With 90% of companies working on artificial intelligence (AI) projects, organizations are realizing the vitality of AI for efficient business operations. Spending money on AI projects could ultimately cut down costs on lengthy manual tasks people would have to conduct instead. This isn’t just a financial cost, but a time cost, as tasks like data analysis and tracking have been done by human hand in the past.

AI brings a convenience and immediacy to data processes unparalleled to prior attempts, which is why 96% of companies said they expect to see machine learning projects continue to skyrocket in the next two years.

SEE: Artificial intelligence: Trends, obstacles, and potential wins (Tech Pro Research)

“Most manifestations of AI in business today revolve around machine learning, and the use cases are quite vertically dependent,” said Brandon Purcell, principal analyst in customer insights at Forrester. “Many customer-facing companies use machine learning to glean insights about their customers.”

For example, “manufacturers and utilities use machine learning for predictive maintenance, and retailers use it to optimize inventory levels; companies are building chatbots to handle routine customer service inquiries; companies are also using speech and text analytics on customer service and feedback data to identify pain points and improve the customer experience,” said Purcell

While AI opens the opportunity for many exciting possibilities across industries, many implementation challenges arise. Previously, problems with AI execution have commonly been attributed to employees’ lack of experience with the technology, resulting in a learning curve for business professionals. Often, companies have to reach for outside talent to help get the most out of their resources, said Purcell. However, humans are not solely to blame for AI’s limitations.

Here are three limits to AI’s use in the enterprise that tech and business leaders tend to overlook.

1. Data

In order for AI to do its job, models need to be trained on data. However, data brings quite a few obstacles to the table. “The most pervasive limitation to AI adoption is data. AI needs data to learn to perform its function,” said Purcell. “Unfortunately, I’ve yet to speak to a company that has its data house completely in order. In most companies, data is typically siloed and rarely consistently catalogued and governed. Without good, relevant training data, a company will find it quite hard to get started with AI.”

Often, companies think they may not have enough data to work with AI in the first place. The key here, though, is to remember that it’s not about having enough general data, it’s about having “actionable data that will help them learn, that is suitable for whatever task they have in mind,” said David Parmenter, head of data science at Adobe.

Another data-related limitation has to do with data standards and regulations. Companies need to determine whether the data has the right parameters, said Whit Andrews, agenda manager for AI and distinguished analyst at Gartner. Organizations need to make sure that their data is able to be shared with different companies based on federal, state, and internal requirements for those organizations, Andrews said.

2. Lack of knowledge

Another limitation to AI is that machines often don’t know what they don’t know, said Parmenter. While AI is fantastic for interpreting large volumes of information, there is no guarantee that the technology will understand all the data.

A great example the occurred in September involves the Nest doorbell, said Parmenter. The doorbell locked a man out of his house because he was wearing a shirt with Batman’s face on it, and the doorbell didn’t recognize Batman. “It’s very funny, but it’s real world,” added Parmenter.

3. Bias

Hidden bias is present in both people and data, and oftentimes bias is transferred to data because of people. “We can’t do these jobs without getting data. Then you go shopping around for data, and the data may have a bias in it that you don’t even know about,” said Parmenter. “You’re just blind to it.”

One example is from the world of autonomous cars, Parmenter said. “You’re going to get more data in wealthy neighborhoods, because that’s where autonomous cars are gonna go first,” he added. “I really don’t think any practitioners in my field are bad actors, but we really have to be open to the implications of what we’re doing and making sure that we are fair and evenhanded.”

The biggest thing companies need to remember when adopting AI is why they want it. “Don’t do AI for the sake of AI,” said Purcell. “Start with a business case grounded in customer insights from behavioral analytics and market research.” Companies will end up wasting a lot of time and money trying to implement AI for no good reason. Make sure your company has the data and reasoning first, then execute, added Andrews.