Beyond the Lab: Innovation at Scale | Dreamforce 2026

Thank you so much to my esteemed panelists for joining us this afternoon.

Thank you. Thank you.

Let's jump right into it. Every company obviously wants to innovate. The harder question today is how to make innovation a repeatable capability, especially inside global sprawling organizations, often with thousands of people, established processes, and existing ways of working. AI today is making that question all the more urgent, as many of you can attest to, and certainly my panelists here as well. It's asking organizations to do more than adopt new technology. It's forcing them to continually rethink how they work, how they build, and how new ideas move from experimentation to real-world impact. Today, I'm joined by 2 people who think about that challenge from very different but also very complementary perspectives. Professor Linda Hill, to my far left, is faculty chair of the Leadership Initiative at Harvard Business School and one of the world's leading experts on leadership and innovation. She has been recognized by Thinkers50 as one of the world's top management thinkers and is the author of numerous award-winning articles and books, including her latest, Genius at Scale, which I started this past weekend and I'm very interested in diving in further.

Oh, thank you.

Itay Asayo is VP and head of strategy and incubation for Salesforce AI Research. He has been at the forefront of Salesforce's AI transformation for roughly 11 years now, just over a decade, helping turn what's possible at the frontier into innovations that can create value for Salesforce and its customers. So today we're going beyond the lab. How do you create the conditions for breakthrough ideas, moving those ideas to a large organization and building a culture that can keep doing it again and again? Linda, Yitai, welcome. Thank you so much for joining us yet again.

Thank you.

Thank you.

Let's start with what it actually takes for an idea to travel through an organization. I think many of us can attest to the fact that you can have a singular brilliant team with a singular brilliant idea or even many brilliant ideas, but it's very hard to scale that across an organization. Linda, what are the conditions that leaders need to create for new ideas to develop and to survive as they move through an organization?

So, pleasure to be here with all of you, and I must go back a little bit in history.

Yeah.

So, in the internet boom, I was sent out here by my institution to understand what it takes to lead innovation. And the first organization I studied was Pixar, and have had fun ever since. So, I came out here to look at Pixar and eBay and Google, and what, what are you doing that allows you to innovate so much? And I just went and spent Monday at Pixar again and was inspired. But I have since then, I just want to say, been looking at this connection between leadership and innovation for now. It's hard for me to believe over 25 years. And so I study organizations all around the world. And very surprising to me, whether I'm at Pixar or I'm at Cleveland Clinic Abu Dhabi or I'm at MasterCard or wherever it might be, the leaders have the same mindset. So I have to— explain that quickly. So the mindset they have that I first heard from Ed Catmull was that leading innovation is different from leading change. And I got to keep that in mind as we're talking through all of this. Leading change is about having a vision, communicating that vision, inspiring people to want to fulfill that vision. Leading innovation is different. It is not about followership. It's about getting people to co-create the future with you, not follow you to the future. And there are 3 muscles you have to build if you want to have an organization that can innovate time and Again, you need to be able to collaborate, you need to be able to experiment, you need to be able to learn together and collaborate across difference, which is one of the things that makes it very difficult. So the conditions you see these leaders do is they think about what are the culture, what are the capabilities, what are the tools people need to be able to collaborate across difference, to be able to experiment and learn in an iterative way as efficiently and as effectively as they possibly can. That is what they worry about. Not so much, do I have a vision? So those are the conditions. And with AI, to be perfectly honest, none of us really know what the future's gonna look like. So again, it's how do we create conditions that people will wanna go on that journey with us to figure out what we want that destination to be and how we're gonna get there? So that experimentation and learning is really a critical piece of the puzzle.

Similar question for you, and I'll ask you the same thing. Make it real for us. As you look at your team, you're really at the crux of this AI's evolution and experimentation in that regard. How are you ensuring that you're creating that environment for your team to not only experiment and play around with AI, but really have it scale through the company?

