Artificial intelligence agents – autonomous digital entities that analyse data, automate decisions, and act independently – are set to drive a new wave of business transformation.
Analyst group Gartner predicts that up to a third of enterprise applications will use agentic AI by 2028.
Yet, behind the hype lies a sobering reality: most corporate AI projects don’t deliver the promised returns, and many never make it out of the pilot phase. According to a recent study by MIT researchers, up to 95% of AI initiatives fail to produce meaningful ROI, leaving executives wondering if agentic AI is worth the risk.
Spark, one of the country’s biggest digital service providers, sees a pathway to real results if firms take the right approach.
Why AI projects fail
There’s a phrase that’s common in the tech industry: “the graveyard of AI pilots”.
“If you start with technology as the answer and work backwards, that’s really a recipe for missing the brief,” says Liz Urquhart, Spark’s General Manager of Business Products and Services.
“Garbage in, garbage out.”
Urquhart has witnessed the landscape shift from isolated pilots to agentic AI deployments impacting everything from customer service to cybersecurity. Based on Spark’s experience internally and with enterprise clients deploying AI agents, she says successful AI projects are the product of rigorous foundations, clear strategy, and smart change management.
Spark’s AI Accelerator is designed to incubate its clients’ AI projects to give them the best chance of success.
“That’s where we get them to really distil the problem they’re trying to solve, or identify the opportunity they want to pursue,” Urquhart explains. “What might success look like? It’s not just technological success. Are you trying to improve customers’ experience? Productivity output? Employee experience?”
Practical AI in action
For Urquhart, the key is to “start small, iterate and have the checks and balances in place, so that the transformation feels more manageable”.
Spark’s journey with agentic AI began in its contact centres, freeing staff from repetitive tasks to focus on helping customers.
“In our contact centres we’ve seen what I think of as a triumvirate of outcomes,” Urquhart says. “One is productivity gains as our contact centres resolve issues faster. But equally, customer satisfaction scores have improved because calls are resolved faster.”
A third, often overlooked outcome, is the increased job satisfaction employees experience as they are freed from mundane and repetitive tasks.
Spark also harnesses AI agents in back office, technical, and security domains. It is early days for the technology, which requires a careful approach to governance and human oversight. But with AI agent technology evolving rapidly, it’s likely to serve as a significant productivity booster in most industries – if businesses pursue the right approach to its development and use.
The five pillars of Spark’s approach for AI success
1. Data excellence
“AI will learn based on what it’s fed, so getting those data foundations right, both the construct of the foundations as well as the quality of the data, is all-important,” says Urquhart.
That doesn’t mean you need to wait until you have the perfect data.
“If you’ve got a specific opportunity or problem in mind, work through what are the critical data that you need to have cleaned up. Start with that and make sure you’ve got that right.”
2. Technical infrastructure
Modern platforms like Salesforce, SAP and ServiceNow have agentic AI built in. But don’t expect to just press a button and put the AI agents to work straight away. Most organisations have run on a mix of legacy systems and applications and cloud platforms. How will AI agents navigate this landscape to get the information required to do the task accurately and securely?
“You need to choose a cloud model that can scale and is cost-effective to address those massive data sets that are going to grow over time as you deploy these solutions,” says Urquhart.
3. Governance and oversight
Nor are AI agents ‘set and forget’ solutions. Nothing stands still, says Urquhart.
“You will build an agent, and even over six to twelve months, there’ll be a proliferation of other sources of data.”
It must be governed and measured. There must be proper guardrails, human oversight, and benefit tracking, she says.
“It doesn’t need to be hundreds of people looking after the stuff all the time, but it’s not nothing. It’s a continued investment.”
4. Change management
If you don’t get buy-in from the people who are going to be working with AI agents, the chances of the project succeeding are greatly reduced. This has long been the death knell for tech projects, with staff sticking to what they know and are comfortable with.
The potential for AI agents to enable employees to do more meaningful work is great. But that requires the technology being seen as an opportunity rather than a threat.
5. Skills and capability development
Crucial to building that trust and acceptance as AI agents change how aspects of an organisation operate is upskilling the workforce.
“We actually have to teach people how to be AI engineers, to oversee the AI agents and do testing and quality control. It’s a different sort of approach now,” Urquhart says.
Spark has invested in AI skills programmes across the organisation, giving every employee the ability to learn to use AI tools and platforms that can augment or transform how they work.
Avoid the pitfalls
Spark has developed a roadmap, based on those five pillars, for navigating the risks and reaping the rewards of agentic AI.
By paying due attention to these key priorities, companies can avoid the graveyard of pilots and turn potential into performance.
Learn more about Sparks Data and AI solutions.
