By Alexander Puutio,Contributor

Copyright forbes

Learn how the 5% club is launching AI deployments straight to the stratosphere

MIT’s recent State of AI in Business 2025 report landed like a splash of cold water when it announced that only five percent of embedded, task-specific GenAI projects have been successfully implemented. Most readers stopped there, calling it proof of a bubble, which it very well might be, but the nuance tells a different story.

A five-percent hit rate for a technology that has barely celebrated its first birthday would make Clay Christensen proud on any day of the week. The same study also shows that forty percent of simpler, general-purpose deployments already deliver results, which most bubble-bursters gladly ignore.

And even then, the real headline hides well underneath the lede. You see, when that five percent is done right, it moves markets, not just individual outcomes.

One company showing how far a well-executed GenAI project can go is AlphaSense, the New York-based market-intelligence platform that just announced it has surpassed $500 million in annual recurring revenue. The company now serves thousands of enterprises, including many of the world’s most sophisticated investors who are dead-set on being in that 5%.

What makes that figure noteworthy is not just the scale, but how much of it now runs on agentic AI. AlphaSense CEO and Founder Jack Kokko says the milestone shows the gap between building something impressive in a demo and getting it to work in the real world. “The hard part isn’t building an AI that looks smart,” he says. “The hard part is making it deliver accurate, consistent results when people’s decisions and reputations are on the line.”

Now all you need is to figure out how to make that reliability repeatable, and you might just have a winner in your hands.

What Differentiates Between Early AI Adopters and Early AI Winners

When great ideas and even greater tools are plentiful, success comes down to execution. And execution, if anything, has proved harder than most anticipated. Even cutting-edge consultancies have stumbled with deploying AI as Deloitte recently discovered when one of its reports for the Australian local government turned out to be pockmarked with hallucinations.

MORE FOR YOU

If the consultants can’t get it right, what hope does everyone else have?

For AlphaSense, getting it right meant a decade-long obsession with accuracy, which is what everyone else seeking to join the 5% club need to urgently emulate.

“Our users make million- and billion-dollar decisions based on the information we provide,” says Chris Ackerson, the firm’s senior vice president of product development. “We’ve spent 10 years building and curating a half-billion-document library and 200,000 expert-call transcripts so that every model answer is trustworthy, cited, and explainable.”

That precision is no small feat, which is exactly why it’s easy to demo an AI agent but hard to make them deliver at scale.

You see, AI agents face a dual squeeze that is particularly poignant in the finance industry. They must be flexible enough to reason through uncertainty yet foolproof in deterministic workflows. This means that Ackerson’s team needs to run daily evaluations to prevent drift, building monthly benchmark datasets that mirror real-world analyst tasks. “The pace we’re innovating at just to keep pace is blistering,” he says. “Clients push these systems to their limits, and we tune against that pressure. There’s no other way to make it all work.”

Kokko frames it more simply by noting how in markets measured by basis points, hallucination is not an option. “When clients rely on your output to make billion-dollar decisions, trust becomes the product,” he says. “You can have the smartest models in the world, but if people don’t believe what they’re seeing, none of it matters. The real partnership providers need to forge is between accuracy and confidence.”

The 5-percenters Are Going For Action

As many companies do, AlphaSense began with frustration. Fresh out of Morgan Stanley, Kokko realized that while Google could parse the open web, nothing comparable existed for professionals drowning in filings, transcripts, and analyst notes. His idea was to make enterprise search act less like a database and more like a junior banker who actually read everything.

Fifteen years later, that vision has evolved from AI-powered search into a full agentic platform. In 2025 the company has rolled out Generative Search, Generative Grid, Deep Research, and now an AI Agent Interviewer, rounding out a suite that allows autonomous agents to source, synthesize, and even cross-examine information.

The shift mirrors what we’re seeing across enterprise software. Salesforce, Amplitude, and Anaplan and dozens of others across all industries are all moving in the same direction, which is decidedly away from static dashboards and firmly toward systems that think, adapt, and act alongside the user. Each is expanding horizontally across data sources while going vertically deeper into outcomes. The emphasis is no longer on delivering information, but on delivering outcomes.

