Investor Perspectives on AI-Native Startups

Good to meet you all. My name is Pratna, and as Laisha said, I'm the former Managing Director of Maple Leaf Angels and currently a partner at FutureSight.

So my talk today is going to be a little bit different than David's. David had a very cool lens around building AI products and what technology to use and what not to use. I'll bring the lens off basically as a capital allocator.

so as Laisha said over the past 10 years for the first half of it I was sort of deploying capital at the pre -seed and seed stage directly into just pure software companies pretty agnostic in terms of sector and it would generally be yeah like I said pre -seed and seed so you know round sizes of under three million dollars now I'm involved with an organization where we are basically basically creating net new AI native companies.

And over the last couple of years, we've built 10 companies in the portfolio. And when I say AI native, I basically mean they are vertical, vertical, sold as software, they're B2B, they're generally targeting the mid -market. And in terms of sectors, it's roughly split between about 30 % digital health, 30 % legal, financial services, a little bit in education, a little bit in prop tech and construction tech. But generally, we're not deeply married to any sector, just won't go very deep into, I guess, crypto or payments. payments.

So there's two things I wanted to cover today and let's see what we can get through. I don't exactly know what will be most interesting for you guys, but let's give it a shot.

A Framework for Evaluating AI Companies

The first thing I want to talk about is a little bit of a short framework about how investors today are looking at AI and how we're evaluating companies.

And then the second is because we run basically a a venture builder. We're a startup factory. We are doing what many of you are doing in your own heads, but we're doing it as an institution.

So I would love to break down every little function that goes into it and how we're AI enabling it. We don't have it perfect, but we're experimenting and I think, yeah, that's all I have to say on that. And sorry, yeah, I think the important important thing is that each of those can be their own deep dives for sure.

Where AI Value and Defensibility Shift

So I think this one idea of when reasoning becomes abundant, where does the bottleneck shift is a question. And we've all been in this golden era over the last year, a couple of years, right? Suddenly feeling all the more intelligent. And businesses are feeling it too.

And so we started to think about where is the the ultimate prize when it comes to AI. And initially, we said it was the model, the brain. Then we said it was the harness, the brain needs hands and legs. Then we said it's the context, it's the know -how and the proprietary knowledge. How do you know what to do with your body once you have your body?

So more and more, we are shifting towards building that structural data advantage or proprietary context as the critical prize when it comes to investing in AI software. And so that's what I would say is the first thing.

Four Factors Investors Look For

Four things investors care about is what I'll talk about today. One is the access mode. How are you entering? How are you earning the trust? Is there some regulatory complexity or nuance where state by state there's a different mechanism that somebody cannot create something in a week and go out and run the same motion?

So I can give you an example. We have a company called Crewscope in the construction tech space. It's actually one of our only Canadian, one of two Canadian companies. They have been in market for a couple of years on job sites, on construction job sites, the CEO and CTO there and the field workers recognize them. they trust them and they've now earned trust so when they like everybody else go to market with an ai product a construction worker is likely going to be inclined to trying a versus b so earned trust is relational is structural that's number one the second one is something i alluded

to at the start which is the data asset today it's almost table stakes that people are saying thing, the only thing I will invest in is something that has a structural data advantage. The obvious thing that that is, is like a point solution or a tool does one thing, something with a data asset has built data either by capturing existing data or creating, generating net new data and building an asset.

Couple of examples for that, I think David talked about voice. Voice is huge now. But with voice, you're able to create net new data, which is amazing. And this data, this is new context that never existed before. We have a company called Addy in the student counseling space, which is having conversations with students. It's new data that's generated for a school counselor where they just never had that access before. for.

Or Mercada, another company in the hedge fund management space, which is creating the company brain or the institutional memory across all the data points, all the decisions, all the data feeds that are coming in on the way a business has operated. So that's number two. First was the access mode, then is the data advantage.

Number three is workforce leverage. So we have this question thrown around pretty often like is it creating massive AI leverage workforce leverage this whole trend around AI will take over our jobs no well there's a step before that it's leverage and so can you put putting more tokens at it does that generate much more leverage and so again in that counselor example those conversations could never have been had before and And when you compute the financial cost of using something like this AI -based counselor versus a human counselor, you will see at least a 4x cost difference on an hourly wage level.

And then comes the next part, which is most interesting because it'll raise some flags should have gone off for you in these, which is human observability, believability, and auditability. Can they see where the result is coming from? can they believe it, and can they go back and audit it? Those are basically table stakes, and I would just call that a gate that exists above the others.

So I'll just recap those four. Access mode, so what do investors care about today? Access mode, data asset, workforce leverage, observability, believability, and auditability.

How FutureSight Uses AI as a Venture Builder

Then I want to move on to FutureSight, our organization, and how we are AI powering our setup at the moment.

So as a venture builder, we basically are thinking about the different steps that you'd be thinking about. One is what idea, what market am I going to go after? Then who's the team I'm going to build it with?

with, then how am I going to get customer validation and development, customer validation and develop my market? How am I going to do product development? How am I going to finance my company?

So, you know, we've spent five years basically building each one of these teams up and processes and functions. But at the outset, I'll just give you a high level about what each one is doing.

Generating and Validating Startup Ideas

So ideation is probably the most interesting thing is like where do good ideas has come from. And you could spend the same amount of time on a good idea or a bad idea.

So essentially, we sat back and thought about our thesis around the labor market, right? With AI coming in and giving a whole lot of leverage to businesses, what is going to happen to certain jobs?

