Imagine having a new team member who shadows your best salesperson to fetch data and learn unspoken rules, like why one client is more responsive to a direct approach while pitches have to be carefully framed for another. This apprentice never forgets a lesson and shares their nuanced understanding with colleagues. 

That’s the vision of AI that Ari Lehavi, the head of applied AI at Moody’s, is bringing to life, shifting the focus from task automation to capturing and scaling the institutional wisdom that companies are built on. Lehavi shared this idea and more during a fireside chat at From Day One’s December virtual conference

The transformative potential of AI lies in human-AI collaboration based on a continuous, two-way learning street that’s designed to augment human judgment rather than replace it, he told moderator Rebecca Knight, contributing writer at Harvard Business Review. 

Shifting From Automation to Augmentation

AI-doomers often frame the technology as the worst thing that’s happened to job security in human history, but Lehavi sees it more as a collaborative tool that enhances human performance and encourages organizations to do the same. 

“I do think that there’s been some orientation around thinking about AI as a way to generate efficiencies and automation, and I don’t think that’s the best use of AI,” he said. “Increasingly, I’m seeing a shift in the way that companies are thinking about it as an accelerant of performance, rather than as a way to generate efficiencies.”

The central question then becomes how to increase productivity and work quality with AI. Lehavi says one of the ways that organizations can accomplish this is by using AI to handle simple, repetitive tasks, freeing up employees to focus on work that requires uniquely human skills, such as judgment, empathy, and innovation. 

“The hard cases, the edge cases, the complex areas, the mentoring of other people, the management, the development of skills in other individuals, the expansion of what’s possible in their role,” Lehavi added, pointing out what humans excel at. 

The Importance of Bi-directional Design

Lehavi says “bi-directional design” is necessary to optimize human-AI collaboration. Most AI tools used today have a single directional design. You ask questions, and it answers. True partnership requires a feedback loop where humans teach AI context and nuance, he says. 

“AI has information that it can pick up from documents, from data that can help you assemble research faster,” Lehavi said. “But that has a very limited kind of lift that it creates.” The exponential gain happens when AI begins to understand how and why you make decisions. “It has to kind of almost get into your head.”

AI provides value, like summarizing key points from a large text library, in a bi-directionally designed system, but it also identifies gaps in its understanding. It learns to ask questions such as “Why did you make that decision?” 

This leads to humans working with AI, explaining the nuanced instincts that come with experience. Capturing the reasoning behind human decision-making enriches the AI model's understanding, allowing it to provide more insightful recommendations in the future. 

The information learned by the AI can be packaged and shared, creating a “collective organizational wisdom” that other employees can access. 

A Concrete Case: Augmenting the Sales Professional

Lehavi shared an example of how bi-directional communication between humans and AI works in the real world from within Moody’s sales department. A standard CRM stores data, but misses the subtleties that define a veteran sales rep’s success. Insights like the unspoken politics of a client company, the specific pain points a key decision-maker is sensitive to, or the historical context of a relationship. 

Moody’s built a system that starts by giving sales team members AI-generated leads, matching market pain points to the solutions it provides. The AI responds with questions such as. “Tell us what we don’t know, tell us, you know this person,” Lehavi said. “We know the general profile, but we don’t know this particular relationship in this particular instance, and what exactly is the dynamic that would make this deal move faster and closer.”

The seller feeds the nuance context back to the AI, which then refines its recommended messaging and value propositions. The system also identifies patterns in these seller-client relationships and provides recommendations such as: “What you’ve told us about this individual and this company seems a lot like three others that we’ve encountered, and this framing of this message really resonated.” 

The sales team member tests the hypothesis, and the result, positive or negative, is fed back into the AI model, expanding its institutional knowledge. 

Lehavi views AI more as an apprentice than an intern. “Initially, the apprentice gets more value from you than you get from the apprentice,” he said. You invest time teaching the algorithm your ways, then the dynamic eventually flips. “You’re starting to get that much more value. And then you know that you have a true partner, so you can move up to the next level in your career.”

With AI managing more of the administrative burden and research, sellers have more time and mental space to focus on the irreplaceably human aspects of their role: deepening relationships with clients and crafting persuasive value propositions. For leaders, it means scaling the impact of top performers, so other employees benefit from the institutional knowledge they help build. 

The Undocumented Layer of Human Judgment

The critical insight Lehavi stressed throughout the conversation is appreciating the vast, often invisible complexity of most professional roles. He points to what he calls “the undocumented layer of human judgment” that exists in every position, from customer service to legal departments. Studies suggest that around 10% to 40% of what knowledge workers do is based on this tacit understanding.

“Whenever I see enterprise implementations that end up where people kind of feel like they didn’t accomplish what they were supposed to accomplish, I often link that to the underappreciation of how much of the work that gets done is unwritten, and is based on judgment and experience,” Lehavi said.

The routine portions of a job that knowledge workers spend 60% of their time on might be automatable. But the high-value edge duties, where crucial relationships depend on nuanced judgment, are where human-AI collaboration must focus. 

The goal is to design systems that bring the right information and context to the surface to help their human counterparts make faster, more-informed decisions. 

Lehavi advises companies to build systems that ask “why.” AI models that learn from human experience and improve the performance of their human collaborators. This allows organizations to move beyond simply automating tasks with AI, and start codifying, scaling, and institutionalizing their collective knowledge–their most valuable asset. 

Ade Akin covers artificial intelligence, workplace wellness, HR trends, and digital health solutions.

(Photo by KTStock/iStock)