Yeah, and it's very good to be here. So as you mentioned, I'm with our AI research group, and it's actually, this year is the 10th anniversary of AI research at Salesforce. So we were founded in 2016. And throughout those years, we've really contributed a lot to both the academic nature and the scientific community as well as to the products. And it's interesting because, you know, if you look 10 years ago, we were doing a lot of AI that was predictive AI. It was, you know, wasn't as kind of in the forefront as it is today.

Yeah.

And so we were able to really, in a way, be a bit more insular, like focus on foundational research, go really deep, And something has happened over the last few years, especially since in 2022, at the end of 2022, the kind of ChatGPT moment, and it brought AI into the consciousness. And what happened is also the landscape really just accelerated. And so the expectations of bringing innovation into a company like Salesforce has also, you know, those expectations have also changed. Now, You're here at Dreamforce. Hopefully you're enjoying your time here. It's a massive conference, right? I think there's like 57,000 people here. Salesforce itself and the whole kind of ecosystem around it is huge. So it is not an easy task to actually bring something from foundational technology to something that's that our customers and people in our ecosystem actually leverages and uses. And so this is really kind of a big part of what we do at AI Research. We don't— not only do we look at the foundational technology, how do we progress the frontier, but how do we connect that to the customer needs?

Yeah.

How do we connect that to the different product teams? And how do we make sure that people are actually connected and understanding, okay, this is a technology that can actually make some real impact for my customer, for this industry. We have so many different types of industries, types of customers. So it is really by being able to experiment and connect the dots that we're kind of bringing that to the forefront.

So let's double-click there because when we talk about innovation, oftentimes that term can feel quite nebulous and I wanna make that very tangible. As you're talking about how you serve as a bridge and you're connecting that customer needs. What does that look like in practice? How do you go about doing just that?

Right. And so as I mentioned before, our AI research team used to kind of focus a lot more on the foundational research, like researchers that are doing kind of, you know, a lot of papers and patents and stuff like that. But over the past couple of years, we've really shifted. And actually earlier this year, we've launched what we call the AI Foundry. The AI Foundry is really a kind of a way for us to connect foundational research with customers. And so we're actually experimenting and creating proof of concepts together with customers. A great example of that is the work that we've been doing with SharkNinja. SharkNinja is an amazing company. They're doing all these consumer products for the kitchen, for the living room. And one of their challenges is to help people actually set up the product, get, you know, actually register it, know how to use it. And sometimes, you know, having instructions by text is not enough. We've actually created an agent for that called the Unboxing Agent that helps people do that with text. But what we've started doing with them is to see, can we actually have the AI look at the machine, look at the product, and help you with video and audio to actually, you know, help you through the setup and troubleshooting and others. And, you know, by doing that, by working directly with the customer, addressing the challenge, applying technology, in this case video reasoning. So we're creating video reasoning harnesses for that. We're able to very quickly iterate and see, Does this actually add value? And then we can quickly either, you know, address it and bring it to the product. If it's maybe it's not exactly the right fit, we can pivot or we can decide, okay, maybe this is not the right thing to actually go after.

That's a great under-the-hood look to how you're thinking about innovation, certainly on your team. Looking at it from an outsider perspective, Linda, you know better than anyone, you know, creating these conditions of innovation often means giving people room to experiment, to learn, certainly to change course. And Linda, you've had the chance to see some of that up close at Salesforce, and you spent some time with Yitai's AI research team at Salesforce. What stood out to you about how they work, and what does it tell us about the conditions that organizations need to innovate at that frontier? Give us that insider look.

So I would say that one of the things that we have been studying lately Is how do you make sure you not only generate innovative solutions, but you actually scale them?

Mm-hmm.

You actually launch and commercialize them. And so one of the things, and I'm going to embarrass Yitai, is you need people who are bridgers.

Yeah.

