Executives Clara Shih, Chief Executive Officer at Salesforce AI, and Asha Sharma, Chief Operating Officer at Instacart, shared a few ways they’ve successfully implemented the technology at Fortune’s Most Powerful Women Summit in Laguna Niguel, Calif.

Their insights shed light on how leaders can successfully implement AI as part of a modern business strategy. Here are their top four insights.

Using data for efficiency

“AI is only as good as your data, you need data to train the AI models, you need AI to fine tune models, and then of course, to ground and give context into the prompts so that the models don’t hallucinate,” Shih said.

Her team at Salesforce is using data to help automate common customer service requests. They’re also working on a product for salespeople that would easily pull the background on an account, saving them hours of research time.

Up until now, practical implementation of AI has been difficult. Salesforce is addressing that with Einstein, a platform that lets customers create their own prompts without code or technical expertise.

Instacart is using data to break down current limitations, such as a 1.4 billion item grocery catalogue that is currently focused on individual ingredients (avocados, seasoning, tortillas) vs. meals. (tacos).

“For consumers, the number one thing we’re thinking about is how do we help them meal plan? How do we help them find what’s for dinner? If you think about it, a grocery store has 30,000 items, it’s really hard to spearfish by shopping for an avocado or cardamom instead of actually shopping for the meal. And that’s what generative AI is going to do,” Sharma said.

Connecting with customers

Instacart has successfully harnessed AI to make the customer and advertiser experiences better, Sharma said.

“When one of our 600,000 shoppers are going into the store, we update our catalog in real time when they’re taking items off the shelf, and that feeds all the way through to our advertisers,” Sharma said. “So they’re not advertising on products that aren’t even available to customers.”

AI tools also help match customers with shoppers who are close by, meaning they get their items faster.

The generative AI from Shih’s team is helping companies like Amazon, Gucci, and Williams-Sonoma address operational bottlenecks and improve customer service.

“For example, if you’re a customer service rep, customer calls in expecting a quick answer, instead of having to look through pages and pages of knowledge articles to read the solution, and then let the customer know, generative AI allows us to automate that,” Shih said.

Adding protections now

With the adoption of AI comes some dangers, such as violating data privacy and bias in large language models.

Shih said Salesforce has over 25 years of experience with these concerns, starting with when they put customer data on the cloud. This required the company to find new ways to encrypt data and keep it safe.

In addition to applying those principles to data privacy, Shih is now working on systems to combat misinformation and bias.

“There’s a lot of not trustworthy sources out there. And so we have to compensate for that with really good toxicity filters, citations. Again, going back to the data is so important to ground all the prompts and trusted, reliable truthful data to ensure that the outputs are reliable and accurate,” she said.

Use AI to solve real-world problems

Sharma cautioned against using AI for the sake of seeming modern, and urged leaders to find practical ways to solve problems.

Her team tests the usefulness of problems with their own teams first and gathers feedback before rolling it out.

“That is the what we’re in the business of doing as technologists, as business leaders, as operators, to use AI to solve the most pressing customer problems,” Sharma said. “We’ll succeed together.”

Prioritizing making AI more accessible to employees and customers will increase its usefulness, Shih said.  

“We’re seeing yet another UX revolution come from AI, which is to go from code and clicks to conversation,” she said. “Whether it’s the end consumer who’s doing it that way, or all the employees and partners that you have in your organization.”