Subscribe
Share
Share
Embed
Making capital allocation decisions in low information environments and with alot of uncertainty is hard. This show talks to people who do this every day and teases out how to be good at it.
Martin Tobias (00:01.603)
Hello everybody. This is the first bet. And every successful person gets interviewed about how they won. Very few people drag them back to the moment before they knew it would work. When the information was thin, when the money was real, and they pushed their chips in anyway. That's what we talk about here on the first bet. I'm Martin Tobias and I've bet three ways as a CEO, as a pre-seed investor, and across the table at playing a lot of poker.
My guest today spent years inside the rooms where the AI products are being built, AWS and Microsoft, the actual machine watching real product seeing what real product market fit looks like before it became a press release. And about five or six years ago, he walked away from that and started betting his own money in YC companies you know, before consensus arrived at volume.
We actually met each other running syndicates on Angelist and now Brian runs a couple of funds. Brian Bell, managing partner of Ignite team Ignite Ventures, welcome to the first bet.
Brian Bell (01:09.986)
Yeah, Martin, I'm honored that you would invite me to to chat with you. And for folks that don't know, Martin was a mentor of mine when I when I got started, you know, about six years ago. I reached out to him, just cold, cold emailed him on Angelist. And I was like, Hey, I'm new. Let's let's chat. And he he was nice enough to take the call. I I I distinctly recall I was at a like a soccer practice and you're like, Hey, can you chat chat now? I'm like, sure. I'm just walking around the field chatting with Martin and he's just unloading his knowledge about investing. This is back in like probably twenty twenty one.
Martin Tobias (01:37.055)
Okay.
Brian Bell (01:39.952)
yeah, so thanks for doing that. Yeah.
Martin Tobias (01:41.919)
It was. Well well no problem and and I I think you you know that I've been obsessed about how do you, you know, make capital allocation decisions in low information environments for a long time and I'm glad that more people kinda join that thing. So w what I want to talk about, I mean you you started doing this, but you've you know, since probably evolved and raised some funds and done done different funds and you said
to me or in the in the opening you've done over four hundred investments. And what we're what I'm trying to do here is extract, you know, from how from your process how you make decisions in low information environments and the frameworks so that other people can understand, you know, the the the frameworks that you use. So maybe you could start with that, not a particular investment, but how do you think about and and in could you mention you know some ways that you think about
how to deploy checks in low information environments, what criteria or decision frameworks that you use to decide to write checks in two guys and a dog and you've written four hundred of them.
Brian Bell (02:48.91)
Yeah, nearly nearly 400 now. And, you know, I got started like a lot of people do, you know, writing angel checks directly into some companies, you know, some some checks into funds and sitting on their investment committees. You know, some of those funds kind of invite some of their LPs to get involved. and then, you know, after a couple of years of doing that, I realized I was having a lot more fun talking to founders and playing Game of Thrones at Microsoft is what I like to say. And and you know, so I I I wanted to do more. So I started a syndicate in Angelist, and that's how we got
Connected. It's a great way to get started. you know, but you you don't have deal flow, right? And or it's very minimal. so one of the ways I bootstrapped deal flow at the beginning was to to fish from a pool of really good good deal flow, which is YC, right? Kind of the preeminent accelerator. they get something like twenty thousand applications per batch and they only let in, you 150 to 200 of those. and so you've already have this huge filter.
Martin Tobias (03:30.485)
It's cracking.
Brian Bell (03:47.211)
of of very high quality startups to to invest in. And now you're trying to make a decision very, very quickly. Talk about low information and high pressure. You know, a lot a lot of these rounds close weeks b ahead of demo day. And it's it's just like a one and done. There's no due diligence you can do really. I mean you get a little demo, you meet the founders, you know, you kind of look through stuff, but you basically need to make a decision that week, almost that day, in order to get into the hot rounds and
Martin Tobias (04:00.04)
it's just pocket.