People who can actually talk across the functions. So you need someone who knows the technology well enough to understand what's possible and what the risks are. You need the person, the people who know the customer. and what needs to go on there. And then you need the people who know the production and the products rather. And so what we often see, and no matter what we look at here, almost all of our organizations become very siloed because of the need to specialization. So going back to this issue of how do we collaborate across our differences?

Mm-hmm.

Etai is someone who can actually translate across those 3 worlds and have them figure out, you know, what experiments should we run and how should we do it? And then how are we gonna actually move it into production to make that product be what it's gonna be? So 2 of the organizations I work with, they discovered in their labs, and I've studied about 14 labs, both in government and private sector and not-for-profit. And often what they find they have to do is when you're ready to move the product, the innovative solution out to be, you know, to be produced so it can be launched, they often have to send the people with the product because in fact you're going to get resistance They're not going to want to do this because they have different KPIs or OKRs than you're going to do. So literally, in fact, the man who was running the labs of Anthropic said, when the product graduates, the team that created it graduates. So they're always having to renew who's in the lab because really it's human. People don't want to take risks. People don't want to experiment and potentially fail with someone they don't know. Or they don't trust, they don't have some sense of mutual expectations. And what we often see goes wrong is you don't have the human, literally the human connection. What AI is helping with that I've seen thus far is a single source of truth. So at least we all have kind of agreed that this is the truth we're working from. But there doesn't seem to be yet a substitute for me thinking, you know, Linda's gonna let me have influence on this. She's not gonna try to shove this down my throat because it's come down from on high that we're supposed to do this. So I found it fascinating that you often have to send the person to the foreign land, if you will, to play that role of being the connection across those 3 worlds of product, science, techno, whatever the innovation is. And finally, you know, the people who have to make also the pricing. So it turns out finance, some financial group, they can't figure out what the pricing should be or how we should work with this. They're going to be the resistors. So the final thing I would say for all of you is that too often we think about innovation as being, there are the people over here who come up with the solutions, and then there's the rest of us.

Mm-hmm.

If you set up your organization that way, you are always going to have trouble scaling. In the organizations that are scaling about as quickly as you can, they actually have a belief that everyone in the organization needs to understand there's nothing called business as usual, nothing called business as usual, and we all have to be innovative problem solvers. So all of us have to create values for customers, create value for customers, and we all have to be game changers for customers, which is why they work on the culture and building those muscles everywhere in the organization, not just in that innovation group. But you often have to send— you have to be the one who goes over to the other side and have the conversations.

Yeah, that's right. And I think a good example of that is actually— so, so sometimes you need to really work hard at translating what we're doing in the lab and kind of like, you know, the very complicated things that are actually very hard to explain, right? Foundational research can actually be pretty technical.

Yeah.

And so one way I think, as an example, something that's happened over this past year is a slightly different approach. We have a model that we came up with in 2023 called MORAE. It's a time series model. I'm not going to go into the details of what a time series model is and how it's different from an LLM, but I'll tell you that a time series model is a model for predictions. And it's been kind of the leading, the industry-leading model since 2023. We've been developing it. And the interesting thing is that by having it addressed and embraced by the scientific community, It's been downloaded on Hugging Face for over 30 million times. It helped also influence our internal teams. It was like, wow, this team has actually made some real impact, and we're seeing that outside externally. And it really helped galvanize a lot of the support that we're seeing now across the teams. And we're working with our Data360 team, and we're working with our field service team, and we're with our service team. So a lot of teams are starting to adopt it, and it really helped to have that kind of external validation for us in order for us to connect internally.

