Welcome to the August 2026 edition of This Month in AI.
Most large companies have something AI-powered running in production now, like a support agent or a forecasting model. As an industry, AI has moved from experimentation to a focus on getting initiatives into production, and now ROI. Line leaders and Finance want to know what an AI solution returns.
This month, five important pieces of research tackle the ROI question from different angles.
AI’s next opportunity may be hiding in the middle office
Often, the most valuable AI opportunities live behind the scenes inside a company.
Harvard Business Review’s 4 Steps to Transform the “Middle Office” with AI builds on a study by economists Anders Humlum and Emilie Vestergaard, who set out to check whether AI tools changed what workers actually earned. They had two kinds of evidence that didn’t agree. Workers surveyed about chatbots said the tools saved them time. But Danish government payroll records, which show what every worker was paid and how many hours they were logged for, showed no change at all. Comparing workers who got AI tools against similar workers who didn’t, two years after ChatGPT launched, the researchers found no measurable difference in earnings or hours.
The authors trace this to where enterprises have aimed AI.
Attention has gone to the front office, where results are most visible. In the middle office, operations have been largely untouched. That is unfortunate because middle office functions like contract review, compliance, risk management, and accounts payable respond well to AI solutions. In each case, AI can be used to automate routine tasks, while complex cases and exceptions can be escalated for human review.
That distinction matters.
AI is often most valuable in work that is hardest to standardize, where people have to interpret information, weigh exceptions, and make a call.
The article also argues that measuring time savings alone is not enough. Organizations need to measure whether AI reduces escalations, then determine where the recovered capacity creates greater value.
At RapidCanvas, we’ve found that clients often gain the most value from using AI for the processes that sit between core systems and frontline operations.
Production is no longer proof of success
A year ago, most companies were struggling to get AI solutions into production. While that situation has somewhat improved, many companies are now struggling to demonstrate ROI from the solutions they have deployed. Forbes’ Most Enterprise AI Is Live. Half Of Companies Can’t Prove It Works reports that 74% of the world’s largest enterprises now run at least one AI solution in production, while 93% are either piloting or further along. Yet half of the production-stage companies surveyed cannot consistently demonstrate whether their AI investments are delivering ROI.
That gap between deployment and proof is now a central problem in enterprise AI.
Part of it starts before an AI system goes live. Forbes points to the absence of clear baselines in many deployments, which makes it impossible to establish what changed after AI arrived. As the article puts it, a use case that goes live without a pre-deployment metric is unmeasurable forever. There is nothing left to compare against.
Data foundations remain the top blocker to scale at 71%, and they stay elevated at every stage of adoption rather than fading as companies mature.
Ownership is shifting too. Only 21% of AI ownership in the survey sits with a CAIO, AI lead, or center of excellence, while 37% sits with functional or line-of-business leaders. That makes sense. AI creates value inside business processes, so accountability increasingly has to sit with the people responsible for those outcomes.
AI programs now need a measurement discipline as rigorous as the technology behind them.
Scaling agents requires changing how people work
Agentic AI raises the stakes because deployment is only the first step to creating value from these tools.
McKinsey’s How to close the agentic adoption gap argues that organizations need to treat agentic AI as a reinvention of work, with change leadership embedded directly into how employees operate. Its survey of thousands of global business leaders identifies change management and siloed ways of working as major obstacles to scaling AI, outranking concerns about adequate technology infrastructure.
McKinsey draws an important distinction between traditional change management and what agentic AI demands. Periodic communications, one-time training, static guidance documents, and top-down rollout plans create awareness. However, they rarely build the trust, habits, and operating discipline that let people work differently.
McKinsey recommends that companies put three dollars into process redesign and five into capability building and adoption for every dollar spent on agentic technology. Most companies invert it, with technology taking the majority of executive attention and investment while capability building and behavior change get handled as implementation details.
An agent can be technically capable and still deliver little value if the surrounding workflow remains unchanged. Scaling AI means deciding where humans exercise judgment, where machines can act independently, and how both improve through experience.
Agents need a foundation they can trust
As organizations give AI systems more autonomy, the quality of the underlying data largely determines the quality of the AI itself.
MIT Technology Review Insights’ Scaling AI Agents with Trustworthy Data reports that agents at surveyed organizations can only access about 45% of company data. Among the least mature, it is 30% or less; among the organizations the research calls data leaders, it is above 70%. Levels of trust reflect the same pattern. Trust in AI solutions is about twice as high at data leader companies.
Agents need data that is accurate, contextual, governed, and available at the moment a decision has to be made.
An agent operating across business processes needs to know which information it can trust, where that information came from, and how current it is. Poor data quality turns an otherwise capable agent into an unreliable decision-maker.
This puts a practical ceiling on agentic AI. Autonomy can extend only as far as the data foundation underneath it, and two-thirds of the least mature organizations report that legacy systems are limiting both their ability to scale agents and their speed of decision-making.
The more responsibility an organization gives an AI system, the more important its information environment becomes.
CEOs are bringing AI closer to growth
The shift from experimentation to execution is also changing what senior executives prioritize in AI initiatives.
Chief Executive’s In Final Stretch Of The Year, CEOs Push For Growth, Lean Into AI surveyed 285 CEOs and found a clear narrowing of focus, with 55% now prioritizing revenue and market-share growth, 43% profitability, and 38% operational efficiency. AI is increasingly considered alongside the business priorities executives are already accountable for, rather than as a separate innovation agenda.
That is an important marker of maturity.
When AI becomes part of discussions about growth, productivity, margins, and operating performance, the standard for success changes. The question for leaders becomes where AI improves the economics of the business, not merely where it has been installed.
Interestingly, just 13% of CEOs name talent a top priority, even as 27% cite workforce constraints as a real obstacle. Yet the CEOs who do prioritize talent report markedly stronger growth expectations. Close to 90% expect revenue increases, against 73% among those who don’t.
From AI use to AI value
AI is becoming easier to deploy. Turning deployment into business results is the harder work, and it is where this month’s research converges.
Four requirements run through all five pieces.
- Find where AI makes a real difference.
- Decide how the value will be measured before the system goes live.
- Redesign the work around it.
- Build the data foundations and the human capability that let it scale.
Maturity is the sum of those four. It comes from knowing where AI creates value, proving that it does, and scaling what works.
Most organizations miss the mark on the baselines, process redesign plans, and data foundations agents need. These are key elements of the RapidCanvas model. We help enterprises move from disconnected pilots to governed, measurable use cases that business and technical teams can trust.
That’s it for the August 2026 edition of This Month in AI. We hope you enjoyed the read.
Sign up for our newsletter to get this curated list of AI articles and more AI insights from RapidCanvas delivered straight to your inbox.
