Copyright Fast Company
You can’t plug tomorrow’s technology into yesterday’s systems and expect transformation, still most companies are learning this the hard way. The market has plenty of revolutionary AI software. Despite us riding just the first wave of many, today’s predictive intelligence models were but dreams until the very recent past. Availability and validity are not the problems, yet the failure point is as old as software itself. Inside nearly every business are all too familiar stumbling blocks: disparate systems, lack of process, and—say it with me—dirty data. The MIT/NANDA report on AI that shook pundits and markets alike, revealed that 95% of enterprise AI deployments fail to deliver measurable results. That number should surprise no one, especially anyone who’s lived through a technology disruption or two. Given that we have been through this cycle before, let’s avoid the inflammatory urge to categorize this as a “bubble” and instead call it what it is—a correction. It’s a wake-up call that plug-and-play AI isn’t coming to save everyone who hasn’t first done the foundational work. You can’t simply layer AI, a solution which requires training, on top of a broken foundation and expect sudden success. If history has taught us anything, it’s certainly that with Hail Mary adoption, the only guarantee is fast failure. DISCONNECTED DATA, DISCONNECTED RESULTS The problem isn’t the tool. It’s the environment it enters. Most businesses still operate with fractured, siloed, and sometimes even contradictory data. They’ve spent years duct-taping systems together, relying on consultants who never fully understood the business. Now, when AI enters that stack, it encounters a mess it can’t make sense of. Subscribe to the Daily newsletter.Fast Company's trending stories delivered to you every day Privacy Policy | Fast Company Newsletters Some of the most expensive AI projects stall not because the model underperforms, but because it has nothing coherent to learn from. A chatbot trained on case records, PDFs, CRM notes, and Slack history isn’t learning the interaction. It’s learning interpretations of the interaction. The signal is buried in noise. Without context, there is no predictive intelligence. THE AUTOMATION LIE There’s a common misconception that AI will automate poorly run systems into high performance. That belief is fueling massive waste. Many enterprise leaders still believe they can skip the painful internal work and jump straight to transformation. You cannot automate what you haven’t already understood. You can’t train an agent to execute a process that only exists in someone’s head. And yet, organizations still pay six-figure consulting fees to help third parties “come up to speed,” instead of addressing the real issue: internal clarity. Until companies own their own excellence, AI will continue underperforming. It cannot replace institutional knowledge that was never available in the first place. THE FIRST REAL FIX: GET SPECIALIZED AND GO DEEP What businesses actually need isn’t broad-based AI. They need an internal shift toward building organizational centers of value. This starts with identifying key business areas ready for innovation. Not generic SOPs or wiki pages, but true mature centers of knowledge that are innovation ready. This often looks like agentic AI adoption centered on specific roles and use cases. Then, again, get specific. What problem are we solving, how are we solving it today, what is the success metric? AI only becomes a force multiplier when the business has a clear and consistent structure to multiply and clear goals for these experiments. Unfortunately, most teams are years behind. They rely on third parties to plug in tools like Zapier, Chili Piper, or HubSpot without ever really understanding the workflows underneath. What results is a patchwork of low-cost tools solving high-stakes problems. It doesn’t scale. YOU’RE NOT UNDERSTAFFED, YOU’RE UNDERPERFORMING The answer isn’t to throw more people at the problem. The answer is quality. In one analysis, a company spending $450,000 annually on outside consultants could have hired three top-tier, AI-literate internal operators. Don’t forget, someone needs to manage agents in similar ways to how you manage employees. Agents are simply more efficient. Fortunately, this talent already exists. Thanks to recent shifts in the job landscape, companies are seeing applications from candidates with top university backgrounds and deep systems thinking skills. The leverage is there. The will to use it is not. advertisement I recently listened to Jason[DA1] Lemkin detail how SaaStr adopted agentic AI with the magic ingredient of intentionality to deploy 11 agents which have driven over $1.5 million in just over 100 days. They did this by first selecting tools purpose-built for the functions they wanted to transition to agentic AI. The areas they focused on were process and data mature, employing the preparation steps I mention above. But that was just the beginning of the journey. The antithesis of set it and forget it, their teams spend 30% of their time training, managing, and running experiments with these agents. We’re talking about daily optimization. The result is a company with a single digit employee count producing eight-figures of revenue. That’s the type of environment you want to learn from. FOCUS ON METRICS THAT MATTER Most businesses measure revenue growth and EBITDA, and those same measures translate well to AI adoption. However, they don’t tell the full story. Inversely, very few companies really focus on revenue per employee. Combined with growth, that metric tells the real story of organizational efficiency. Across my companies, hitting $350,000-$400,000 per employee was possible, but only because systems were designed to avoid overhiring from day one. Without that kind of operational clarity, AI simply becomes another line item. In the end, AI success really comes down to commitment. You don’t need another pilot. You need to approach AI with the same conviction you would have when hiring an executive. You need to build the conditions for success, select the best tools based upon purpose, and steward those tools until their output is worthy of your customers’ expectations. The technology is ready. Most businesses are not. Justin Gray is cofounder and managing director of In Revenue Capital.
