This article is part of our Future of AI series from Imagination in Action 2025 Silicon Valley Summit — where founders, leaders, and investors explored what’s next for AI. Explore the magazine.
Most enterprises are struggling to integrate AI into their businesses and realize value from the technology.
According to a recent MIT study, The State of AI in Business 2025, despite investments of $30-40 billion in AI, 95% of companies have seen zero return.
This report sent shockwaves through the AI sector. What hasn’t received as much attention is that the report’s authors told Fortune that some large companies and startups are “really excelling with generative AI,” extracting millions of dollars in value while the rest flounder.
Companies that can close that gap with innovative AI solutions tailored to meet the needs of large and deep-pocketed enterprises stand to bank a fortune.
Enterprises need more customized and powerful AI tools, it’s true. But they could also benefit from a greater willingness to embrace change, especially with respect to their workflows and processes.
That puts technical leaders in a pivotal role, said Sharon Mandell, CIO of Juniper Networks. “Their role is to help those who don’t naturally understand the technology,” reducing friction and resistance to new technologies.
Why AI adoption takes time in the enterprise sector
Erik Brynjolfsson, a Stanford professor and director of the university’s Digital Economy Lab, agreed that there is a lot of disappointment about AI’s seeming lack of value. But, he added, that shouldn’t come as a surprise at this stage.
“That’s actually a very common pattern with general purpose technologies, such as the steam engine and electricity. You have these (powerful) capabilities, but getting that to turn into business value often takes much longer.”
The pattern is familiar: Transformative technology arrives, expectations soar, reality disappoints, and then—after painful adaptation—value finally materializes.
We’re currently in the “painful adaptation” phase.
The data readiness crisis
There are other reasons for AI’s weak results in the enterprise, such as a lack of data readiness. Older companies may have data in a wide variety of incompatible and hard-to-access formats. They may have enormous data lakes filled with unstructured and poorly managed data.
Or, they may have troves of valuable intellectual property accumulated over decades. Many companies, especially in the financial services and healthcare industries, need to worry about complying with regulatory requirements and privacy concerns.
These companies have all sorts of data controls, and are typically wary about anything that might affect their data, said Randall Lane, chief content officer at Forbes.
There are other reasons for AI’s weak results in the enterprise, such as a lack of data readiness. Older companies may have data in a wide variety of incompatible and hard-to-access formats. They may have enormous data lakes filled with unstructured and poorly managed data.
Or, they may have troves of valuable intellectual property accumulated over decades. Many companies, especially in the financial services and healthcare industries, need to worry about complying with regulatory requirements and privacy concerns. These companies have all sorts of data controls, and are typically wary about anything that might affect their data, said Randall Lane, chief content officer at Forbes.
“For an agent to be truly effective, it needs a lot of freedom.”
Randall Lane, Forbes
“For an agent to be truly effective,” Randall said, “it needs a lot of freedom (to manage data).” That’s going to make more conservative companies reluctant to roll out agentic or other AI technologies very fast.
The knowledge gap
Sometimes the problem isn’t bad data—it’s no data at all.
Laura Shact, a principal in the human capital practice at Deloitte Consulting, said the COVID-19 pandemic revealed to many executives that they had no idea what work employees were actually doing.
“When organizations examined job roles or titles, most people within the organization admitted, ‘Oh, that’s not what people actually do,’ Shact said. “This (knowledge) gap makes it extremely difficult for companies to look at people’s roles, personas, and skills to determine their future needs.”
Other times, attempts to build new apps that automate or streamline workflows are stymied by the use of outmoded systems or processes. Companies may be too attached to legacy processes. Shifting to a new way of doing things may require too much change management and culture shifts.
Without understanding current processes, building AI systems to improve them becomes impossible.
Big migrations are getting easier
Even today, many companies in regulated industries continue to operate their own data centers rather than becoming cloud-native—and that can limit AI’s effectiveness too.
In some cases that’s a calculation about minimizing perceived risk. Other times, decision makers may simply want to avoid the hassle and cost associated with migrating to new tech platforms, which often involves upgrading code.
Years ago, it was typical for banks and other regulated companies facing a large platform migration involving 20 million or 30 million lines of code to hire consultants; the projects could require millions of dollars and more than a year’s worth of labor.