Martin Tobias (04:13.459)
That's the that's what giving
Brian Bell (04:15.352)
That was a that was a big a big reason we even raised a fund in the first place. You know, we were happy to run syndicates. I never thought I'd be a fund manager. I just wanted to keep investing in startups. And, you know, I would three things would happen. You know, one, the rounds would close before demo day, which was painful. two, the valuations would change. So I'd have a syndicate open. you know, I'd raise a hundred K, 200K, go to wire the money, and the founder's like, sorry, the the round, you know, the rounds closed, or hey, I I told you told you a twenty million dollar val
Martin Tobias (04:43.132)
It's now a higher cap.
Brian Bell (04:45.058)
I told you 20 cap, but it's a 30 cap now. Why is it a 30 cap? Well, we doubled revenue and the rounds oversubscribed. or the worst thing the of the three would happen, which is they'd say, I don't want to syndicate how how many people in your syndicate? I don't I don't want to send it out to like 600 people, let alone, you know, Team McKnight now is 16,000 people, right? I don't want to send it out to that many people. And so we raised a fund on Angelist, a little rolling fund. That was like kind of our fund zero. We call it fund one, a little million dollar fund. So we could move quickly with small checks and get into the hot rounds.
And then you talked about kind of how do we underwrite that. You know, when we, you know, when we got started, we kind of looked it around and and looked at what other smart people thought. You know, we we chatted a lot about how to make decisions back then. I mean, obviously you're looking for an amazing team, you know, good founder market fit, a good a good why. I used to really care about their why. Why are they doing this? I don't I don't know if I care about that as much anymore. but you want like I I kind of envision this like, do you envision this person
Martin Tobias (05:16.774)
Okay.
Martin Tobias (05:41.414)
Like I said to the stride.
Brian Bell (05:45.977)
Being this Nasdaq bell ringer in 10 years. Do you see them being interviewed on CNBC? and if they're not that kind of quality, that star quality kind of person, they're probably not gonna be successful because they have to like go sell people on leaving the comfort of their big tech jobs in most cases, because we invest mostly in, you know, technology startups. And so you got to be able to attract the right talent. And so you got to have that kind of star quality, that recruiting ability.
Martin Tobias (05:54.822)
Yeah, or public that's still often not that like twelve thousand people on TV, but they're taking
Yeah. can you post the Wilmot? Actually, added that from thank you.
Brian Bell (06:12.942)
You know, coachability, velocity of learning, velocity of understanding the market. Obviously you gotta gotta look for a big, big problem with l you know, I learned I learned this from you, lots of friction. I actually added that from you into my framework. And I think about that. you know, how like what friction is removed. you know, timing is a thing. I I wanna write a whole book on this actually. You know, timing is is really tough to to get right. You can be too early, you can be on time, you can be too late.
Martin Tobias (06:33.329)
I'll provide all four kinds of that's going to easy work if you don't have to laugh, like Google the one.
Brian Bell (06:42.702)
You know, I think about this a lot. Like Google was the seventeenth search engine to come along. And I c I think about this a lot. Like if I was a angel investor in like nineteen ninety seven and L Larry and Sergey walked through the door, would I invest? I think about that a lot. because I I c yeah.
Martin Tobias (06:55.248)
Well, I was an LP in Ron Conway's fund, and he did give Larry and Sergey their first $250,000. And thank fucking God, because as an angel investor, I would have never seen that deal. and I still own some of those Google shares. My cost basis is 50 cents. and I haven't actually talked to Ron yet, but I would want to go back to that point where he met them and say, Okay, how did you decide? Because there were lots of other competitors.
Brian Bell (07:05.954)
Right.
Brian Bell (07:09.379)
Right.
Brian Bell (07:14.966)
Incredible.
Martin Tobias (07:24.555)
you know, looking back a little bit later, my own personal d decision on on Google why I was glad he invested is that the product actually was an order of magnitude better than the existing things. If you just did one search on Google, one search on Excite, one search on Alta Vista, you got better results immediately from Google. I switched from Alta Vista and Excite in like one search.
Because the product was just that much better. Now, how did they do it and all the technology? I'm sure he did some diligence around that. But in that case, I think the product was demonstrably better than the competitors, despite there being lots of competitors at the time. So let me just summarize what you said. The the the the first way to reduce risk on low information is to try to pick from a a better curated pool.