It’s no surprise that Ackerson describes Deep Research as “an autonomous analyst using our own platform the way our customers do.” What once took a team six weeks now takes thirty minutes, and it comes fully sourced.

Kokko reflects on the agentic revolution the industry is going through the analogy of a force multiplier for the clients. “You can now walk into an M&A meeting smarter than anyone else in the room because an AI team of a thousand analysts just worked overnight on your behalf.”

In one case, a private-equity firm used AlphaSense’s Deep Research to analyze the U.S. mid-market banking sector. “Their internal team took five weeks; our AI did better in ten minutes,” he says.

When agentic AI hits the spot in finance, the productivity gain is measured not just in saved hours but in deals won. And that shows how transformative the 5% can be.

Deploy and They Will Come, As Long As They See the Value

The MIT numbers do not show that AI fails, instead, they prove that innovation is difficult. Even Einstein once said that it would not be research if we already knew the answer, and the same applies here.

What the success of agentic deployments like AlphaSense’s show is that the fundamental problem is not of ambition but of misalignment. Many AI projects chase technical novelty instead of client value.

According to James Luo, General Partner at CapitalG and an early AlphaSense investor, that is exactly what sets most companies back in their AI adoption processes today. “You have to double down on accuracy and workflow integration, including the boring plumbing of it all, if you want to see the ROI some of the early AI winners are seeing.”

Ackerson admits the temptation to chase the shinies is real. “We’ve seen a lot of sixty- or seventy-percent demos that look amazing on stage, but layering a bunch of human effort on top of weak AI isn’t adoption. Getting to reliable autonomy is.”

Reliable autonomy, in AlphaSense’s case, means domain-specific orchestration, something which companies like PwC betted on early in the current wave of AI adoption. Instead of one giant model, the company uses many specialized ones tuned to different reasoning styles and chained together. “It’s almost a taste thing,” Kokko says. “Some models write better for analysts, others for investors. We run them all and let the system and the client pick.”

What experts like Luo hasten to remind us, is that it’s never just about the models themselves. “What matters are what clients can do with them, and whether the models integrated with the right interfaces and workflows deliver results,” Luo notes The last decade of SaaS was about more tools to help users, but the next decade is about fewer tools that deliver exponentially more value, with each tool requiring less of a crank by the end user.

And if Ackerson has his way, that value will be delivered automatically instead of on-demand. “Deep Research today is transactional,” he says. “The next step is letting it run continuously in the background, monitoring the same sectors our clients care about and surfacing changes in real time.”

That shift from “use the tool” to “trust the agent” marks the maturation of AI adoption, and it’s not at all obvious whether most of the five percenters have made it to this level. Marketing AI tools is one thing, but embedding them into the core of an enterprise is an entirely different matter.

AI Adoption is Racing Towards an Augmented Workforce

Inside AlphaSense, few believe in the dystopia of mass job loss.

“We don’t buy into the displacement narrative,” Ackerson says. “What we see is reallocation. Analysts spend less time catching up on broker notes and more time with clients and management, more human-to-human work.”

Kokko agrees, seeing in AI a way to fix the burnout culture he once lived. “I remember analysts crying at two in the morning because we were building a deck for a client while their families were in crisis,” he recalls. “Those humans didn’t want to do that work. Now AI can.”

This evolution points to something deeper about the wave of AI adoption we’re witnessing. Value can no longer be measured only in returns. It must be felt at every node of the workflow, from the employee whose time is freed to the client who trusts the outcome. The companies that understand this will build systems that elevate labor, instead of simply automating it.

With half a billion in ARR AlphaSense is still “scratching the surface,” Kokko says, including in how far the company itself will go in reworking itself with AI.

Asked what comes next, Kokko does not hesitate. “We’re building the most intelligent brain in business. A system that reads everything that matters, connects every dot, and helps you act, not just analyze.”

Agentic AI may still face a five percent success rate, but those odds look less like a warning and more like an opportunity. Every technology has its early winners, and they’re already building the world the rest will inherit.

Now it’s up to us to study what makes their deployments succeed while others keep spinning their wheels.

Editorial StandardsReprints & Permissions