Speaker A
[music] I'm excited to be joined here today by Aaron Cannon, co-founder and CEO of Outset. Outset is the leading AI customer research platform used by some of the biggest brands in the world like Google, Microsoft, and Nestle.
Aaron Cannon discusses how Outset created AI-led customer interviews, pioneering a new research category with major brands like Google and Microsoft.
Key Takeaways
- AI-led interviews combine the scale of surveys with the depth of qualitative interviews.
- Creating a new research category requires significant market education and behavior change.
- Enterprise buyers initially struggled to fit AI interviews into existing budget frameworks.
- Digital twins represent a novel way to simulate and understand customer segments.
- Outset’s success relies on leveraging AI advancements and strong early customer partnerships.
What the video covers
- Outset developed an AI-driven platform that conducts deep conversational interviews at scale, combining survey speed with interview depth.
- The product uses GPT models to lead interviews and synthesize insights, initially offering a conversational survey with CSV exports.
- Early adoption challenges included educating enterprise buyers who lacked budgets and frameworks for this novel research method.
- Outset had to create a new category in customer research, differentiating from traditional qualitative and quantitative methods.
- Market education was critical, with early customers like Weight Watchers helping demonstrate the product’s value.
- The platform evolved with advances in AI, enabling deeper probing and synthesis of customer insights.
- Outset introduced digital twins to simulate user segments, allowing faster and more cost-effective research.
- The company’s approach helps enterprises understand customers in ways not possible with traditional surveys or focus groups.
- Behavior change and adoption hurdles were overcome through persistent demos, education, and leveraging founder networks.
- The video highlights the transformation of customer research through AI and the creation of a new market category.
Chapters
- 00:00Introduction and Outset Overview
- 01:08Product Concept: AI-led Interviews
- 02:25Founder's Background and Inspiration
- 03:30First Product Version and Features
- 04:43Challenges with Enterprise Adoption
- 06:04Market Education and Buyer Resistance
- 07:14Early Customers and Network Leverage
- 08:25AI Advancements and Product Evolution
- 09:41Digital Twins and Future Research Methods
Full Transcript — Download SRT & Markdown
Speaker A
Thanks so much for being here, Aaron.
Speaker A
Thank you for having me.
Speaker A
Cool. Well, why don't we start a little bit by telling us exactly what Outset's product does?
Speaker A
Yeah. So, I find it useful to kind of look at the world of research and like to put into context what we do. So, when you're trying to listen to your users or talk to them, usually you've
Speaker A
had to use things like surveys, which are very, very kind of shallow capturing of data, or you actually do user interviews and you call them one by one. The problem is that's super expensive, slow, and time-consuming, and so
Speaker A
what we invented back when we started was this idea of AI leading interviews. That means AI is actually leading a conversation with a person, with a real person having a back and forth and digging really deep. So you get this
Speaker A
kind of the scale and speed of what we would, you know, do with a survey, but you're getting it at the depth of an actual interview. And so that's what we built.
Speaker A
Cool. And can you give a sense of kind of what scale you operate at today?
Speaker A
Yeah. So, as you mentioned, we work with a lot of the biggest brands in the world. And we've now done millions of customer interviews all led by AI. So it's become a really key way that people
Speaker A
are actually learning from their markets and their users.
Speaker A
Cool. So, as I remember it, when you went through YC in winter 2023, this whole idea of AI doing customer interviews seemed nonobvious. It wasn't like a market that clearly
Speaker A
existed. Tell us what that felt like back then and what was it like trying to get the first few users and people to believe in this.
Speaker A
Yeah. So, back in 2023, it was like ancient history at this point from the world of AI, the AI timeline. And so, we had this idea leaning very much on my own background. So, I started my career doing research myself, a lot
Speaker A
of ethnographic research in people's homes and doing very one-on-one interviews. So, I had a deep kind of care and passion for how we understand human beings. It's like what drives, you know, business forward everywhere. And so, I cared a lot about it. So, I
Speaker A
thought, you know, at the time, ChatGPT was really good at, you know, conversation. And so, we thought, well, what if we put that into the position of an interviewer, right? And ultimately, an interview is just using
Speaker A
conversation to elicit good insights. But the problem was the market was not there yet, right? Like in summer 23 it was like people were barely absorbing the idea of AI in the first place. And so, you know, what we put
Speaker A
forward was super novel. So, we launched the first week of the batch. We were the first one in our batch to launch, which we thought was like our special hack, and
Speaker A
a lot, we got a ton of demos booked during the batch, which is like an incredible position to be in, but honestly very few of them converted, and so it's like that's what it feels to be too early to a new category, right, where
Speaker A
it's like everybody wanted to come and learn about what was new and interesting and possible, but no one had internalized where exactly how exactly to deploy this kind of thing. Do you remember exactly what that was with the first version you
Speaker A
launched with? What did it do?