And so we built a tool that basically scans all the white -collar jobs in North America basically ranks them by salary be built an agent that will basically be able to assess the level of AI leverage that you can get and we generate ideas that way

we've inserted data feeds of different companies around the world startups and high growth and then we're scanning for what markets are experiencing growth without raising venture capital dollars just to kind of assess where is the heat map and so we have a couple of mini agents they're working that short lists for us then a few different ideas that we look at every week we run that process every week every month and that's how we generate two new ideas per week from there we go out and find founders how do we

Finding Founders and Developing Markets

find founders combination just like you guys would find founders our own networks then we We put a call out for founders, we go to events, we talk a lot about our thesis in a market and so on, and then we built a screener, an AI screener internally to just look at all the profiles, and we gave it scoring around specifically what we look for, and then we have humans involved throughout the get -to -know -you process.

Then there's the market development, customer validation. foundation we are experimenting with and have experimented with AI SDRs and synthetic interviews and of course nothing really trumps a founder coming in with their own network or a partner having an existing network in a vertical but there's a reasonable amount of experimentation there.

Prototyping simple and obvious it's too easy to do that today using AI so we're able to to do that.

Capital Raising in an AI-Driven Market

And lastly, on capital. Capital, I would say, and it'd be cool to get into this at the panel, is a good question.

I don't have the verdict in terms of can you raise capital easily with AI and replicate the human aspect in it. I do believe there is a, I don't know what the term is, cottage country something feel to it, and largely human run.

But we are experimenting with with AI outreach, double -opted AI outreach for venture investors. There's a whole universe of investors out there online that can be found that we would not even have in our networks. And transparently, we are experimenting with that as well to see if we can get hold of them.

So that's the factory. We invest about a million dollars into every build that we build out. and yeah our job is to make money with all of this thank you thank you so much

Audience Questions

wait we still have questions for you yes anyone who has a question oh wow okay we're gonna take three questions maximum before going to panel discussions all right just go ahead thank you so for

Building Defensibility Without Industry Access

For someone who's starting out new, you mentioned about access mode. Somebody who does not have access mode, no trust because they might be new in the industry, how would you evaluate their defensibility on an idea that cannot be replicated because ideas now can be replicated very easily using AI. So what would be your parameters for someone who's starting out completely new without a lot of access mode?

so suppose your certain industries you know the person you just don't have a right to play i don't think for instance um you know things obvious ones where there is an extremely high education bar and so on that you need to have spent 10 years working in it um with that said

i have seen many successful companies where the ceo has not been from a vertical and um they've They've had very strong co -founders who have been from the vertical. So I think that can work as long as we solve the domain access gap somehow. And I think that's not replicable.

So in another framework that I put together, market knowledge is a parameter, which a CEO without the sector can get, and market access is another one. is that can you make it rain with the first 10 customers and if i've actually worked worked in

a scenario where the ceo did not have domain expertise and their right hand person or ceo did and they did the job there of identifying the talent and and driving it for them to make it work but as a founder i think you've got to solve it quickly because that's number one great thank

Data Strategy: Renting Versus Owning Intelligence

you and i think you have another question yeah um thanks for speaking um you mentioned uh data modes being uh important um i'd like to understand how you think about how your portfolio companies think about renting versus purchasing intelligence and so i know that generally like i follow the story between like model harness and context but i think there's still your your data is not particularly clearly valuable until you either sell it to somebody else or you can use it to actually power one of your products so just maybe walk through with a couple examples how your portfolio companies think about renting or buying intelligence yes it's a good question i think it's actually a

mix um if i think about it um so in the areas where we're creating new data that there's there's no question of renting or buying but you're talking about the companies themselves renting or buying data, right, in order, yeah, yeah, I think renting, renting is generally the approach,

I think the, let me think here, so I have an example with a company, Mercada, and they're working in deep institutional fundamental investor land, and that company is an asset management firm and has data feeds coming in from different places that has to be in play for Mercado to work and how they and and so they kind of have the the position is that the customer is responsible for securing the necessary data and Mercado can then sit on top of it am i answering your question with a

a different story yeah I I actually I I'm not totally sure yeah I'm not totally sure yeah

Investing in Profitable AI Startups

all right we have one more yeah please go ahead hi thanks for speaking my name is Nick I have a question I think there's a general conception or perception that we're not truly paying the cost of AI so like consumers of you know AI agents or AI models so I'm curious how VCs find potentially profitable startups to invest in because think of the the amount of money on Tropic for example has burned down the AI journey and they're not truly profitable as a company right so how do VCs go finding in those rare AI startups, not just any software startup, but like an AI startup that can truly yield return on investment?

Yeah, it's a super good question. I think the job of the investor during this last six or eight months has gotten really confusing. And during periods like this, when people don't have clear conviction, they end up following and chasing the themes, right?

And one of the major, this phenomenon that's happening with SpaceX and Anthropic and OpenAI, this idea that you don't, the shift is actually happening where people are saying, I don't need to invest in AI startups when I can get so much of a healthier return this way. And so if you have the ability to get access there through the secondary market, market you're likely going to sit back a little bit and I personally am seeing that especially so so that's the first thing I'm seeing.

The second thing I'm seeing is because of that there is some level of belief that and this happened a few years ago as well some level of belief that we have to invest in foundational model technology or the infrastructure or robotics. We have to invest in really, really deep tech because we don't know where the moats will form in vertical software. And that's relatively understandable, but there are

certain class of investors that have doubled down almost with greater confidence on vertical AI with the conviction that AI is just going to change the way software was being delivered and work was being done. So it is a big bet. And as I spend more time here, I understand that ultimately the bet has not changed from 10 years ago. The bet remains the executionability of the founder and some level of ambition and vision to see into the future that the technology is going to change buyer behavior in certain way so there's I almost maybe one to two

times a week I'll hear from an investor or an LP who will say that you know I'm just gonna sit out of startup land and profitability that's what people might say I wouldn't say that that's happening at early stage for sure and so it remains a long game