And that's why you do a lot of YC things, then would you get into that hopefully better curated pool and and I and I would agree with you on that. Like it if if there's 150 companies in YC, even if you're throwing darts at 150 companies, if you are you're gonna get a better result from those 150 companies in YC than 150 random companies on Tech Stars or you know, on Pitchbook or something, right? So you've already sort of constrained your thing a little bit.
The second thing that you said is that you raised a fund so that you had dedicated capital and the ability to execute quickly in some of these times where you may have to make a quick decision. And and and that was a response to a prior investment strategy where your capital was sometimes a little bit slow to be able to commit, it which caused you to miss opportunities. So that was a compensation.
Brian Bell (09:17.453)
All the time.
Martin Tobias (09:19.884)
of deal structure which allowed you to to to move quicker and then you got into of those 150 how how do you pick them what would you do you have a a a rubric you know like percent of the founder versus the market versus the product versus traction and could you you know share that when you're looking at an individual company inside of yc
Brian Bell (09:46.062)
You're asking for my proprietary algorithm? Martin, come on.
Martin Tobias (09:48.523)
Well we've had other people share them. If you don't want to share it you you could say
Brian Bell (09:51.815)
No, I'm just messing with you. No, I think, yeah. And, you know, I led AI at Amazon, launched Sage SageMaker Marketplace. I built a lot of AI and in in my time, know my way around a Jupyter notebook and how to train models and stuff. And, you know, so when I when I started investing, I just started thinking about first principles, right? how would I set up feature extraction? What are the features that matter? And what are their weights and biases in the model that would output a score that would predict
Martin Tobias (10:17.215)
Okay.
Brian Bell (10:21.452)
Success, right? And so what you start doing is you start collecting data. luckily with YC, you have a lot of past data. They've done 5,000 startups. There's lots of really good data sets out there on which ones have been successful. But it's it's there's an art and science to it, right? So we've built up this AI system over the years where we can ingest lots of data about the founders, their resumes, their some, you know, the transcripts I've I've all all the transcripts of the thousands and thousands of calls I've done, all the pitch decks.
Martin Tobias (10:40.36)
Lost the yeah about the
Brian Bell (10:51.182)
and we can basically do some feature extraction there with a weighted scorecard. And the AI has actually figured out the weights in that in that scorecard. And you can do the same thing, right? Like a lot of a lot of machine learning is just that. It's feature extraction, weights and biases and and and outputting a number between zero and one. And in our in our case, it's one to five, because that's better like
Martin Tobias (10:54.57)
And the guy has actually figured out a way to take that program.
Martin Tobias (11:08.042)
Take the lines and
I'll get that doesn't.
Brian Bell (11:15.394)
That's how we think. I don't think of like a point nine startup versus a point six startup. It's better to think of like, the startups like a four and a half versus the startups like a two and a half. And so actually everything that we do when we underwrite is we've done all that feature extraction. I think there's about 20 features in the model. and these features have friction. they have founders, they have moat traction. you know, obviously,
Martin Tobias (11:19.294)
Yeah.
Actually right is if you stop on if you just track and I think that's
Martin Tobias (11:33.247)
Okay. and also the first find those things care about is like hey I kind of pretty rendered application for me. Right. Right. basically I kind of go through the features. I've got over by it. Okay.
Brian Bell (11:40.655)
Problem category defining. there's about 20 of those things that we care about. And so the AI, by the time I even look at a startup, it's already rendered an opinion for me about like how good the startup is. And then basically I kind of go through the features and I sometimes I override it. I'm like, actually, no, I don't think it's a four founder. I think it's a four and a half because of this. You know, I you rated the traction of four. I think it's a five because they're doubling, you know, they're doubling every month. and I think that's a five out of five. And so I'll go back and kind of play with the features.
Martin Tobias (12:01.492)
you read this trash before, I think it's a five different double.
I think that's fine. But I kind of like the feature of it. And then what I could do is I just need to have it actually five times there. And I think this kind of investment possible.
Brian Bell (12:10.466)
The weights are decided by the AI, but I kind of play with the feature ratings a little bit. And then what I do is I ingest the entire 200 company batch for YC four times a year. And I basically stack rank it from hey, this is the best possible startup and this is the worst one in the batch. And then that way I prioritize my time. I do end up manually reviewing everything. and I do this outside of YC too. So half of what we do is YC, half is non YC, I would say, about 50 50.
