77% of leaders say AI spending is stuck in incremental mode

By Marcus Bennett

Seventy-seven percent of business decision-makers say their organizations devote too much AI investment to incremental improvements, according to a survey commissioned by Microsoft-focused IT advisory firm Avanade.

The Avanade survey covered C-suite leaders, directors and managers. Only 11% expect AI to fundamentally change how their organizations operate, create value or compete in the next year.

Nearly two-thirds say AI has not yet delivered returns that justify the investment and effort. Three-quarters believe their organizations have overstated their AI maturity.

Four years into the new AI era, Ben Beath, Avanade’s chief AI officer, put the gap plainly: “Confidence is high, but confidence isn’t the same as change.” Copilot licenses and a few dozen use cases do not signal maturity if work remains unchanged. For Beath, weak returns indicate low maturity and a misguided strategy.

According to CIO.com’s State of the CIO survey, 59% of IT leaders report that some or most AI initiatives have met or exceeded ROI goals. ROI, or return on investment, measures whether expected gains justify the cost.

What separates incremental AI from transformation?

Many IT leaders chase tools that summarize email or generate basic reports rather than reinvent the business. Beath says these tools can improve employee productivity, but they will not transform the organization.

Incremental projects remain easier to approve and measure, and they carry less risk for the executive sponsor. Leaders know the difference, even when they choose the safer project.

Adding AI to broken or legacy work merely produces a faster legacy process, Beath said. He described the alternative: “What would this process in our business look like if it was reimagined with AI?” In these cases, AI sits in every step rather than on top of an existing workflow.

The leading CIOs Beath cites are reimagining business processes. AI becomes integral to the redesigned operation instead of a separate productivity layer.

Who must own the strategy?

Beath argues that most organizations face a leadership crisis. He calls it a leadership conversation, not an IT-only one. CIOs pursue AI projects without a broader leadership coalition, leaving strategy inside IT.

If AI strategy sits only with the CIO, the focus stays on efficiency. Beath says shared ownership with the CEO or CFO and business leaders opens a path to transformation.

A CIO should ask what business the company would build today from scratch with AI, not where AI might fit. Beath rejects two easy explanations: models need not be bad, and pilots need not be poorly selected. The hard work sits where data meets applications, technology meets process, and technology meets the business. He argues that moving beyond siloed efficiency can improve both AI maturity and business results.

Can incremental and transformational AI work together?

Ed Keisling, chief AI officer at Progress Software, says the two modes go hand in hand. Incremental projects help teams explore what AI can do and can compound across a full workflow.

• Leadership updates: Progress Software uses AI agents to synthesize updates from workplace collaboration, project management and knowledge management systems for leaders.
• Sales quotes: Sales agents combine data from the company’s customer relationship management system, data platforms and customer systems to generate quotes faster and more accurately.
• Time-sensitive work: The company is deploying AI for requests for information and proposals. Processes that took days now take minutes, improving responsiveness to opportunities.

Data sovereignty, meaning control over where data resides and who can access it, plus security and cost, hold organizations back from more transformational uses. Keisling calls scaling from an individual to a team, then to teams of teams or the organization, a major challenge.

Organizations must balance the risk of action against the risk of inaction, while weighing market opportunities and threats. AI ROI remains elusive, so IT leaders must focus on budgets and results.

Yet Keisling says AI creates economic value without needing a transformational label when it removes meaningful time from a high-volume workflow, increases throughput, improves quality or frees capacity. Throughput means the volume of work completed.

What blocks AI from scaling?

Digital adoption software provider WalkMe began with incremental uses and continues to evolve, according to Tal Carmi, its CIO. He calls incremental improvements steppingstones: Copilot solutions evolved into autopilot workflows after the company reviewed their success rates, changing how work gets done.

WalkMe has reached transformation in some areas but not others. Carmi identifies systems limitations, face-to-face processes, risk ownership and employee enablement as constraints.

AI magnifies existing knowledge and data problems, Carmi says. Asking a large language model, or LLM, to extract data and build a dashboard is easy; getting the right data and processing it correctly is difficult.

Many legacy systems people want to automate lack an application programming interface, or API, and may not gain one soon. Workers then improvise, creating shadow IT: software and workarounds that bypass approved systems and controls.

What should product teams do on Monday?

Use two approval paths, with different evidence requirements.

• Incremental work: Track meaningful time removed from high-volume workflows, throughput, quality and freed capacity. Keisling’s value test matters more than whether the project earns a transformation label.
• Transformational work: Redesign the process with AI in every step. Give the CIO shared ownership with business leaders, then resolve data access, application connections, security, cost, risk ownership and employee readiness before scale.

Enterprise AI’s next advantage will come less from adding tools than from redesigning how work crosses data, applications and business ownership. Measured increments can serve as stepping stones to that shift, but only when leaders share the risk.