Key Takeaways
- Less than a quarter of pilots scale
- What blocks scaling is the definition of success
- Defined by speed, cost, quality, and risk
- One thing to check this week
A report released this week by Infosys surveyed more than 1,000 senior executives across various industries. While most respondents cited AI as a key driver of new revenue opportunities, 72% said less than a quarter of their AI pilots had successfully scaled. Two-thirds said they struggled to measure ROI well enough to substantiate the benefits of AI that leadership touts.
The survey also revealed pressure for short-term results. Nearly three-quarters of respondents said demands to demonstrate short-term ROI make it harder to experiment with more transformative, long-term AI initiatives, and only about half said they had a balanced set of KPIs to evaluate AI's value. The report noted that this approach pushes organizations to rush results based on partial, inconsistent, or misleading signals.
Anant Adya, EVP and Global Head of Cloud, Infrastructure and Security Services at Infosys, said in the report that team alignment and how success is defined in business terms—such as speed, cost, quality, and risk—have a greater impact on outcomes and are essential to building a successful roadmap.
Spending itself hasn't slowed. Gartner data released in July projected that global end-user spending on AI models and platforms would grow 63% year-over-year to reach $64 billion by 2026. In a recent Deloitte survey, nearly half of respondents said they were running more than 30 AI pilots, many of which they continued even knowing they wouldn't reach completion.
Bali D.R., EVP and Global Head of AI and Automation at Infosys, said companies continue AI investments based on signals that demonstrate the technology's potential, but noted that translating this into financial performance requires clearer business cases, success metrics, and consistent leadership strategy. Infosys CTO Rafee Tarafdar said exponential gains come when AI capabilities are productized on an enterprise-wide platform and scaled and opened up across the organization.
Implications for Korean Enterprise AX
What this survey identified as the barrier to scaling wasn't model performance—it was how success was defined and what metrics were used to measure it. If your organization is considering AX transformation, there's something to decide before selecting tools: write down, before starting, which task's which number—processing time, unit cost, defect rate, or rework count—you intend to move, and by how much. The four categories Adya mentioned—speed, cost, quality, and risk—are the boxes you can use to select that number.
Running more pilots also carries a cost. If you launch multiple pilots assuming most won't reach completion, the bigger cost isn't the failures themselves but lacking grounds to judge which to cut and which to scale. Writing down termination criteria alongside success criteria at the outset means that judgment gets settled by data, not by meetings.
There's one thing to check this week. Count the AI pilots currently running in your organization and verify whether each has documented success criteria and termination criteria. The proportion of pilots without such documentation is the proportion of spending you won't be able to explain at your next budget meeting.