Speaker A
So the first version, it was basically taking the GPT models and it just wrapped around a basically chat interface where it would ask questions and a person
Speaker A
would answer. The synthesis of that data, which is the other half of our product,
Speaker A
is like AI leads interviews and then it synthesizes the results. The other half was literally a CSV export. So, it was like super, super bare bones, and so that's all it did, right? Is like you sent it to almost like a survey. You
Speaker A
sent it to your users and they would answer questions by AI. So, it's like an interactive survey.
Speaker A
It was basically a conversational survey was the way to think about it. Why did these demos not convert? What was the resistance you met from buyers?
Speaker A
So, I think we found pretty early on that there's a lot of interest from enterprises, which we loved. It's like should you be so lucky from the beginning to actually work with enterprises. The problem was that these teams and buyers were not—they did not have budgets for this. They did not have any framework for fitting this into their budgets. So it's like when an enterprise buyer, like you're talking to them,
Speaker A
they already have line items. They're already there. So either something is swapping something else out of a line item, or they need to make a case for an entirely new line item. And to make that case, they need enough
Speaker A
external validation that their boss or their boss's boss believes it. So the problem is when we're coming in with something novel, and it does not fit into an existing line item, then it's incumbent on us to kind of
Speaker A
will this new category into existence so that line item appears. What was sort of the alternative? Like what were they doing as the alternative and maybe where was the budget for that coming from? So there's a bunch of
Speaker A
incumbents in the research world, right? So we're talking about like Qualtrics and Medallia and these user testing, these kind of older school companies, and those had line items, right? So but those were like just static research, you know, the idea of
Speaker A
hearing from lots of people at once. You just send a survey and it asks very static questions. And this idea of a conversational one, it didn't fit into that bucket because it's not quantitative research in the way that
Speaker A
people think about surveys, and it wasn't qualitative research in the way you think about a focus group or a one-on-one interview. And so there just wasn't a nice home for it when we started. And so the kind of
Speaker A
challenge was how do you convince them that this is the best of both, that we've changed the economics of this thing and put that new line item there. And so it was a challenge getting the teams to understand how
Speaker A
to use a tool and adopt the tool because it wasn't clearly qualitative or quantitative, or was it? Did you have a different pricing model and people didn't know how to think about that and compare the price to like Qualtrics?
Speaker A
I think it was more of the former. So it was more about how people, the hard part was to get them to understand where this fits into what they do, right? Do they replace a project where they
Speaker A
otherwise would have done interviews? Do they replace a survey? And the hard part is if it is something net new, which it definitely was, then sometimes it's not an exact rip and replace. It's actually behavior change.
Speaker A
And so behavior change, especially when they're a prospect, is really, really hard. And so we had to ultimately do a lot of market education. So our first kind of couple years of existence, it was all about kind of
Speaker A
market education, not straight sales, not classic sales, right? It was like we have to help bring you on the journey of this thing.
Speaker A
So you launch sort of as quickly as you can during YC, lots of demos, demos don't convert. You realize you have to do more sort of educating the buyers because the category, the future isn't quite
Speaker A
sort of where you want it to be. What does that look like again, like tactically? Like how do you educate buyers? I mean, so hard to get their attention just for a sale. So how do you kind of
Speaker A
get them around with it? Our first customer ever was Weight Watchers, which was an amazing logo to get as your first one. I happened to have at the time somebody I knew there, which is my advice always for YC
Speaker A
founders of like shamelessly use your own network.
Speaker A
Um, bu—
Speaker A
next customer and that was I think the best way to start educating the market was like through examples.
