4 minute read<\/p>\n
How can product managers leverage AI for human-centric product design? Expert Cal Al-Dhubaib talked about this important topic and shared some tips for product professionals.\u00a0<\/em><\/p>\n <\/p>\n AI is no longer on the sidelines of product development. It’s embedded in how we build, scale, and deliver value. In 2024, <\/span>U.S. private AI investment surged to $109.1 billion<\/span><\/a>, and the number of <\/span>FDA-approved AI-enabled medical devices reached 950, up from just six a decade ago. <\/span><\/a>These shifts aren’t theoretical. They’re reshaping expectations for how modern product professionals operate.<\/span><\/p>\n But<\/span> while AI’s technical capabilities are advancing <\/span>quickly<\/span>, product managers face a <\/span>different<\/span> challenge: integrating AI into products in a <\/span>way<\/span> that feels useful, ethical, and trustworthy to real users.<\/span><\/p>\n This<\/span> is the heart of human-centric product design, not replacing humans with automation, but designing AI features that support and empower them. As Cal Al-Dhubaib, a leading data scientist and <\/span>entrepreneur who launched Pandata,<\/span><\/a> puts it, “There’s this misperception that it’s AI or human. One of our core beliefs <\/span>is, it’s<\/span> a mixture of both.”\u00a0<\/span><\/p>\n Keep reading to learn about this topic as recently discussed on our podcast or listen to the discussion here.<\/a><\/p>\n For product managers, the most valuable AI isn’t the most advanced. It’s the most usable. <\/span>That<\/span> means building tools that <\/span>fit naturally<\/span> into workflows, support decision-making, and <\/span>allow for<\/span> human oversight when <\/span>needed<\/span>. This calls for organizations to not see AI as a trend<\/a> and focus on including it where it actual adds value, not to just check a box.\u00a0<\/span><\/p>\n Cal frames this shift through an umbrella term he identifies as “generative experiences.” These are interfaces that allow users to interact with AI intuitively, without forcing them into unnatural or rigid processes. What does this look like in everyday life?\u00a0<\/span><\/p>\n “Imagine your inbox with a summary bar along the side telling you action items requested of you from each email, and that you can then say, “Who is that individual?” or “Remind me some information about what their role is?” or you can dig for information <\/span>in a more intuitive way<\/span>.<\/span><\/p>\n In this context, AI hasn’t <\/span>been tacked<\/span> onto a product, but the value it could add to the user experience has <\/span>been considered<\/span> first. <\/span>It’s natural and <\/span>doesn’t require context switching<\/span><\/a> and<\/span> it <\/span>is<\/span> a more human-centered inclusion of AI into a product.<\/span><\/p>\n In other words, the best AI doesn’t lead. It assists.<\/span><\/p>\n A fundamental principle Cal shares is deceptively simple: Start with what humans are already doing well.<\/span><\/p>\n “When humans can’t agree, the model can’t converge,” he explains. That’s because AI relies on consistent patterns and labeled examples. If a task is ambiguous or disputed among human experts, an AI model won’t find stable ground to learn from.<\/span><\/p>\n That’s why one of the first questions Cal asks when advising teams is, <\/span>“What are humans doing today?”<\/span><\/em> And if the response is, <\/span>“Well, we couldn’t, but AI can,”<\/span><\/em> that’s a major red flag.<\/span><\/p>\n “There are very few situations where that <\/span>actually<\/span> makes sense,<\/span>“<\/span> he explains. Product professionals need to be skeptical of AI solutions designed to fill gaps where no solid human process exists. In those cases, the problem is often too ill-defined for a model to solve effectively.<\/span><\/p>\n Strong candidates for AI include:<\/span><\/p>\n Red flags for product managers:<\/span><\/p>\n If the product team hasn’t built or trusted a repeatable human process, AI won’t be able to either.<\/span><\/p>\n AI should <\/span>be designed<\/span> to intervene at the right moment, with the right level of confidence, and in <\/span>a way that’s actionable<\/span>. One of Cal’s examples from healthcare contrasts two models: on decisions with high accuracy but delayed availability, and another with lower accuracy but available at the point of decision.<\/span><\/p>\n The team didn’t dismiss the lower-accuracy model. Instead, they mapped where the model was most likely to be right and used it to trigger selective alerts. Where confidence was low, it triggered no alerts. Where confidence was high, <\/span>alerts were triggered<\/span>. <\/span>This<\/span> helped clinicians <\/span>to better focus their efforts<\/span> and more appropriately consider treatment paths. This system also prevented “alert fatigue”.\u00a0<\/span><\/p>\n “We’re not trying to make AI the decision-maker. We’re flagging points where humans may want to re-look.<\/span>“<\/span> This inclusion of AI in the workflow treats AI as an assistant to, <\/span>not<\/span> a replacement for, human discernment, insight <\/span>and<\/span> decision making<\/span>.<\/span><\/p>\n Product management takeaways:<\/span><\/p>\n When AI <\/span>is framed<\/span> as a co-pilot, <\/span>not<\/span> a controller, users tend to trust it more and <\/span>use<\/span> it more.<\/span><\/p>\n AI will get things wrong. <\/span>Google admits that its AI overviews <\/span><\/a>produce “odd and inaccurate<\/span>“<\/span> results <\/span>and<\/span> a <\/span>study found that around 52% of <\/span>questions about programming<\/span> were answered incorrectly by ChatGPT<\/span><\/a>.<\/span><\/p>\n Cal agrees that AI will be wrong some amount of time and reframes the conversation, “It isn’t about, can we get AI to work on this problem? <\/span>It’s<\/span> can<\/span> we get it to be useful in the context of this problem?”<\/span><\/p>\n And further<\/span>, the fact that AI <\/span>gets things wrong<\/span> is not a problem if <\/span>you’re<\/span> prepared<\/span> for it.<\/span> It’s only a problem if you aren’t.<\/span><\/p>\n “Have a good answer to ‘What happens when it’s wrong?<\/span>‘”<\/span> Cal advises. Product managers need to plan for AI’s edge cases just like they plan for UX gaps or system outages.<\/span><\/p>\n Build these into your product design:<\/span><\/p>\nAI Is a Tool. So Design It Like One<\/span><\/strong><\/h3>\n
Use AI Where Product Management Structure Already Exists\u00a0<\/span><\/strong><\/h3>\n
\n
\n
Integrate AI Seamlessly into Product Workflows<\/span><\/strong><\/h3>\n
\n
Build Safeguards and Metrics into AI-Driven Products<\/span><\/strong><\/h3>\n