How to Fail in Front of Your Executives and Get Invited Back
If you're asking me for help, it's because you've tried everything you know how to do and you're stuck.
Head of Commercial, Radiant AI
The demo was in Italy, and it went badly.
Ben Cappellacci had given a lot of them by that point. At Palantir he was the person executives called in when a prospective customer or an investor needed to see the software properly. “The purpose of a demo is to help tell a story,” he says. “It tells the story of business value. It tells our story about our philosophy in terms of how we think about the world.”
None of that landed. The customer did not understand what the product did. They did not think it was a fit. And the way he had explained it was not how the executives thought it should be explained.
Then his boss emailed him about it.
“This was pretty bad,” he says. “I didn’t know how to react. It was like, oh, I made a mistake here. But what could I do to make this better?”
What he did next was not work harder.
He went to senior people at Palantir he respected and asked them for an awkward favour. Watch me give this demo. Let me explain why I pitched it this way, to this particular customer. Then tell me how to do it better.
“In the end, I was invited back to do many more demos,” he says. “It was a really good learning for me, to draw on the wisdom of other people within the company who I respected. Basically, I had miscalibrated my understanding of what this demo was supposed to do.”
Miscalibrated. Not unskilled, not unprepared. He had solved the wrong problem competently, and more effort would never have shown him that. Only someone watching him do it could.
Your boss is the wrong person to ask
Cappellacci is specific about where advice should come from, and it is not up the reporting line.
“They tend not to be my bosses or people that are directly responsible for my day to day stuff,” he says of the people he learns from. “Your boss has goals in mind in terms of accomplishing targets and projects, and those aren’t always aligned with your overall growth as a person.”
He treats this as structural rather than personal. A manager who needs something shipped by Thursday is badly placed to tell you to spend six months getting worse at your job so you can get better at it later. The conflict comes with the role.
“It’s important to find someone that’s adjacent to you, or a peer, that can give you that advice from an external perspective, that doesn’t necessarily have objectives that you need to reach for them.”
What he wants from those people is not encouragement. He looks for people “who have perspectives different from mine, who understand the goals that I have, and are able to question the assumptions that I’m making.”
Make asking for help mean something
The best thing about Palantir, he says, was a norm most companies never manage to install.
“You could go on Slack and text almost anyone in the whole company, even the number two guy, and they would usually get back to you fairly quickly with some kind of a solution.”
It worked because of the assumption sitting underneath it.
“If you’re asking me for help, it’s because you’ve tried everything you know how to do and you’re stuck. So if I know how to do something, I’ll give my time to you.”
That inverts the usual reading. In most companies a request for help sounds like someone admitting they cannot cope, so people put off making it. At Palantir it meant the opposite. Whoever was asking had already run out of their own options, so the request arrived pre-filtered, which is why answering it was worth a senior person’s time.
Tell people who benefits
Ask him what motivates a team and deadlines never come up.
“I think a lot in our world, we do things because we’re supposed to do things, not because we believe those things are important,” he says. “As a leader, part of your job is to explain the value of why something is important to the team.”
His example is specific. On a project for the National Cancer Institute, he did not tell his engineers the ship date. He took them to hear from a scientist who was using what they had built.
“They’re like, yeah, this has been amazing for me. I used to have to wait weeks for my research to come back. Now I can do it in hours.”
Engineers with no particular interest in oncology came away understanding what they were doing and who it was for.
“It’s like, why? For who? What will happen when we do?”
The same thinking shapes how he builds teams. Asked what he is proud of, he mostly declines the question. “It’s difficult to take credit for anything in a company that has a lot of really smart people.” His job as a manager, as he puts it, is to “find what people love to do and really create an environment for them to do that thing”. He wants a broad mix of skills with each person placed where they are strongest, and says plainly that he is not building a team around a star.
Ask what one wrong answer would cost you
Cappellacci is now Head of Commercial at Radiant AI, which helps engineers spot generative AI systems misbehaving in production. The problem has an awkward shape. A language model always returns an answer, including when the answer is badly wrong.
Out of that came a question worth borrowing, whatever you are building.
“How sensitive is your project? And if there’s one bad response, how dangerous could that be for you?” he says. “Trying to understand that can help you actually pick what kinds of things to do with generative AI products.”
It explains why the deployments he sees working are often dull ones. Sorting thousands of inbound support emails into categories, for instance. “If it gets one or two of the categorisations wrong, that’s not the end of the world.”
On the noise around AI generally he is calm, having watched a version of this before. “A few years ago, big data was the big hype thing, and every company needed a big data strategy. We’re seeing an echo of that.”
He is not calling it a bubble. “It doesn’t mean that it’s a failure. It just means people have to level set their expectations of what actually matters.”
-
№ 01When something you did publicly goes wrong, ask two or three senior people you respect to watch you do it again and tell you what they saw.
-
№ 02Pick your mentors from outside your reporting line. Your boss is measured on what you ship this quarter, which puts them in a real conflict with your long term growth.
-
№ 03Set the expectation that anyone asking for help has already tried everything they know. Senior people answer much faster when the ask arrives pre-filtered.
-
№ 04Teams work harder when they know who benefits. Take them to meet the person using what they built instead of telling them the deadline.
-
№ 05Before starting an AI project, work out what a single wrong answer would cost you. Build the things where the honest answer is "not much".
Ben Cappellacci
Ben Cappellacci is Head of Commercial at Radiant AI, which builds tools that help engineers find when generative AI systems misbehave in production. Before that he was at Palantir, where he gave demos to prospective customers and investors and worked on projects including one for the National Cancer Institute.