Speaker A
So getting Weight Watchers to understand the right way to think about the product and and adopt it. How did you overcome that? I mean I I just there's no replacement for just spending a lot of time with them. And so I think it was a
Speaker A
very uh felt pretty forward deployed right we were like spending like a kind of irrational amount of time with that early customer and uh we kind of convinced them to try it in a few instances and once they actually tried
Speaker A
it in kind of a particular instance where it was really successful that became the blueprint for them and it became much easier. So that was like boulder rolling downhill for them. Now we had to bring that knowledge to the
Speaker A
next one and the next one. You guys seem to me like the classic case of a startup where it's like as the models have got better, you've benefited from that massively. Um maybe talk us through that like how exactly have the
Speaker A
underlying models becoming more powerful and there being more intelligence to build on top of. How has that improved the product you're able to build and how has that maybe changed some of the attitudes to the enterprise buyers you're talking to? So I think it's
Speaker A
actually worth stepping back and like think about the world of customer insights or customer research. Like out in the market there's actually like a uh insatiable demand for it. It's like an uncapped thing. So we're not just automating an existing workflow but
Speaker A
rather we're saying hey you can actually get way more insight about your user about your customer which means you can build a much better product which means you can do much better marketing which means you can grow your business. And so
Speaker A
the way we sell is actually a growth story right? is like a how how like you can by accelerating your ability to do research you can accelerate your own business and so because of that there is not like a single kind of beginning and
Speaker A
end to a use case that needs to be automated and so hey uh you know second tier model will do it it's like the better the models get the better we can do interviews the better we can synthesize data and pull in other kinds
Speaker A
of data the better we can actually drive more insights and so our first product was on Da Vinci right like forever ago um and now we use all the frontier models in various ways that work together. So we positioned ourselves
Speaker A
basically as like the more the models get better, the more insight you are going to get.
Speaker A
Have there been particular types of insights, maybe even specific insights some of your customers have gotten where it's it's very clear they would never have been able to get that from a traditional survey or like a focus group. They could only have got it from
Speaker A
their customers talking to the outset AI. Yeah. Okay. So I'll give you one example uh which I'll actually stick with the Weight Watchers example from early days.
Speaker A
So I think what you know when we started this product we we we could tell hey there's an opportunity to do something in a more scaled way than we've done before. And so we tested that and Weight Watchers had the first pilot project.
Speaker A
They did a study about weight loss and what motivates people. And you could imagine for a company like Weight Watchers, that's actually a really uh kind of strategic question that matters a lot for how they speak to their market
Speaker A
and how they advertise, how they do everything, right? And so we did a study uh where a 100 people were interviewed um about what motivates them and about their journey. And what was so interesting is not only did they get a
Speaker A
hundred interviews done in a matter of a couple of hours, which the traditional methods would never have gotten anywhere close to that. A hundred interviews takes months. But also people shared more with AI than they would than they
Speaker A
would otherwise with humans. And that's the really like counterintuitive thing here. So like it's not just that this was a kind of better economic decision to run AI interviews, but it's actually that you are getting data that was
Speaker A
impossible to to unearth otherwise. And I think it's like a it's the result of like, you know, people like love those anonymous apps where they can say anything and they they feel like they're able to express themselves, but there's
Speaker A
no kind of personal judgment and you're not trying to come off a certain way.
Speaker A
It's like that that's just transferred to research now, right? Which is like the most kind of uh incredible thing you can get is like people's ground truth.
Speaker A
Like what actually matters to them, what do they really think of your idea? All of those things really matter. It kind of reminds me when um when chat GPT was first taking off. I remember the people who really were deep in like early power
Speaker A
users of it um could understand that this was clearly going to replace Google search. But I think the people sort of at Google or working with search engines felt like oh like people were still want like like the fastest way to find a
Speaker A
result and it takes too long to like chat to a thing and wait for it to tell you like which things you should buy or go for. But the like once open AI once the um once the models have all this
Speaker A
context on you like because you share so much with them, they can actually get you to the thing that you want really really quickly. Um and it just feels actually more efficient than trying to like browse through it or look through
Speaker A
it yourself. And so I imagine you have like the similar thing like early on I bet a lot of incumbents will know like if we're going to do surveys and we need to like go through all of like the um
Speaker A
the responses ourselves to understand what's going on or focus groups need to be really um hightouch and done in person but people will like bear their souls to these things. That's like really rich information for you. Well, I
Speaker A
I think about this like even on a more macro level where you just AI can make us more efficient at things we want to do anyways. But then there's like AI, you know, the the the the uh has changed
Speaker A
actual human behavior. And so I think what I didn't realize when we started, I thought about it being an efficient way to get a thing done. And the reality was is it was creating a new set of human
Speaker A
behaviors that we've really really benefited from. It's funny because like understanding humans is of course what our product does, but it's also like we had to do that to take advantage of what's now possible with AI.
