Published on the 02/07/2026 | Written by Heather Wright

Highest value AI projects may not be the first you should fund…

Enterprise AI enthusiasm is creating a new problem for tech leaders: Too many opportunities.

One Gartner client began with more than 600 potential AI use cases competing for investment. Unsurprisingly, Gartner’s Luke Ellery recommends companies don’t pursue everything. Instead, he recommends creating a ‘AI funding funnel’ that ranks initiatives based on business value and technical feasibility before significant budgets are committed.

“It’s a great way of helping CFOs understand AI investments.”

For CIOs facing pressure to ‘do more with AI’ the message at Gartner’s recent Data and Analytics conference in Sydney, the message was surprisingly simple: Stop launching pilots and start managing AI investments like a portfolio.

Ellery, Sydney-based Gartner VP analyst, says while organisations have been spending billions on AI, ‘you could say there is no R in ROI, only investment’. A Gartner survey showed 80-95 percent of Gartner clients saw limited financial returns from their AI investments and only 11 percent of CFOs were able to concretely measure financial returns from big AI investments.

Ellery’s assessment? Based on other Gartner surveys, he says the number one thing is picking the wrong use cases. Deploying Microsoft Copilot might help employees save time and feel happier and more productive, but that’s not financial value.

Value vs feasibility

Gartner’s funding funnel requires ideas to be vetted and prioritised across two dimensions – business value and feasibility.

Business value focuses on the outcomes executives care about – higher revenue, lower costs, reduced risk or improved service delivery. It’s an area that will vary for each organisation, Ellery notes. Time horizon is also an important factor here, he says. While last year CFOs were very focused on getting AI returns within the year, that stance has softened somewhat with the realisation achieving AI goals, and returns, will be harder than anticipated and returns will come on a longer time frame.

Feasibility, meanwhile, looks at the practical realities of implementation, including data readiness, technology requirements, skills availability, security controls and governance obligations.

The sweet spot, of course, is funding initiatives that score highly on both measures and will provide easy wins.

But Ellery notes: “Often organisations won’t have many.” All, however, is not lost. Ellery says leaders need to think about how projects work together, evaluating AI initiatives as a portfolio, rather than in isolation.

“In some scenarios it is very hard to get any of the individual projects off the ground by themselves, but if we collect a number of projects together, we can combine the cost and value and risks.

“It’s a great way of helping CFOs understand investments so we can actually create those larger returns.”

Why boring projects matter

He used customer personalisation as an example of an initiative with potentially high business value. The ability to tailor products, offers and interactions to individual customers has long been viewed as an AI holy grail, with obvious implications for revenue growth and customer retention.

The challenge, however, is that many organisations lack the foundations required to make personalisation work. Customer data may be fragmented across multiple systems. Governance controls may be immature. Data quality may be poor. Technology platforms may not be integrated.

As a result, what appears to be a high-value opportunity often scores poorly on feasibility.

By contrast, a much less glamorous project – such as automated client call summarisation – may deliver only modest business value on its own. It is unlikely to transform an organisation’s growth prospects or dramatically improve profitability, yet it may be significantly easier to implement.

More importantly, the project may capture and structure customer information essential for more advanced personalisation initiatives later.

Seen that way, the lower-value project becomes a stepping stone to something much bigger.

“We need to look at all these initiatives as a group in a broader view,” Ellery says. “The problem is that a lack of consistency prohibits us from actually achieving differentiation,” he says, arguing that organisations often need to fix underlying operational and data problems before they can use AI to create competitive advantage.

Building foundations before differentiation

Many AI projects fail not because the models are inadequate, but because organisations do not have accessible, well-governed data available for those models to use.

Technical debt presents a similar challenge. Legacy systems often store valuable information in formats that are difficult to access, integrate or analyse. Organisations may also need to invest in security, governance and compliance frameworks before deploying AI more broadly.

These investments can be difficult to sell with boards and executive teams naturally attracted to customer-facing use cases and ambitious transformation projects. Few get excited about data architecture, governance frameworks or system modernisation. Yet those investments can determine whether more ambitious AI initiatives succeed or stall.

Ellery pointed to the banking sector to illustrate that, noting years of core banking modernisation efforts as an example of organisations improving consistency across technology environments.

“That’s probably why some of the banks are more ahead in the AI race than others, in terms of realising the value of AI.”

In some cases capability-building projects lacking immediate business impact may be the highest-value investments an organisation can make.

Making the financial case

After vetting and prioritising – providing the litmus test for funding – Gartner’s funding funnel focuses on the crux of the issue – the funding. He noted there are different types of value for organisations. ‘Green money’ provides direct, measurable financial outcomes that appear in profit and loss statements, such as increased revenue or reduced budgeted costs. These are the benefits CFOs tend to care most about because the cause-and-effect relationship is clear.

‘Blue money’ by contrast, includes measures such as customer satisfaction, employee experience and Net Promoter Score. While these metrics may contribute to future financial performance, they don’t guarantee a return and can be difficult to directly link to revenue or profit with less direct correlation.

Ellery also warned against relying heavily on ‘cost avoidance’ arguments, such as delaying a system upgrade, noting that financial leaders often place less weight on savings that have never formally appeared in a budget.

“What we’re doing is putting all this information together on a slide so we can make a determination with the CFO.”

When cumulative costs are higher than the benefit – as with the earlier customer personalisation example – there’s a chance to flip the conversation and argue it as a cost leader to enable you to do other projects – or to bring it into a wider program.

Ellery noted the success of a multinational DIY and home improvement goods company. “They used a system dynamics model and did scenario simulations over three and five years to actually figure out how to order, how to do promotions with feedback loops.”

Targeting 10 percent revenue growth, they nearly got it, coming in at nine percent.

“But they also stabilised their supply chain, which unleashed about $120 million in working capital, and they also deferred the expansion of their processing facilities, which saved them about $25 million in capital costs.”

A courier company increased on-time deliveries by 25 percent, cut idle time by 20 percent and reduced fleet expansion to three percent through use of ‘a very interesting agent-based model, where they had agents represented for vehicles, drivers, customers as autonomous agents’ and micro-hubs.

And a hospital, aiming to reduce emergency department wait times by 15 percent in three, months used AI in its demand forecasting and rostering. Their wait times dropped 77 percent and – as a side benefit – they also saved $100,000 in overtime through data-driven staffing.

“If we can target the value that we’re trying to optimise, it makes it so much easier for us to build these AI systems to actually realise those outcomes and to optimise them.”