Businesses shift from AI pilots to trust & results

Senior executives from five technology companies have outlined what they see as the next phase of artificial intelligence in business use. Their comments point to a shift away from experimentation and toward trust, integration, and measurable results.

Across the responses, a common theme emerges: companies are moving beyond early enthusiasm for generative AI tools and asking whether the technology can be trusted in core operations. Several executives said the debate is no longer about simple adoption, but whether AI can reduce workloads without creating new risks or adding more checking.

Dharam Gurbani, Chief Growth Officer at Ascendion, said organisations are now testing AI against practical business demands rather than broad strategic claims.

"The real test of AI, in the UK and beyond, isn't ambition, pilots or strategy. It's whether AI makes critical work more reliable in the institutions people depend on every day. Across the UK enterprises we work with, leaders have stopped asking whether AI can be adopted. They are asking a harder question: where does it improve productivity, ease operational drag, strengthen compliance and improve customer experience, without adding more pressure on already stretched systems? The answer is that AI has to prove itself inside the operating fabric of the enterprise, not just bolted on. In practice, that means modernising outdated systems, clearing bottlenecks, improving decision support, making compliance easier to evidence and giving teams the confidence to do higher-value work. The most credible AI programmes keep accountability with people. They help teams move faster while governance and judgement stay firmly in place. That is the distinction that matters. AI earns trust through measurable outcomes in the institutions people rely on, which means it must be production-grade. That is the AI worth appreciating," Gurbani said.

That emphasis on trust also appeared in comments from Foxit, which highlighted how much time workers still spend checking AI-generated output. Its research found that while a large majority of executives see productivity gains from AI, those gains can be eroded by verification work.

"Of course, AI Appreciation Day is an opportunity to recognise how quickly AI has evolved. But let's also take the time to ask whether organisations actually trust the work it produces. Our latest research found that while 89% of executives believe AI is improving productivity, those gains are often being undermined by the time spent verifying AI-generated outputs. In fact, executives save an average of 4.6 hours a week using AI, but spend almost as much time checking its work. Without confidence in the output, productivity gains quickly disappear. The burden is now on technology companies to help close this Trust Gap. The conversation now needs to move beyond AI adoption and towards AI confidence. For the past few years, organisations have focused on where they can deploy AI. The bigger challenge is knowing where they can rely on it. Success won't be measured by the number of AI tools a business adopts, but by whether those tools help people make better decisions, reduce rework, and enable work to move faster with confidence. As AI agents become more capable and take on increasingly complex tasks, trust will become the defining competitive advantage. That starts with the quality of the information AI is given, alongside the governance and verification processes that support it. Organisations that get these foundations right will give employees the confidence to spend less time checking AI and more time acting on its insights," said Evan Reiss, Senior Vice President of Marketing and Innovation at Foxit.

Core processes

Derek Thompson of Workato argued that many deployments remain limited to basic tasks such as summarising and editing, leaving broader business value unrealised. He said greater impact will depend on whether companies allow AI systems to operate within core business processes and measure them against defined metrics.

"Despite the rapid rise of AI adoption, the majority of use cases to date have been fringe experiments; straightforward tasks such as summarising research or rewriting emails. Many businesses are cautious, or unsure, in how best to deploy AI for work which is more complex and time-consuming. Yet by restricting agents to surface-level tasks, we are also limiting the technology to surface-level business impact. Deeper integrations and therefore deeper impact require the CIO to trust AI with the business' core processes. For this, KPIs should be the North Star for every agentic application. By aligning the use of AI with the metrics that are most important to the business, the CIO can ensure that all agents are producing results that have tangible impact. The success of agentic AI will be determined not by the sophistication of individual agents, but by how effectively organisations integrate them into the fabric of the business," said Derek Thompson, Senior Vice President and GM, EMEA, at Workato.

OutSystems took a similar view, but focused on narrower use cases where AI can already support staff in routine or data-heavy work. Luis Blando said companies are finding more value in systems that assist human decision-making than in fully autonomous tools.

"Despite the hype around fully autonomous systems, most enterprises today are using AI in far more practical ways, and that's where the real value is showing up. The strongest use cases cluster around three areas: processing documents that would otherwise require human review, handling high-volume transactional work, such as mapping incoming orders, and supporting decision-making by making sense of unstructured data. In these scenarios, AI excels at summarising complexity and offering recommendations, but not at making final calls - unless organisations are willing to accept mistakes. Used poorly, AI behaves like a team of interns: fast and prolific, but still requiring oversight and double-checking. Used well, it becomes a force multiplier for simpler applications, especially when fuelled with the right data and guardrails. Trust doesn't come from autonomy alone. It comes from knowing when AI should assist, when humans should decide, and how the two work together," said Luis Blando, Chief Product & Technology Officer at OutSystems.

Model choices

Malte Ubl of Vercel addressed another issue shaping enterprise AI use: cost. He said teams are becoming more selective in their model choices, using lower-cost systems for higher-volume work and reserving more advanced models for tasks where accuracy and reliability matter most.

"Days like today usually surface one of two conversations: how fast the models are improving or what could go wrong. What we see in the data is more practical. Teams are getting deliberate about where they spend their AI budget, and where they don't. Traffic through the Vercel AI Gateway keeps climbing, even as costs rise. But how teams spend is changing. Cheaper models handle high-volume work; the most capable models are reserved for tasks where quality, reliability, and accuracy actually matter. No single model wins every job, so the work is routing each task to the right one. The teams getting the most out of AI are the ones who know which model fits which task, and balance quality, speed, and cost accordingly," said Malte Ubl, Chief Technology Officer at Vercel.

Taken together, the comments suggest a more restrained phase of enterprise AI adoption is taking shape. Rather than asking how widely AI can be rolled out, companies are increasingly asking where it can be trusted, how it should be supervised, and whether it delivers results that justify the cost and effort.