Speaker A
And you guys have hit a real inflection point on growth over the last year. So tell us what's um what's been driving that.
Speaker A
Yeah. Well, I think it goes to the throughine of like we create a new category after YC. uh we continued to push on the market and I know my co-founder and I at the time we were like wanting things to move faster,
Speaker A
wanting to hit an inflection point but we kept hitting conversations with people where they were really intrigued but did they have budget yet and were they overcome like were they able to overcome their skepticisms um of the category and so that's where we pushed
Speaker A
on the education we pushed on the awareness we we brought case studies to bear and we kind of scraped and clawed our way through and then really in 2025 I think there's a a big inflection point and what happened is the market like
Speaker A
this became went from experimental thing that are you willing to take a bet to consensus and so this is like a you know if if you're you know if you think about like hitting tailwinds of AI right like
Speaker A
this is a you want to be there and already have mind share when the inflection point hits right and like we had already been there at beating the drum and so the moment you know the conversation shifted from
Speaker A
should we explore this to we should explore this we were already synonymous with the category which was like I think in category creation like we just wanted to be associated with that net net new category and that's the position we got
Speaker A
and so the last 18 months have you know been been completely uh have been massive I feel like Ian I feel like going back to the fact that this was such a non-conensus non-obvious idea in 2023 going back in time and explaining to
Speaker A
people then that you'd be able to like talk to your agents because voice would have really taken off would have just seemed like crazy talk um if you project out forward sort of what's coming next for you guys like what are some of like
Speaker A
the the crazy capab capabilities that you're excited for that will let AI go even deeper into sort of figuring out what people want and and going and probing much further.
Speaker A
Well, it's funny actually. Uh I remember in 2023 during the batch having calls with people where I literally had to educate them on what were LLMs.
Speaker A
Yeah. Like I actually had calls with people who was like what what what do you mean LLM? And I had to go. So it's like such a different world. In the last three years now we like AI can do so much.
Speaker A
It's talking to your user's voice. And we've actually included a ton of visual intelligence, too. So, now we're using computer vision models to say, let's just say, you know, AI is running an interview and it can watch your screen
Speaker A
as you're interacting. It can watch your face. It can watch your environment, probe on those, dig deeper. We even have a way where AI will get feedback from people on a design and then literally change the design in real time and get
Speaker A
the person's further feedback. We call that codeesign. So there's like so many ways that AI has now like kind of uh uh uh become his superpower for research.
Speaker A
But um I'm really excited to announce and to launch that we're uh we're shipping outset digital twins this month.
Speaker A
Oh, cool. And so what that is is we're we're able to start synthesizing actual customers, actual users into what we call digital twins. So the idea here is that when you're doing research, sometimes you're able to and have the time and money and
Speaker A
investment to go out and learn from people, but as you do, you're able to put together actual twins that you can interact with as well, which are representations of those people. And so, um, this is part of our new simulations
Speaker A
lab that we've stood up at outset and we've developed a methodology where you're actually able to very accurately, uh, simulate what a person would say. If you think forward, right, businesses um ultimately just want to predict human behavior, right? That is the thing that
Speaker A
matters. That's why research exists. Predict will somebody buy a thing? Will somebody use a thing? Right? You're all you're it's always, you know, you're always trying to figure out the future.
Speaker A
And so now we have the ability to both gather the data to inform the future and also actually simulate it with real uh digital twins.
Speaker A
So that sounds really cool. Could you tell us maybe a little bit about how exactly do you generate these digital twins? like what what inputs are going into creating them and then what's the value to your customers of using the
Speaker A
digital twin for feedback versus just going to like the actual cumid customers and asking them.
