A software company recently sought consultation, convinced of an AI problem. It had bought subscriptions, given the team access to every model on the market, and built several automations. The fundamental operations of the business had not changed.
The issue was not the technology. It was a confusion between owning tools and building systems.
That pattern shows up often. Across the agencies and businesses Hexona Systems works with, spanning six continents and sectors from SaaS to logistics, companies that struggle with AI rarely struggle because they lack access. They struggle because access was never the hurdle.
Recent research points in the same direction.
A $40 Billion Gap
In July 2025, MIT Media Lab’s Project NANDA initiative published a report titled The GenAI Divide: State of AI in Business 2025. Drawing on interviews with 150 leaders, a survey of 350 employees, and an analysis of 300 public AI deployments, the researchers found that roughly 95 percent of the organizations they studied reported no measurable impact on profit and loss from generative AI. Enterprise spending on those initiatives ran to an estimated 30 to 40 billion dollars. About 5 percent of integrated pilots captured significant value.
The authors described the split as high adoption and low transformation.
That is the gap. Nearly everyone has the tools. Few have the results.
Two caveats belong with a figure that stark. The report presented preliminary findings rather than peer-reviewed research, and analysts have questioned its sample size and the decision not to publish the underlying data. Project NANDA also builds agent-based AI infrastructure, which gives it a stake in the argument that current enterprise approaches fall short. The direction of the finding has held up in the wider debate over enterprise AI returns, but the precise percentage remains contested.
The instinct is to read a number like that as a verdict on the technology. It is not. The models work. The tools work. What breaks is everything that happens after the demo.
Why the Demo Always Works and Production Rarely Does
A pilot is a controlled environment. Clean data, a narrow use case, a motivated team, and no real consequences if it fails. Of course it works. That is what pilots are built to do.
Production is the opposite. Data arrives incomplete. Edge cases appear that no one scripted. Volume climbs. And the moment something breaks at an inconvenient hour, someone has to own it.
Most AI projects don’t stall because the model was inaccurate. They stall because no one has defined what winning looks like in business terms, and no one is responsible for getting there.
A tool automates a task. A system produces an outcome. That difference is the entire game.
Businesses Are Spending in the Wrong Place
One finding in the report should change how leaders budget. Money flows overwhelmingly into sales and marketing, the visible and demo-friendly corner of the business. Returns concentrate elsewhere. The report found stronger returns in operations and back-office work, the unglamorous processes that move information from one place to another.
This aligns with industry observations. The automations that truly transform a business are rarely the most conspicuous. Instead, they are the routine, high-frequency processes: lead routing, onboarding, follow-ups, and reconciliations. The work that is least likely to be showcased is often the work that yields the greatest compound interest.
The report flagged one more pattern worth repeating. Tools built with external vendors and partners succeeded roughly twice as often as internal builds. The authors do not read that as a talent shortage inside companies. Durable automation is a discipline in its own right, and treating it as a side project is how it ends up abandoned.
The Missing Role: The Operator
Companies deploy AI and then wonder why it drifts. Often the answer is that no one owns it.
IT builds it. Operations uses it. When it breaks, responsibility dissolves into the space between them, and small failures compound until someone quietly switches the system off.
The businesses closing the gap treat automation the way they treat other core infrastructure. They assign an owner. Increasingly, that owner is an automation operator, someone who understands both the business logic and the technical execution, who monitors performance, refines behavior, and steps in when reality does not match the diagram.
Among the builders in the Automation Institute community, the ones who create real value are not always the strongest engineers. They are the ones who take ownership of outcomes rather than tools.
Close the Gap by Lowering the Bar to Ship
One more habit separates the teams that get results from those that do not, and it runs counter to instinct. They deploy earlier than they are comfortable with.
Perfection is where AI projects go to die. Teams polish a system in testing for months, chasing a version of ready that production will never respect, because real complexity only reveals itself in real use. The teams that pull ahead ship a smaller, sturdier version, watch what breaks, and improve it in the open. That is a recommendation drawn from practice rather than a finding in the MIT data.
Automation is not a configuration you finish. It is a system you run.
The Gap Is a Choice
The AI automation gap is not a technology problem, and no new model will close it. The companies stuck on the wrong side of it bought tools and expected transformation. The companies on the right side built systems, assigned ownership, and aimed at outcomes, iterating until the results were real.
The tools are now available to everyone. That is why they are no longer the advantage. The advantage is what a company builds around them.
Hamza Baig is a Toronto-based AI automation entrepreneur and the founder of Hexona Systems and the Automation Institute. Hexona Systems received the Platinum SaaSpreneur Award in 2024 and the Diamond SaaSpreneur Award at HighLevel’s 2025 LevelUp Summit in Dallas. HighLevel awards the Diamond tier to partners managing more than 1,000 active sub-accounts. According to the company, the Automation Institute has trained more than 40,000 automation builders worldwide.