Martin Tobias (12:27.75)
Okay. I I could side of IT.
Brian Bell (12:38.262)
And so by time the batch comes along, and I'm just about to head head as we record into another batch because you know, demo day is coming up in about a month, I'm starting to look through that stack rank now.
Martin Tobias (12:50.394)
Okay. So one way to which is new maybe in the last year or two to make a low information decision is to just get more information. And AI is is is an amazing thing to close that information gap in ways that you couldn't even five years ago. You know, five years ago you're doing your own Google searches for competitors and trying to troll their LinkedIn and now
AI can do a lot of information gathering and and you have over year the years def figured out which of the things you know about a company or a founder is correlated to returns by backtesting it against some other data. So even though people think there's not a lot of information, there's actually more information than you would think in in some of these things.
by looking at past performance of of Y C companies and being able to extract the the data. I I mean I remember
Brian Bell (13:55.823)
And all your past performance too. Like the the benefit that we have being in the industry, you've been in it a lot longer than I have, is I've probably looked at, you know, 10,000 companies a year now for six years. So I have a data set of, you know, probably 40 to 60,000 companies. A lot, a lot of things that I invested, I shouldn't have. A lot of things I I did I passed on, I should have invested. And so I have a lot of these, they call that F1 score in machine learning. You know, true positives, true negatives, false positives, false negatives.
And so can actually calculate your, you know, investor F One score over time and see how good you personally are and and your AI is. And together, I think the future of investing is probably some sort of a hybrid. and there's a lot of research on this that hybrids, you know, human AI hybrids perform better than either alone.
Martin Tobias (14:43.247)
Yeah. I agree. So so you collect some data and then you you stack rank them and then you decide out of the hundred and fifty companies I'm gonna meet these twenty. And then what are the human elements that that you apply? Is it a feeling? Is it a trust of the CEO? I mean, what are green flags and what are red flags that you see when you actually meet the CEO?
Brian Bell (15:11.064)
Yeah, that's a good question. and a lot of it comes down to hunch, right? I think and pattern recognition. You know, when you've been in the room with Andy Jassy at AWS and he's grilling you on a project, you kind of know what excellence looks like. when you're surrounded by that level of competence, you're just you you I've been around the A players. I don't I don't even think I'm an A player. I'm probably like a B player. but I'm a B player who can spot A players. And and so you're kind of just trying to spot those A players, right?
A players attract B players and so on, is as how the saying goes. the second question was around red and yellow flags, which our our model does pick up on, and we actually have an output for that as well. I'm looking it up right now. I have this long framework in front of me. there's all kinds of rules that we've developed over the years, around there yeah, there's caps and all kinds of different gates.
Martin Tobias (16:06.906)
and all kinds of different cases, you know, really triggering that we can type the first
Brian Bell (16:09.09)
You know, the where the AI will trigger and say, Hey, this is this is a pattern that we've detected that doesn't work in the past. And this is a pattern that we've detected
Martin Tobias (16:19.093)
And what would be some of those that don't work the cases against or the the the yellow or red flags?
Brian Bell (16:24.854)
Yeah, it could be, it could be as simple as like a capital inefficient growth. it could be things like retention. you know, like, hey, yeah, they're growing 100% a month, but did you notice that their logo retention is, you know, 67%? and their net revenue retention is, you know, 80%. And how does that compare to all the other startups in the data set? Right. What does excellence look like? So there's a little bit of like,
Martin Tobias (16:29.823)
Okay.
Brian Bell (16:52.162)
Here's what the positive signals look like, but there's also all these like negative signals as well that can happen. and we could sit here and I could read this model out to you. If I read I read this whole model to you, it would probably take at least an hour to read verbatim every word in this model, which I won't do. Obviously, it's proprietary, but
Martin Tobias (17:08.576)
No, we we we we don't want to do that. But let me ask you another question about YC and one of the things that they're very good at. So you you you know, they go in with a particular plan and this and that. And I I invested in in a company that got into YC and what they're doing today is completely not what they were funded coming out of YC. The amount of pivots that we see coming out of YC, I don't know what the numbers are. It's probably fifty percent.