But Brian Elliott, CEO of Blitzy AI, a maker of an autonomous software-development platform, said the tools now exist to enable worry-free technology migrations. Blitzy is built around a large-scale platform that deploys multi-agent systems to complete development tasks such as large-scale migrations.
“Our developers were able to do that in two and a half months,” Elliott said, also noting customers get the added benefit of not having to pay a consultant’s markup fee.
The technology exists. The question is whether enterprises will embrace it.
Breaking down the walls
A growing number of C-suite executives are recognizing that data silos stand in the way of AI success. Execs at these companies welcome services and tools that can tear down the barriers separating data from the people who can use it.
In the AI era, this is what’s required to unlock value from a company’s data: Toppling silos and providing AI with access to as much data as possible.
Even when attitudes shift, plenty of potential pitfalls remain. Individual business units may adopt a range of unintegrated AI tools or isolated AI agents to suit a narrow set of needs. A lot of organizations are still without a single source of truth and data sources remain scattered. In cases where companies intend to enable AI to make decisions based on the totality of their data, they can inadvertently wall off information to “the detriment of the larger organization,” said Philip Rathle, CTO at Neo4j, a database company.
It helps to take an expansive view of data. Abhishek Mehta, CEO of Tresata, a data and predictive analytics company, says that the least successful companies he sees are those that think “the only data that is valuable to you is within your four walls.” In contrast, Tresata’s most successful clients combine their company’s data with information from other sources. He added that this strategy generates “supernatural value.”
These experts all agree that unleashing AI’s potential means combining isolated stores of data. That is what Lisa Dolan, managing director Link Ventures, calls the “big unlock.”
The shift away from DIY AI
Some of the earliest enterprise AI efforts centered on creating bespoke AI models. Companies hired data scientists and engineering teams and set out to build their own LLMs and custom data integrations. Some of those companies quickly became disillusioned with the expense and complexity of this approach.
“Now buyers want specific AI teammates and want to buy a complete service, not just the technology.”
Patrick Salyer
That’s all changed. Many companies have now concluded that it no longer makes sense to build their own AI and data management infrastructure. They don’t want to pay their employees to organize, store and clean data, or to create models from scratch—not when it’s easier and more effective to outsource those chores or use best-in-class LLMs from the leading AI vendors.
“DIY didn’t work well,” said Patrick Salyer of Mayfield. “Now buyers want specific AI teammates and want to buy a complete service, not just the technology.”
The shift has opened opportunities for AI companies upgrading the enterprise stack. Those include the agentic systems that enable goal-directed execution rather than task-oriented execution. Instead of asking AI agents to perform individual tasks, agents increasingly have the ability to figure out what tasks need to be performed to achieve a stated goal, then perform those tasks autonomously, said Rao Surapaneni, vice president of Google’s CloudAI Business Application Platform.
“Agents are now able to collect, collate, and synthesize lots of disparate information; summarize it; and reduce cognitive load.”
Rao Surapaneni, Google
Another way customers are deploying agents is to perform deep research, which Surapaneni says he himself uses every day.
“With current reasoning capability,” he said, “agents are now able to collect, collate, and synthesize lots of disparate information; summarize it; and reduce cognitive load.”
Agents also are evolving into the primary interface for interacting with technology, according to Nancy Xu, vice president of Agentforce for Salesforce. “I think the conversational interface that many of us are familiar with in terms of how we interact with agents is becoming a new user interface,” she said.
The path forward for Enterprise AI
Companies targeting the enterprise market should focus on agents that perform mission-critical tasks in observability, governance, security, and performance measurement.
Long term success requires better pricing models. Charging by the number of tokens used, as many AI companies currently do, will not be a viable long-term strategy for the enterprise market.
Vendors must also help enterprises understand that their own data can create competitive moats. But to accomplish that, enterprises need to embrace new ways of leveraging data, breaking down internal silos and developing new workflows and processes.
Will enterprises be ready to take advantage of those capabilities? That depends entirely on their ability to overcome their own inertia. For the companies that can help them close that gap, the opportunity is extraordinary.
Founder Takeaways
Explore The Future of AI | This article is part of our Future of AI series from Imagination in Action 2025 Silicon Valley Summit — where founders, leaders, and investors explored the next revolution of AI. We explored how AI is changing scientific research, creating new startup economics, straining power grids, and challenging us to rethink everything from enterprise software to regulatory frameworks. Dive into the Future of AI magazine to see the full picture.
