the way they already operate.
“People often think about AI as just sprinkling a little bit of AI on top of an existing workflow,” explains Sami Ghoche, Vice President of Applied AI at Zendesk. While that approach may generate modest improvements, Ghoche argues that potentially greater gains are possible if businesses instead “reimagine the whole thing.”
Stop sprinkling AI on existing workflows
Independent research reflects that view. A 2026 McKinsey survey of 750 employees and leaders showed that leaders were 5.3 times more likely to report enterprise value from AI when workflows had been redesigned rather than left unchanged.
Customer service is one example. AI might initially help an agent by completing part of a response. It could then generate an entire response and find the relevant information automatically. The next step might see AI reason through an issue and respond directly to the customer.
But Ghoche argues that businesses can go further by asking why the customer needed support in the first place.
“Anything that makes its way into a customer support issue or ticket is actually less a complaint about support and more an operational failure somewhere else in the business,” he says. That creates an opportunity to use the data generated through customer service to identify and address the underlying problem. In fact, AI could potentially help organisations prevent some of the need for support altogether.
“A lot of those issues can actually go away at the source if you’re thinking about AI this way,” Ghoche says.
Build AI confidence through the work itself
Reimagining work requires people to understand what AI can realistically do in the context of their own jobs. That understanding often develops through hands-on use.
Ghoche points to software development as one area where adoption has progressed particularly quickly. Engineers are accustomed to experimenting with technical tools and working out how new capabilities can be applied to problems.
The situation can be very different when AI is handed to a customer service manager, operations team, or another business function. They may know the problem inside out without knowing what a general-purpose AI system could make possible.
Applying AI to real business problems can help employees recognise new applications and make better judgements about when to use it.
“You’re talking to a support manager, support admin, operations admin, and you’re saying, ‘Here are these general language models. They can do anything. So just go ahead,’” Ghoche says. The result is that projects can get stuck because people need help connecting the capabilities of the technology with the problems they’re trying to solve.
That can lead organisations to focus on obvious applications, such as using AI to answer FAQs, instead of identifying new ways of operating. Ghoche gives the example of analysing customer interactions to identify somebody who has dropped out of an onboarding journey, then proactively contacting them. It’s the kind of intervention that may not previously have been feasible to perform manually.
“There are all these other types of interactions and use cases where it’s not economically viable to staff up a person on it, but it’s now possible,” he says. “You just need somebody to think creatively about it.”
AI needs to do things, not just say things
There’s another dividing line between incremental and transformational AI: whether the technology can act.
An AI system can answer a question or retrieve information without necessarily being deeply connected to the rest of an organisation. But issuing a refund, changing an order, booking an appointment or completing another transaction requires access to the systems where the work happens. That introduces an organisational challenge as much as a technical one.
The business team often understands what it wants to achieve. IT, product and engineering teams, meanwhile, control many of the systems and integrations needed to make it happen.
Ghoche describes those technical teams as holding the “keys to the kingdom.” Without alignment between the two sides, AI can be limited to informational applications rather than being allowed to perform useful work.
“If it's based on the knowledge you need to actually do something and not just talk about it, you need a developer to give you the keys,” he says. This makes AI transformation a cross-functional undertaking. Ghoche suggests organisations should embed technical specialists within the teams pursuing particular business outcomes, rather than relying on the traditional model of handing requirements from one department to another.
The business experts bring the context. Technical specialists understand the organisation's systems and how AI can connect to them. Working closely gives business teams more opportunity to explore what the technology can do in practice.
The risk of doing nothing
Giving AI greater access to business systems inevitably raises questions around trust, governance, and security.
The risks also vary considerably according to the application. In software development, organisations need to consider whether AI-generated code introduces vulnerabilities. In customer service, they need to ensure systems don’t expose information to the wrong customer.
But Ghoche warns against assuming that the principal danger facing enterprises is AI running out of control. In his experience, organisations tend to be conservative. When teams lack confidence, alignment or access to the systems they need, the more common outcome is that the project doesn’t progress far enough to deliver the expected return.
“Usually, the issue when you talk about ROI not being realised is you’re not deploying these transformational tools on a large enough surface area,” he says.
What happens when organisations can’t overcome those barriers? Nothing happens.
Trust therefore needs to enable responsible action rather than become a reason for indefinite inaction. Clear guidance can help employees use AI with greater confidence. People need to understand where they can use it, what they remain accountable for, and when human judgement is required. As workflows change, they also need clarity about how their roles are evolving.
Treating AI programmes as broader digital transformation initiatives can help. Bringing IT and other technical stakeholders into a project from the outset means governance, access, and security requirements can be considered alongside the business outcome, rather than emerging as barriers later.
Making AI easier to use
If AI is going to become part of everyday work, the way people interact with it matters. Familiar interfaces can make the technology easier to use and reduce the effort required to learn new ways of working. Voice AI could become an important part of that shift.
Speech allows people to explain themselves naturally and provide significant amounts of context without having to formulate and type a detailed prompt. That makes voice potentially useful not only for customer-facing applications but anywhere that requires a conversational exchange with AI.
Voice technology itself isn’t new. What has changed, says Ghoche, is the quality of the interaction. Previously, speech could be transcribed and processed as text, but the latency involved in that pipeline made natural back-and-forth conversations difficult. Improvements in AI models are removing that constraint.
“The models are faster, better, more reliable,” he says. That’s opening up applications requiring genuine two-way interaction, with customer support currently one of the clearest opportunities.
From adoption to transformation
The organisations moving furthest with AI are rethinking how work gets done across the business.
McKinsey’s research tracks a progression from using AI for individual tasks to improving workflows and ultimately reinventing work around the technology. Only 11% of leaders surveyed said their organisations had reached that final stage, but 48% of those reported realising enterprise value, compared with 13% at the first stage.
The organisations getting the most from AI are focused on the business outcome they want to achieve and prepared to rethink the workflow from scratch. Crucially, they also bring together the domain experts who understand the problem and the technical stakeholders capable of unlocking the systems AI needs.
“The area experts have the context; they just don’t have the keys to get the stuff done,” says Ghoche. “If you can align your organisations well in this way, that’s necessary. That’s always the ones that do the best job.”
As access to powerful AI becomes commonplace, advantage will depend more heavily on how organisations put it to work. The bigger opportunity is to reconsider what outcome the organisation is trying to achieve, what's now technically possible, and whether a process designed around human constraints still makes sense when some of those constraints no longer apply.
For business leaders, the next stage of enterprise AI is about asking whether there is now a better way to design the work from the outset.
Discover how Zendesk can help your organisation put AI to work.
The post Making enterprise AI work means rethinking the work itself appeared first on Raconteur.