Feature
Recent IDC research suggests companies are reaping almost $4 for every $1 invested in generative AI. Former Google leader and business consultant Louisa Loran argues that figure reflects idealized conditions rather than how enterprises actually operate.
"Most enterprises are overestimating short-term ROI,” Loran said. "The often-cited $4 return per $1 invested reflects potential, not performance. It assumes flawless execution and unlimited internal capacity, which few organizations have."
Why Generative AI ROI Is So Hard to Measure
Calculating accurate ROI on GenAI investments is almost impossible, said LSE fellow Rebecca Homkes, because any calculation that doesn't account for the substantial fixed-cost investment in preparation and scalability is inaccurate and "fairly inauthentic."
Companies announcing big gains from GenAI were early adopters who invested millions over several years in infrastructure, data systems and model preparation, Homkes said. Companies joining the rush only after ChatGPT's public launch are much further behind. Even Amazon walked back some of its tangible gains announcements the same week they were made.
Data and tech infrastructure in most companies is pieced together across years of inorganic growth and bolt-ons, Homkes said. Getting ready to do anything meaningful will take years, she said.
On the other hand, Dan Herbatschek of the Ramsey Theory Group is more optimistic. In his experience, the $4-to-$1 return is not only realistic but conservative. The actual number may be higher and will certainly be higher in 2026 based on his clients' experiences, he said.
Yet even Herbatschek concedes the fundamental problem: Companies are tracking the wrong key performance indicators, focusing on technical metrics such as number of models, data volume ingested or features engineered, instead of business outcomes.
Surveys, Not Audit Logs
"A lot of reported gains come from surveys, and not an audit log,” said Nuha Hashem, co-founder of Cozmo AI. Businesses focused on proof-of-concept rather than proof-of-performance are overestimating short-term ROI, she warned. They add up pilot results and early savings but that is not the same as long-term gain, she said.
Hashem’s company tracks what she calls the Outcome Ownership Rate, or how often an AI “employee” completes a task on its own within company rules. Finished calls, closed claims and resolved issues show work that's done right and can be checked later.
Few of today's reported gains are measurable because most organizations still account for productivity in static terms, Loran said. "Anything else is responsible maintenance disguised as transformation." True impact appears only when efficiencies alter the financial structure, such as reducing selling, general and administrative expenses; shifting cost-to-serve or enabling new revenue models, she said.
NetMind's Chief Commercial Officer Seena Rejal is also skeptical, saying that most productivity claims remain projections rather than proven outcomes. It's still early days, he said.
The Hidden Costs of Moving Too Fast
Hidden costs appear where visibility ends, Hashem said. The pilot phase may look smooth, but retaining accuracy and explainability over the long term requires more work refining models and retraining humans for new roles as systems mature.
Companies are moving faster with GenAI than their data quality handles, Hashem added. Many projects launch on data that was never meant for AI. If the input is messy the output stays messy, she said. Strong habits around data quality are more important than new tools applied to that data. Companies that don't prioritize quality data will see up to 15% productivity losses, Rejal warned, citing IDC research — a result of rushed deployment.
The most underestimated expense is friction, Loran said. "When the speed of technology meets the speed of governance, a two-minute demo turns into a two-month approval cycle,” she said. “That delay is not just inconvenience; it's lost capacity and compounding cost."
In regulated sectors, governance costs spiral particularly fast, Rajal explained. Banks need model risk management teams to validate AI systems and maintain audit trails for regulators. Legal firms must build technical safeguards preventing information leaking between clients. Insurance companies using AI to set premiums or assess claims must prove systems don't unfairly discriminate. These costs are permanent overhead that increases with every new AI deployment.
These hidden costs often represent the majority of total AI expenses once data quality, infrastructure and integration are included, Homkes said. Too many organizations treat governance as a guardrail, something that slows and protects, when it should function as infrastructure supporting speed, clarity and control. Without fast-track lanes for experimentation and risk-based decision frameworks, innovation slows, compliance debt builds and total cost of ownership rises long after the initial investment, she said.
In many organizations, the pace of AI deployment outstrips the systems designed to keep it safe and credible, Loran said. The pressure to show momentum leads to pilots built outside normal enterprise guardrails, creating a gap between experimentation and accountability. That gap is not just operational, but also reputational and financial. "The real test of leadership is to align speed with integrity by embedding governance into the build itself, not as an afterthought,” she said.