Speaker A
Yeah, good question. So when we work with a customer, we identify what are the audiences they care most about, right? Maybe it's their own customer base, maybe it's net new prospects they want to get, you know, bring into their
Speaker A
to their business. And so we generate these set of twins based on them. We reach out to real people and we actually do hourlong grounding interviews. These are extensive AI run interviews that are digging into all sorts of dimensions
Speaker A
about their personality, about what they value, about what they care about. And through that, we're able to develop this persona core that can best represent that person. So, we've now created this digital twin that represents them. And what's cool is now anybody in your
Speaker A
business can interact with that digital twin at any time. Traditional research is great and you you know you want to keep doing it in all sorts of uh uh use cases but it costs money, it takes time.
Speaker A
Not everybody can go do this research but um with a digital twin anybody across the business can be tapping into that knowledge can be asking for example maybe your marketer wants to test a new messaging with that kind of digital twin
Speaker A
or your finance person is working on new pricing models and wants to simulate what would happen if I change that price. This is this infrastructure that should exist for companies to simulate the future. Yeah, it seems like it
Speaker A
should unblock like a huge amount of creativity because there have to be lots of good ideas are generated by people in these companies that they talk themselves out of and it's too expensive to go out and get budget or whatever to
Speaker A
go and run them like do like a focus group. But now, yeah, if you just have a digital twin, you can like test it out and build the case for it um without talking to anyone and that that should
Speaker A
be pretty exciting. That's right. It's like amazing how many decisions are made at big companies based on someone's gut or instinct. And like yes, they pull on the research team for the big projects, but there's so many decisions day-to-day that don't
Speaker A
have that. You can't do that. You don't have the budget or time or or capital to do that. And so being able to just talk to a twin of an actual customer or talk to a hundred twins of an actual customer
Speaker A
like that is a really powerful mechanism. Yeah, that's definitely one of the things I feel founders talk about as the companies grow in scale as they sort of in the sort of YC stage. It's almost so it's so easy in certain ways cuz you
Speaker A
just you have an idea and you go out and talk to users and like you just do it yourself and then the company gets bigger and there's all this other like stuff going on. But maybe with outset you actually like help bigger companies
Speaker A
be able to come up with new innovative ideas. That's right. And I think it like comes down to two things. It's like one accelerating the ability to get this input, get the insight, right? Right?
Speaker A
It's like we're if we're shipping code at the ridiculous rate that we are now, we're actually blocked on understanding what people actually want, right? That becomes more of a blocker. So, we need to accelerate that too. But the other is
Speaker A
like going and understanding humans in a way that we haven't before, right? And whether that's a direct outside interview with a person or a digital twin that's representing the most, you know, impactful audience that you have, both are ways to just capture more
Speaker A
insight more quickly. Okay. Well, maybe take changing gears here a little bit. A lot of people watching this are hoping to start a company themselves one day. Uh maybe tell us a little bit about um your journey to becoming a founder. Um
Speaker A
because I think you worked at some places before. So maybe tell tell us about that story.
Speaker A
It all started I walked I walked into Harsh's office and he No, but but but uh but my actual my background was working uh doing product at a number of startups. So I so I worked uh Yeah, we worked together. You were our
Speaker A
first product hire. That's right. That's right. Har took a bet on me. Um, and I was the the head of product under under harsh and team at triple bite and and I think like for me working very closely with at existing
Speaker A
startups. So if I you know at the time I wasn't ready to go start my own thing and so like being able to work really closely with founders who are in the trenches and understand that experience and start you know pattern matching like
Speaker A
out of different phases was a really big way that helped me. the other the other the other advice I always give of like if if folks are you know thinking about it one day is basically uh life is short
Speaker A
and do it now and uh you know I uh decided to uh leave my job and start a company and I hadn't yet figured out exactly how that was going to go or what that was going to be and I was still
Speaker A
kind of exploring you know kind of uh I just decided decided to start working with Michael my co-founder just deciding to do it and not sitting around waiting for the right idea to strike me right like walking down the street was the
Speaker A
most powerful thing I think that that happened. Well, I think that's all we have time for today. Thanks so much for being here, Aaron, and can't wait to see the digital twins launch from Outset.
Speaker A
[music] Thank you for having me. [music]
Topics:AI interviewscustomer researchOutsetYC Root Accessenterprise adoptiondigital twinsGPT modelsmarket educationqualitative researchquantitative research
