Is AI a totally novel thing? Or is it just the latest in a parade of new technologies to be evaluated and adopted by businesses — one that will face many of the same speed bumps as the cloud, mobile apps, and the internet did before it?
Abbie Lundberg, Editor of the MIT Sloan Management Review, moderated a session our recent Impact conference to explore what it takes to successfully introduce new technology to a large organization. It featured three leaders who’ve spent their careers dragging advanced technology across the gap from demo to deployment: Mikell Taylor of General Motors (previously at Amazon Robotics); Caroline Sherman, Head of Growth Initiatives and Market Intelligence at Fidelity Investments (previously at Tableau/Salesforce), and Deborah DiSanzo, formerly CEO of Best Buy Health and General Manager of IBM Watson Health.
The through-line wasn’t what AI, robotics, and other bleeding edge technologies can do — it was why so much of it never gets deployed, and what separates the innovations that stick from the ones that die in the pilot.
1. Start with the problem, not the technology. All three kept circling the same discipline. Taylor invoked roboticist Joe Jones’s rule: “Anytime you want to build a robot, the first thing you should do is try to figure out how you can avoid having to build a robot.” Sherman, a product person, put it plainly — “it’s amazing how we all do just want to use AI” — when the real question is the customer’s pain point. DiSanzo offered herself as the cautionary tale: she green-lit using AI to detect depression from billing calls, spent nine months, hit high accuracy, then ran into the wall of, “But then what? Why are you doing it?” It was an expensive detour.
2. Being right too early is still being wrong. DiSanzo has now twice tried to create a “cognitive companion” for seniors — Coco in 2016, and a Lively-on-Alexa launch in 2022 — and twice watched the market reject it. “It was an utter failure,” she said. “We sold 400 of them, and for a retailer, 400 is [nothing].” The technology worked; the timing didn’t. With an average user age of 74, “they did not want to talk to Alexa about [needing emergency assistance.] They really wanted to call her.”
3. Design with the people who work next to the machine, not just the buyers. Taylor draws a hard line between customers and users: warehouse associates “are not the customers of the robotic systems, but they are the users of them… they are co-workers to these systems.” To hear them candidly, she goes where the filter comes off — “one of my favorite voice-of-the-customer sources is the Sub-Reddits for the Amazon fulfillment center workers.” Watching that thread evolve from “it’s going to take our jobs” to “they’re kind of cute little guys” tracked how the associates actually experienced the robots, and co-designing with them made the systems safer and more intuitive.
4. The unglamorous “plumbing” is what breaks. DiSanzo’s hospital-at-home program had a sophisticated platform — software, tablets, integrations — and none of that is what failed. “What broke it was the plumbing,” she said. The killer was logistics: getting equipment to the home on time, training families and nurses, then retrieving and reprocessing it after discharge. “We spent all this money on the platform, and what broke it was the logistics of getting it into a home and getting it out,” she said.
5. In change management, protect identity and mind your words. Sherman changed her organization’s values before changing its mission statement, and deliberately kept one existing value — deep domain expertise — in the first slot, so “people felt this attachment and their identity was maintained” while three new values (collaboration, innovation, action) moved in behind it. Her bar isn’t universal buy-in: “you don’t ever get 100% of people excited for change. That’s just not the goal” (she targets 80–90%). DiSanzo told a story that involved changing a single word — a draft OKR read “develop AI to increase caring center efficiency,” which reads as a threat if you are a caring-center agent, so “we took AI out completely,” landing on “develop tools to help caring center agents care for members.”
6. Scaled empathy was the surprise — people actually liked the bot. DiSanzo long treated live humans as the competitive advantage; her caring centers fielded roughly 9 million calls a year from users averaging 74, many of whom just wanted to talk. That didn’t scale, so they rolled out a generative-AI bot through their caring-center software vendor in 2023. It absorbed “40% of the customer service calls… way higher than we thought, and our Net Promoter Score went up,” precisely because it was patient and endlessly available. One caller — 73 and newly blind — told them, “I love calling because they just talk to me. They’ll talk about anything.”
Sherman shared this advice to end the session: “Lead with curiosity, lead with inquiry, so that everyone continues to focus on where the clients really need us, and then how we bring in innovation and AI to support that — rather than the other way around.”
You must be logged in to post a comment.