AI is everywhere–at least from a high-level perspective. But does it belong in your business? That depends.

There’s little argument that AI offers companies plenty of capabilities and opportunities, and most executives know it. When polled by PwC, 86% of professionals said AI was a fairly established part of corporate life. Yet even if it’s mainstream, AI shouldn’t be treated as just another tool.

To get the most out of AI, you have to understand it first so you can deploy it strategically in the way that makes the most sense for you, your team, and your goals. Ultimately, you’ll optimize your spending and get the most out of every solution and system you bring into your organization. The following steps will help you figure out how to bring AI into your fold.

1. Conduct research on AI’s strengths and limitations.

AI can do quite a lot. Nevertheless, it isn’t right for all situations. For example, if you want to aggregate and analyze large pools of data, AI can be helpful. The same holds if you’re interested in adding an AI-powered chatbot feature to your website to give customers self-service mechanisms.

What if your focus is on driving innovation within your company, though? AI can’t brainstorm on its own. An AI-enhanced program may be able to help your employees come up with concepts, of course. However, AI itself won’t turn out clever ideas because it has limitations that the human brain doesn’t, namely that it doesn’t really think or have common sense, as pointed out in an MIT Technology Review by Boston Globe writer Brian Bergstein.

So, where can you begin if you want to weigh the pros and cons of AI for your company? One method to sift through all the information available about AI is to determine how your industry competitors use it. Then, evaluate whether those uses will be just as advantageous for you. By being selective, you can make well-informed choices.

2. Start with small AI experiments.

It can be tempting to bring all kinds of AI products into your tech stack simultaneously. However, making too many sudden changes could backfire. Plus, you might not be able to tell which AI solutions are working and which ones need to go.

To avoid this problem, Tiago Ramalho, the CEO of AI consultancy Recursive, recommends taking a scientific approach by setting up small tests. “Break [AI] down into minimum viable objectives,” writes Ramalho. “Develop a model for the lowest-hanging fruit and build up from there.” He points out that tackling everything in micro-bites will show any flawed assumptions quickly. When they appear, you can address them right away.

Let’s say you wanted to see how AI could improve your overall customer service capabilities. Instead of revamping all your systems, you might want to try an integration that uses AI such as a notes transcription tool. Your customer service supervisors and representatives could take the tool for a test drive before incorporating something else into their workflows.

3. Tie AI into specific objectives.

AI systems with plenty of bells and whistles may make it feel like you’re moving faster and farther than other businesses. But if they aren’t helping you achieve your objectives, they’re only trappings. This requires that you find a way to measure the effectiveness of any AI implementations. The metrics that will work for your team are 100% dependent on your plans.

Take sales, for instance. Gartner shows that nearly 9 in 10 sales teams leverage AI or plan to in the near future. Maybe you want your sales team to convert more prospects. Consequently, you purchase AI software meant to analyze sales calls and pinpoint areas of missed opportunities. Once highlighted, those missed opportunities could be used to help sales team members close faster and more frequently.

By tracking everyone’s conversion rates and changing nothing but the application of an AI-based sales tool, you could see whether the AI was working. Having this type of “hard” evidence could be useful, especially to achieve buy-in for other AI investments from executives.

4. Set aside time for employee training.

It’s not enough to just purchase an AI system and expect everyone to use it. Your team members will need time for training. And not just in-class training, either. They deserve the space to fiddle around with the AI and familiarize themselves with what it can do.

Keeping this in mind, try not to set AI deployment goals that are overly ambitious. It takes time for most employees to adapt to change, particularly if the change affects their normal processes. Additionally, some workers may be at different places on the tech learning and comfort curve. Not everyone falls into the “early adopter” category, and some individuals feel intimidated by the thought of adding yet another tech tool to their everyday operations.

A good way to make any switchover to AI more frictionless is to make sure the workers who will be affected are brought into the discussion sooner rather than later. Allowing them to have a voice, ask questions, and make contributions will improve their ownership of the AI later. Plus, they might come up with valuable concerns or ideas you wouldn’t have thought about.

Does AI belong in your business? Probably. Your role is to figure out how to maximize the AI that’s on the market and make certain that any AI you choose serves a purpose.