Yeah, because I think you have to build credibility, right? So with all of these, you know, is this a credible person who's trying to ask us to do something that's going to disrupt our work? And so too often, if you tell other people, if you're the only people who innovate and I'm not supposed to be an innovator, then I'm going to resist you. But if you tell me I'm supposed to be as innovative as anybody— so I mentioned Cleveland Clinic Abu Dhabi because they gave them— and this is true, so this is not a sales pitch— they introduced the nurses to Tableau so that they could understand probability and see it in a way that they could visualize. Within really months, the nurses, about 400 nurses, were running A/B experiments to improve their practice on their particular wards. So much so that they had to step back and say, well, wait a minute, too much innovation happening here. We need to prioritize this a bit. But once people, people know, have a problem that they want solved and you give them a tool that, and they can understand what it is, they'll, you can see, use it. And they begin to refine it very quickly. And they push back on you all about other things they wanted from Tableau. I do remember that. They said, we Oh, you know people at Salesforce? Tell them we want this. Why is this there? Or whatever. So I was like, okay, I'll bring that message back because I, I kind of know some people. I don't know that you changed it, but you see that they're so engaged. And when you have people in the organization actually calling the people in the labs, or— of course you don't want them to disrupt the work, but that means you're actually doing something right. That's not a problem. That's a sign of— Success. They want this and they want it to be better because they find it useful.

We've been stressing for the bulk of this conversation the importance of building that muscle of innovation. However, especially at the top, leaders still have to answer to boards. They have to answer to shareholders. Teams still have day-to-day quotas to hit. They have KPIs and other success metrics and a business to run. So how do you align all of the incentives? Because it sounds very easy to say, think more innovatively, be more innovative. How do you actually embed that into a company and ensure that you're aligning incentives?

So from a Salesforce perspective, and in order to have that traction, you know, we need to be very careful of kind of not just having, you know, blue sky ideas that are not connected to the business.

Yeah.

So Celia Savarese is our chief scientist, and he kind of helps to define the vision and the direction of our AI research team. And so we have this vision of where we want to go probably over the next, you know, 12, 18, 24 months. But that helps us actually ground it in what we call big bets. The big bets are like the actual kind of lanes of how we're actually going to get there.

Wow.

And at Salesforce, I don't know if you've heard this yet, but we have this called— this thing called the V2MOP. This is like our goals and metrics way that we measure everything. It stands for vision, value, Methods, obstacles, and measures. And so our big bets are actually organized against those V2MOMs so that we can not only go towards the vision of the overall kind of like, okay, this is our long-term vision, but have the steps that we're taking towards there and have that defined and measured in a very specific way. So it's— this is like the sense, and I'm curious actually, Linda, if you're seeing this in other—

I'm not.

organizations and how you kind of make that connection between the longer-term vision and the steps that it takes to get there that are more connected directly to the immediate KPIs of the company.

Because most people are focused on their short-term wins.

Yeah. No, most of, you know, it's funny, I was on the boards of 2, you know, Fortune 100 kinds of companies, big companies, and I left those boards to try to do startup boards, which has just been a different experience for me. But still, you have the short-term and the long-term. There's always pressure. I think that— so some of the leaders I've studied have been, you know, CEOs and very senior. Some have been more middle managers and individual contributors. So the levers you actually have access to, to create those conditions vary. But at Procter& Gamble, when they ended up with an activist on their board— I don't know if there's anyone from Procter& Gamble here. One of the things they decided to do was innovate on how they innovate. And Procter& Gamble, huge complex organization, and they had to get faster. So the head of innovation and research, she said, you know what, we're no longer in a product economy, we're in an experience economy. We've been a product company, but we have to change our culture to do the end-to-end, which means it's not just R&D, it's not just innovation lab. So what she did actually with obviously the C-suite, I should say, not just her, they went to the heads of the business units and said, how should we measure your innovation efforts? Who should know better than you what percentage of your compensation should be based on innovation, the new, and what your percentage should be based on whatever? And the new could be about your core business, about diversifying your business or building new businesses. But you tell us, you tell me and the C-suite, how should you be compensated? What percentage of your pay? Now think about that. If your, if your senior bosses came to you and said, what percentage of your pay should be based on innovation? You do the argument, you come to us, you tell us, and then I will go with you to the board and you will tell the board. And guess what? The only thing I'm going to require is next year you're going to have to come back and tell the board how you did. Now, you might imagine that some people lowballed it a bit, other people maybe stretched it, but you said it should be based on your business.

