7 Reasons Your Company Isn’t Profiting With AI, and How to Fix That

Entrepreneurs are going big on AI. But most are seeing small returns.

The numbers are staggering: Despite $30 billion to $40 billion of investment into generative AI tools and systems, 95% of companies are seeing zero return on that investment, according to a recent report from the Massachusetts Institute of Technology. The Boston Consulting Group (BCG) found that 74% of businesses reap no tangible reward from AI, and only 4% have generated substantial value.

Does that mean AI is useless? No. Those same reports show something telling: Among companies that have successfully integrated AI into their workflow and processes, their bottom line rose significantly. BCG identified 50% revenue growth as a result of AI.

So what are they doing right that everyone else is doing wrong?

We called a wide range of AI experts, and identified seven mistakes that entrepreneurs are making.

Here they are.

Mistake #1: Relying on your team, instead of yourself

A lot of CEOs share an attitude about AI: It’s something for their staff to figure out. “They are bass-ackward wrong,” says Steve Ferman, an AI strategist who works with entrepreneurs and business executives. 

Other experts agree: Don’t delegate your AI adoption. You need to be familiar enough with the technology’s capabilities and behavior to lead the charge. “CEOs actually getting their hands dirty and working with this for their own personal life or workflow is super predictive of organizations that are getting the most value out of AI,” says Amanda Luther, senior partner at BCG and global lead of their AI transformation topic (and a coauthor of the report highlighted earlier). “Just playing with it is worthwhile. You might find opportunities you didn’t know existed.”

Mistake #2: Experimenting randomly, instead of targeted applications

Too many companies treat AI like an experiment — trying a little of this, a little of that. But to drive growth and profitability in a meaningful way, your AI efforts should be connected to a focused strategy that targets your core business. This, the BCG report found, is where most of the ROI for the winning companies came from.

So how do you do it? Sol Rashidi, a senior Harvard Kennedy School Fellow who’s been an AI leader at companies like AWS and IBM, has a good framework: It’s the difference between using AI and actually doing AI. “‘Using AI’ means buying a few Copilot licenses and having everyone write emails with it. That’s not gonna get you those big lifts. ‘Doing AI’ is fundamentally picking a business problem that has plagued the company, or taking a function within the organization like financial planning and analysis or procurement, and then understanding where there’s value to extract if they were to run more optimally,” Rashidi says. 

Take that function, look at the day-to-day workflows, and break down the tasks and subtasks. Assess which low-complexity and  low-risk tasks can be outsourced to machines, and which must remain human — either because it requires a certain level of expertise or judgment, or because there are relationships involved that you can’t erode.

Mistake #3: Looking to cut costs, instead of creating market advantages

Yes, AI can help you snip costs here and there. But don’t forget that your competitors are doing that too, points out Goutam Challagalla, professor of sustainable strategy and marketing at Switzerland’s International Institute for Management Development (IMD). So instead, ask yourself: “How can I use AI to amplify an advantage in the market that’s hard for others to copy?”

In biopharma, for example, it may be about speeding up drug discovery, says Luther. A software company might focus on building the next generation of its product. A retailer might use AI for marketing. Once you’ve identified your target, think big, Luther says. “It’s saying, ‘How could my marketing function look totally different — everything from ‘What am I spending marketing dollars on?’ to ‘How am I creating content?’ to ‘How am I measuring that?’” she explains. “And then, ‘How am I creating a feedback loop that much more quickly senses what’s happening out in the market, so I can quickly respond to that?’” 

Mistake #4: Focusing on the tech, instead of the people

When leaders explore AI, they often get hung up on which tools and bots to use. But the BCG survey found that, in order to get AI transformation right, 70% of the focus should be on the human team.

First, assess where your people are. Many are likely already hitting up ChatGPT and Claude for help in their daily work. Others may not be yet, perhaps because change is hard or they view it as threatening. An investment in AI training goes a long way, says Ferman. “It’ll help the team feel more secure about using it and less like, ‘Oh, my God, it’s gonna take my job.’ It’ll help them secure their environment. And now you’ve bonded as an organization all working together on this new AI project.”

Mistake #5: Dumping your data into AI, instead of prepping it first

Companies expect AI to work miracles. They’ll dump a ton of data into a system, and think it will just…understand. But it won’t.

“They’re not scrubbing their data before putting it into large language models or setting up guardrails and rules and regulations within operating procedures to protect it,” Ferman says. Nor are they tackling questions like how to integrate separate data streams, who has access, and what ongoing maintenance looks like.

Other companies experience the exact opposite problem: They obsess over perfecting their data, and then get stuck on it. “I’ve been doing this for 20 years. I have never once talked to a company that thinks they have great data,” Luther says. “Data’s always going to be inherently messy, and there’s going to always be new sources that aren’t quite integrated in the right way. So there’s this middle ground of: OK, how do I get pieces of my data infrastructure that matter for the thing I’m trying to do, up to a good enough bar that I can start? Do I get 80% of it pretty clean and then do something with it? That’s the challenge to work through.” 

Also, with AI evolving so rapidly, data security is scrambling to keep up. “The major focus pre-2024 was about protecting data at rest — password protection, tokenization, identity access management, authentication layers,” says Harvard Kennedy Fellow Rashidi. “With the advancement of generative AI and agents, the focus has to pivot into protecting data ‘in motion’ and ‘in use’ through a variety of mechanisms, core capabilities like prompt provenance and prompt security, but of course there’s a lot more to consider in this space.”  

Mistake #6: Jumping on an AI agent, instead of asking whether you really need one

AI agents are clearly the future. And you should be preparing for that. “But doing agents for the sake of agents because you’re bold or your investors are pushing — that’s a very problematic way to start,” says Nufar Gaspar, an executive AI consultant and trainer for companies from startups to Fortune 500s. “You have to have the justification that this is the only technology that will solve the problem and go in with an eyes-wide-open understanding of the risk.”  

Just to be clear, people sometimes refer to a simple website chatbot or custom GPT as an AI agent, but the technology we’re talking about can make decisions and act on its own using tools to achieve a goal, rather than just responding to prompts. For example, a chatbot can answer customers’  basic questions, but an agent could also issue refunds, handle returns, and order replacements. Although agents can do impressive work,  it’s very hard to predict all the ways they might behave, says Gaspar — not to mention, they’re fairly new and still evolving.

In light of the risks, she advises businesses to use simpler technology any time it can meet their needs. If you’re sure that an agent is, indeed, the only thing that will truly solve the problem you’re trying to address, then keep in mind that the ROI may not be positive because of how complex they are to build. “There is a steep learning curve here,” she says. 

Mistake #7: Letting AI speak for you, instead of keeping your voice

It’s tempting to have AI write your marketing copy — and pretty much everything else. But the result isn’t pretty: The internet is now full of generic AI content and synthetic images. And the companies doing that will just drift further away from their customers.

Even if you’re using AI, it’s critical not to lose your unique voice and human touch. Ultimately, your company will thrive because of your relationship to your customer…not just your ability to reach them efficiently. “Being able to project authenticity,” says BCG’s Luther, “is now differentiating for the brands that can do that. And every day, it is becoming more important than ever.”