AI ROI: Beyond the Hype
If you look at the headlines, the verdict on the AI revolution is becoming increasingly cynical.
From Goldman Sachs questioning the $1 trillion capital expenditure to Sequoia Capital asking "Where is the revenue?", the narrative has shifted from euphoria to skepticism. The prevailing data suggests that for most companies, AI initiatives are becoming money pits with low returns.
But if you are a leader in a Tech business, accepting this "average" at face value is a dangerous strategic error.
The "low ROI" narrative is technically accurate in aggregate, but it is practically misleading. It conflates experimentation with production, and it ignores the massive variance between legacy enterprises and modern tech stacks.
Here is the evidence-backed case for why the "AI failure" narrative doesn't apply to tech businesses executing with discipline.
The Average Hides the "Power Law"
The primary issue with pessimistic AI reports is that they average out the results. In the world of AI adoption, there is no bell curve—there is a power law.
Recent analysis from McKinsey (Global AI Survey) highlights a stark bifurcation. While the median company is struggling to break even, a top tier of "high performers" (roughly 10–15% of organizations) are attributing significant portions of their EBIT (Earnings Before Interest and Taxes) to AI adoption.
The "average" ROI is dragged down by the 80% of companies that are merely dabbling.
According to BCG and Forrester, ROI on AI is typically negative until a company scales past a specific threshold of production use cases. The "losers" are stuck in the experimentation phase; the "winners" have pushed through to scale, where the unit economics flip to positive.
The “Pilot Purgatory”
Why are so many companies failing to see returns? The failure mode is rarely the AI model itself. It is almost always operational.
Gartner predicts that through 2025, at least 30% of GenAI projects will be abandoned after proof of concept. This is known as "Pilot Purgatory."
For a non-tech company (e.g., a 100-year-old insurance firm), 70% of the effort in an AI project is non-AI work: cleaning unstructured data, fixing legacy APIs, and fighting bureaucratic resistance.
Tech-native businesses usually skip this "tax." If you already have a modern data stack, structured APIs, and a culture of CI/CD, you are immune to the friction that kills ROI in legacy sectors. You aren't spending millions cleaning data; you are spending it building features.
Focusing on the Wrong Return
In the rush to justify costs, many businesses are measuring AI ROI using the wrong metric: Headcount Reduction.
This is a trap. In a technology business, the currency of success is Velocity and Quality, not just cost savings.
Research from GitHub and Microsoft regarding Copilot usage found that while developers completed tasks significantly faster (up to 55% in some trials), the real value was in the reduction of "drudgery" and context switching.
If AI allows your senior engineers to spend 30% less time writing boilerplate code and 30% more time on architecture and complex problem solving, the ROI isn't found in "saving a salary"—it is found in shipping a better product three months earlier.
Vertical vs. Horizontal AI
The gloomiest stats often come from companies buying "Horizontal AI"—generic licenses (like ChatGPT Enterprise or Office 365 Copilot) for every employee without specific workflows.
Sequoia Capital’s recent analysis suggests the real value is accruing to "Vertical AI"—systems designed to solve specific, high-friction problems.
Companies that sprinkle AI dust on everything see low returns. Companies that build "Agents" for specific workflows—like automated QA testing, Level 1 Support triage, or dynamic pricing—are seeing measurable impact.
Tech companies are uniquely positioned to build these vertical integrations internally, rather than waiting for a vendor to sell them a generic solution.
The Verdict
The "AI ROI Crisis" is real for companies that treat AI as a magic wand. They are stuck in pilot purgatory, paralyzed by bad data, and measuring success by how many people they can fire.
Here are three take-aways to ensure your AI implementation puts you in top tier of performers:
3 Take-Aways
- Avoid "Magic Dust": Don't just sprinkle AI on everything. Pick one high-friction workflow (e.g., "Customer Support Triage" or "QA Testing").
- Human-in-the-Loop (HITL): Design for augmentation, not full automation. This lowers the risk of failure to near zero because the AI only has to draft, not decide.
- Measure Velocity, Not Just Cost: define ROI as "features shipped per quarter" or "time to resolution," not just "hours saved."
Audit your operations
Get a roadmap to fix bottlenecks and scale your GTM engine.
Restez informé(e)
Updates du monde RevOps, automatisation, IA et productivité, livrées dans un format de tweet.
Cet article vous a plu ?
Partager
Écrit par
Marvin Luksenberg
Co-Fondateur @ KLAP