Mm-hmm.

And that sort of— and once they understood and got used to that, then, you know, now they have metrics that they put in place and they— but they had to learn because they actually didn't know what the right answer was. So that was— but as she said, that's not really all that risky because who should know better than the head of the business? who's got to collect the information from the frontline. What's not working? How much should we do? So I actually thought that was one of the— I don't know if it's bravest because I don't think it's that brave because who— and the corporate doesn't know as well as the head of the business what needs. But they actually had to collaborate about that to figure out what the pay was.

Yeah.

So I think that we're seeing different ways that— and also actually at MasterCard, when I studied MasterCard, Ajay Banga, who's now the head of the World Bank, was the CEO. And if any of you know the MasterCard story, unbelievable growth in their market cap as they became more of a tech company. He literally said to the company, don't bring me a budget unless 50% of your resource is going to the core and 50% is going to diversify and new. So if you bring me a budget that doesn't reflect that, if you can't tell me you're spending your time that way because we're going to be disrupted by fintech.

Mm-hmm.

So You know, again, having someone at the top who's sort of like, he's got to be kidding. No, we're going to die.

Right.

You don't— maybe don't see it. You don't see that urgency, but we are. So often at the top can set the stage. But that's, that's to try to force people to— he had to talk. I'm sure he had many conversations with the board. I'm sure they had many conversations at Procter& Gamble with the board. But these are very established businesses, regulated, etc., saying, We're dead if we don't learn how to use these tools because they're so critical to our success.

Really about agility. We were talking about this a bit in the green room. Well, first of all, a lot of promising ideas, it's very easy for them to lose momentum, especially in large organizations. But on the flip side, you often see ideas that are kind of middling, they're not really going anywhere, and they're allowed to continue to foster and foment without them really coming to fruition. So how do you navigate that tension? How do you ensure that giving the best ideas, the most promising ideas a real chance, and on the flip side, not overinvesting or taking too long on ideas that aren't proving their worth, so to speak?

Yeah, so I think this is something that we actually work with and work through all the time. Working with— so I mentioned already our big bets, our different lanes that we have, and we have different experiments, different proof of concept, different pilots that we're working through. And it does require a lot of, you know, in a way, bravery to admit, okay, this is not working. If you're testing it, you're testing it with whether it's with customers or internal groups and look at the data and see like, okay, this is not working. What should we do next? Should we actually pivot this? Should we actually discontinue and prioritize something else?

But it's always data-driven.

And it's data— well, it's data-driven. I wouldn't say only. I would say that it's mostly data-driven because you also need to have a certain amount of imagination.

Yeah.

It is, you know, if it's true innovation, it's something that hasn't really been done before. Are we actually measuring it in the right way? And so another element that goes into this is momentum. Having momentum is incredibly important. And I think I look at momentum almost as a KPI on its own. Like, do you have momentum for this? Do we have other teams customers that are interested. When you are presenting this to people, do they get excited?

Yeah.

And so, that momentum is another kind of— it's a soft measure of understanding, like, does it have— is it keeping going? And to your point, though, if, you know, if the momentum kind of slows down, if it meanders and becomes something like, okay, there's another version and another version and we're not really seeing progress, you do need to kind of Like with, okay, do we need to prioritize something else? Do we need to actually kill it? And I know that you have some interesting stories about that.

Yeah, Linda, you're nodding your head.