The paradox is that the pace of change slows investment. Companies that weren’t early adopters fear overinvesting in the wrong technology. Custom GPTs offer a cautionary tale: companies moving too early could have spent millions on something that cost a few hundred thousand a year later.
From Boardroom Optimism to Frontline Reality
Boards want evidence of speed, Loran said. Executives respond with promises of efficiency and participation in latest trends, even when not understanding the effect nor organizational implications underneath. "This is the incumbent's dilemma in human form: the desire for progress without disruption,” Homkes said. Most enterprises are still built around structures that protect continuity, not capacity, she added.
Automated AI “employees” help take the load off repetitive tasks and are advancing quickly toward decision making, Hashem said. But human employees deal with the gaps daily, she said. They see where the AI tools stop working and where humans need to step in. "The boardroom stories sound smoother and more optimistic than the truth in the call center, sales office or collections department,” she said.
Half of all AI-enabled enterprise applications will require new oversight positions dedicated to governance, risk and accountability, according to IDC predictions Rejal cited. Human oversight remains essential and can't be discounted from the process.
Most companies across most industries are still dabbling rather than integrating AI into value-creating strategies, Homkes said. Lacking centralized strategic insight, organizations rely on individual teams to select their own tools and use cases. While this boosts individual productivity, it eventually slows the organization down with hundreds of science projects but no meaningful return.
Webinar
Dec
11
Fuel Your Front Line: Workforce Trends Shaping 2026
Get expert insight into the tools and tactics top frontline teams will need in 2026.
Register
Webinar
On demand
Fix the Disconnect Between Projects and Profit
Your margins are hiding in disconnected data. Let's find them.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
Webinar
On demand
Accelerating Healthcare Ops with AI
Sign up to see how your peers are putting AI to work — and making it stick.
Watch Now
Webinar
On demand
From Nice-to-Have to Non-Negotiable: Prove the Value of Your EX Stack
Transform your tech stack: prove its value, secure your future.
Watch Now
Webinar
On demand
The Great Workforce Breakdown (And How to Stop It)
Disconnected employees. Failing communication. It’s time to turn things around.
Watch Now
The Risk of Inflated Expectations
Projections about doubling workforce productivity by 2027 assume perfect implementation and ideal conditions, Rejal said. When reality doesn't match projections, budgets get cut and promising initiatives get abandoned.
High hope leads to an AI bubble if people buy into stories rather than proof-of-performance results, Hashem warned. Hype always starts high but actual adoption grows iteratively. Proof will be how many systems are running in production a year from now, she said.
Inflated expectations around agentic AI may trigger another wave of disillusionment, not because the technology lacks potential, but because organizational readiness lags behind ambition, Loran said. "Agentic AI is being presented as a leap toward autonomy, yet most enterprises still struggle with basic data discipline and decision accountability,” she said.
Many of the most valuable business problems remain better addressed through optimization, prediction and human judgment, where intelligence helps decision-making rather than replacing it, Loran added. "Disillusionment will come if executives over-index on novelty instead of structure,” she said. “Transformation doesn't fail for lack of innovation; it fails when leaders underestimate the redesign, data integrity and governance required to make it real."
AI and its transformative effects are real; we can put that debate to rest, Homkes said. But she would like the hype to calm down so executives trying to create value for their shareholders have more realistic expectations of investment and return. Whether businesses are building the foundations required to capture AI's value, or racing toward disappointment wrapped in optimistic projections, remains the central question.
Editor's Note:
- 5 AI Metrics Every Leader Should Track — While everyone is eager to chase down that elusive ROI, fixating on that number alone won’t give you the full story. Here are five metrics to keep your eye on.
- What It Takes for GenAI Pilots to Scale — Alan Pelz-Sharpe, Rebecca Hinds and Craig Durr join Three Dots to discuss why individual productivity gains with GenAI haven't scaled across the organization.
- How Businesses Are Turning Generative AI Into Measurable Value — Successful GenAI rollouts share at least one thing in common: the initiative starts by identifying and aligning with a core business goal.