Yet despite all the potential and interest in AI technology, last year she said that in 50% of situations where generative AI was being considered, the right call ended up being to not move forward with using it. 

In some cases, the idea came too early, creating challenges the company wasn’t ready for, she explained. In others, the cost was too high or business goals could just be hit faster, cheaper, and with more reliability without it. Sometimes, a different opportunity simply won out instead.

“When we’re looking at all of the potential investment opportunities, sometimes generative AI is the best, and sometimes it’s not. Sometimes, our money is better spent elsewhere, and that’s actually what we want to see,” she told Fortune. “We don’t want to treat AI like this special, shiny object. It is a tool, and we want it to be held in comparison with all the other opportunities and ways we could be spending our energy.”

With the pressure to adopt generative AI and a rapidly growing ecosystem of potential AI solutions to choose from, true strategy comes down to being discerning and never assuming AI is the best solution or a guaranteed value add. Speaking with Fortune, Doan detailed the decision-making processes behind occasions where AI ultimately wasn’t pursued and shared advice for business leaders on how to decide when to use—or not use—AI. 

Why three potential AI products didn’t move forward 

One instance where Doan and the team decided not to move forward with a generative-AI-powered use case involved a job with a large retail manufacturer. The company was considering using the technology to build an internal-facing tool for employees that would allow them to access company information. Because it was very important to the company that it would surface data with 100% accuracy, and because hallucination is an unavoidable issue with generative AI, they ultimately decided to use traditional lookup methods instead.

For another company she worked with, a health-tech startup offering AI transcription products, they discovered a compelling market opportunity that AI would allow them to pursue: using the company’s large amounts of data to forecast diagnoses for patients. While they felt it was promising, they realized this would shift the firm to legally be considered a medical device company, thus subjecting it to a whole new level of regulation.

“That just made the cost-benefit analysis not worth it for that particular company. They didn’t want to go down the path of becoming a medical device company,” Doan said.

In another instance, a pet services company she was working with homed in on a market opportunity to do real-time analysis of photos to deliver health insights. But when Doan actually started to understand the total cost of the project, it turned out to be a much bigger initiative than the company could take on.

“What I found upon further investigation was the cost of real-time processing in delivering this consumer experience at scale would take us into a new price point where we would actually have to either upcharge, change our subscription model, or find some additional revenue angle to pay for those high costs, making it no longer financially viable,” Doan said.

In this case, Doan and team decided to reassess in three to six months, showing that the decision not to pursue an AI feature or service is never final. Sometimes, it’s just not the right time, and waiting is what will make it viable. 

“If the technology costs have come down, or if we’re also redoing our pricing and packaging, that could be a great opportunity for us to potentially layer in this new feature set,” she said. 

How to decide if AI is a no or a go

According to Doan, decisions around tapping AI all stem back to some very fundamental questions: What are the problems you’re trying to solve? What are you ultimately trying to deliver on for your customers?

These questions should guide business leaders at the beginning, and ideas should always be gut-checked against them. As for deciding how to actually execute, she emphasized that business leaders should lean on their technical experts to weigh in on the best technologies and approaches.  

“All of us have access to consumer-facing AI and LLMs, but that is fundamentally different from the expertise of infrastructure engineers, AI engineers, and AI product managers,” she said. “The expertise they bring to a conversation will help you understand the cost, the potential upsides, and the potential downsides at a much deeper level of understanding.”

Lastly, it’s also vital to make sure your tech stack is sufficient and prepared for the AI use case, such as ensuring the foundational data you plan to tap is cleaned up and ready to have AI layered on top of it. This is the reason another potential AI initiative she was working on last year didn’t move forward: The company’s data just wasn’t ready for it. Doan suggests that when building out a road map, the order in which you tackle initiatives is equally as important as what you decide to pursue.

“Especially when we think about user-facing experiences—they build on top of one another,” she said. “So you really do want to think about that sequencing. When it comes to AI, there can be foundational pieces that you need to get in place first, because then, when you layer AI on top, it’s going to have this kind of compound effect, as opposed to just looking at those opportunities in isolation.”

Death of the first-mover advantage—and the risk of moving too fast 

With the rapid pace of AI development and implementation—not to mention, the constant messaging to “adopt AI or get left behind”—the pressure can feel very real. From Doan’s position developing AI products across industries, however, she’s seeing less and less of a reason to rush.

Because development cycles in engineering are generally getting shorter, it’s now possible to put out features far faster than ever before, she said. Now, if you see something that surprises you in the market and you want to react to it, you can move quite quickly to develop a competing offering within your own company.

“Unless you are an AI company—like, that is the thing you are producing—I actually don’t think there’s as much pressure to be in that first-mover kind of category as we’ve had … in previous years,” Doan said.

It’s also important to consider the risks of moving too fast. Customer satisfaction, trust in the products—these things are hard-earned but easy to lose if a product or feature is launched prematurely, leads to a price increase, or otherwise damages the customer experience.  

“The companies that I work with have worked really hard to build up their brand reputation and the relationships they have with their consumers,” Doan said. “And we all know that the cost of losing a customer and needing to go out and acquire a new customer is really expensive.”