I feel like you have something to say. So I'm an anthropologist, and I— so I'm nosy. I look at all of you all and see what's going on with what you're happening. So one of the things that I have seen in most organizations is if you want to innovate and if it's an experiment, the idea is if you learn something, it's okay. But that's not the way we experience when an experiment fails. So if I could go back to Procter& Gamble, when they put in what they called lean innovation, which was sort of lean startup methods, the first thing they had to say, and they really had to work at it, is fall in love with the problem, not fall in love with your solution. Because when you have very capable people who are, you know, Type A's, whatever, they fell in love with their solution, not the problem of the customer. So that mantra had to get around. But what I found in most organizations is they could not kill projects that needed to be killed because mostly— and I hate to say this, I know I'm at Salesforce— but people don't simply rely on data. Not that you necessarily should. People don't rely on data as much as we'd like to think in making their decisions. And emotion is a big piece of that. So in one organization, actually at MasterCard, Ajay told people, if you kill your own experiment, I will give you a bonus. It was so hard to get people to kill things because it looked like it reflected on them and might impact their— you get a bonus if you kill it yourself. If I have to kill it, other things are going to happen. But you kill it. The other organization that I was working with also had very much— they were stuck. And what happened was the middle managers came to the senior management and said, you keep telling us to cut our budgets, but if we could kill some of these things that we all know don't work, but none of us want to hurt each other's feelings. They were just too polite. There wasn't enough. Then we wouldn't have to cut our budgets. So we're not going to cut our budgets until like a certain percentage of these things are killed. And again, it's— I think people don't like conflict. People don't like to hurt other people's feelings. So going back to collaboration, if you cannot debate and really speak honestly— so I'm not surprised about all this work on psychological safety. People don't say what they really think. And if you can't do that, you can't make decisions in a timely way. You can't kill things when experiments have told you. And sometimes in the end, it is about boldness saying, we just don't have the right data yet. And I still think this is promising and we need to persevere. Or you never throw anything out. You save the learning. Because again, the other thing is organizations, it's hard to learn. I'm in an academic setting. Do we learn across and share with each other enough? Absolutely not. The dean is right about that. So collective learning. One thing I will tell you about GenAI and all of this, too often the training is too individual. You should do the training in intact teams so that the learning happens. across the team and they have imagination about how the work can be done differently because you're actually working with a real group that you work with, as opposed to me individually learning and acting like I'm going to pass on to you what my experience was.

We're running up on time and I want to give all the leaders here in this room something tangible to take back to their respective organization. So let's end with this. What's one thing that you would tell a leader in this room to do starting Monday, not, not 5 years from now, starting next week Monday, in order to create room for innovation, in order to cultivate an environment of innovative thinking and doing?

He does it more than I do it. I get to observe all kinds of people. So I don't know, maybe you should have the last word. Yeah, actually, where my head is now, I'm having a little trouble hearing here, is you need to look at your own capacity to manage conflict.

Why?

Because if you're not good at managing conflict, embracing those differences and helping constructively do that, sort of be the bridger, which is one of the roles that we write about in our book, you don't know how to help with that. I would say look at yourself because your leadership really does matter. And if you're not able to embrace, embrace and work through difference, work on those skills, have a coach that helps you with that, those micro skills, you will do a real service to your organization.

Please, final word.

So for me, and this is something that I tell my team a lot as well, I have a couple of things. The first one, which I think is really important, and sometimes it baffles me why people are not doing it more, is just simply show your work. Share what you're working on, not when it's final, not when it's ready, but when it's raw, when it still has mistakes, when it's still kind of in the process. And, you know, share it with people across different teams, across your own team. It will help you not just necessarily even to get feedback, but when you share something with somebody else, it helps you see it from their eyes. If you just try it, right? You just, like, just by seeing it, you don't even need to get feedback. Just by doing it, you get that. And so, it starts this kind of iterative process. And the second thing, if I can have a second—

Please.

Is diversity of thought. So, if you're just having and sharing that or kind of working on it with a group that they're all kind of like coming from it from the same angle, you're not really going to get all the different point of views that you need in order to really progress it forward. So, having a diverse set of people that work together is probably one of the most important things.

I think that's the perfect future-looking place to end. Linda Hill, Itay Haseo, thank you so much for your time. Thank you everyone for joining us and enjoy the rest of your conference.

Thank you.