Of the companies that come out of YC
Brian Bell (17:37.4)
Yeah. It's it's it's it's probably not higher than fifty, but it's probably not lower than like fifteen or twenty. Yeah. I don't know exactly the pivot. Yeah. Yeah.
Martin Tobias (17:43.699)
It it it's it's a high per so how do you think about pivots in in in your underwriting framework? Are you okay with them? Is that what you expect? Like the company that I invested in, they were they went they had a generative AI search product to help brands show up better. They got into YC on the basis of that, they got funded in YC on the basis of that.
And then basically about three months later, the CEO said, There's too much fucking competition. This product sucks. My churn's too high. There's I I just don't think I can create a defensible business here. So he took his four million dollars of YC, pivoted into a completely different business. That business is actually doing better. But the but but the the fact is that the CEO was smart enough, and if you you know go back to the kind of founders it's v YC wants to back, they should be smart enough to figure out
The the the the pivot, right?
Brian Bell (18:38.498)
And that's why founders is usually the the first and last thing. You know, if you ask a lot of VCs out there, they'll say it's it's it's founders w first, founders second, and founders third, because especially before you're like one or two million of ARR, you you don't know if you have something. Right. And even sometimes when you reach one or two million of ARR, you still have to pivot, right? it's why I actually developed this rule. Maybe you'll if you end up
Martin Tobias (18:42.643)
Yes.
Brian Bell (19:02.998)
Interviewing your fa your founders is I don't invite founders to the podcast anymore until they're at about one or two million of ARR to the Ignite podcast. Because I've had founders come back to me and say, Hey, can you remove that podcast? Because it's talking about something that we no longer do. Yeah, we and I've had to do that a few times now in three hundred almost three hundred episodes. They're like, you know what, that talks about something we don't don't do and we've pivoted away from. and yeah, so that that's challenging because these YC companies.
Martin Tobias (19:16.466)
That we don't do anymore.
Brian Bell (19:31.961)
They're raising at 30, 40 caps, right? With, I don't know, half a million of ARR in most most cases at that cap. and so they feel like they have something, but you know, that they may not, right? They may decide that I can't capital a capital efficiently grow in that space as as good as I can. And we actually have this 11 point fragility score in our framework. and these are picked up from the data set over time. Founder fragility is one.
you can have like a single point of failure. So solo founders kind of are weighted down a little bit in our model. or or if you don't have like if you don't have strong complementary skill sets, right? Like if you don't have like a strong business person, strong technical person, market fragility. So you're expecting the market to shift. And the thesis is if the market shifts this way, then everything will work out. But if it if it turns out to not shift like turn in the way that you thought it was gonna turn.
Martin Tobias (20:10.427)
has to become complicated.
Martin Tobias (20:17.848)
It's the whole year as action.
Brian Bell (20:31.054)
if you're betting that this thing becomes a unicorn based on how the market's moving and it doesn't move that way. So we have a market fragility score as well. There's also product and capability fragility. so what what does the does the success depend on unproven research or hard physics, deep tech that you haven't proven out, that you have no IP for? there's GTM fragility, right? So you can have like a good GTM score, which is in our framework.
Martin Tobias (20:57.212)
That's the
Brian Bell (21:00.206)
But you can also have the inverse of that, which is GTM fragility, which means you depend on one particular kind of go-to-market motion or partnership or platform, which creates a lot of risk. And I could go on, there's eleven of of fru like of fragile vectors that you have to analyze too with all these startups.
Martin Tobias (21:04.86)
kind of market right the purchase platform that creates a lot of risk I go on as a bit of
Martin Tobias (21:21.144)
And and what percentage is the fragility score in your overall thing if the the founder is fifty percent?
Brian Bell (21:26.574)
Well, our yeah, it's interesting. Our model is actually it has multiple scores. It has a power loss score, which is kind of like, hey, if everything works out, is this like a $10 billion company score? Then it has like kind of the weighted scorecard, which are these 20 factors that we care about. Then it has this fragility score, which is like, hey, how fragile is the thesis here for this startup? Yeah. And then it has the the kind of the what you were kind of getting at the red and yellow flag score.
Like are there anything is there anything here that doesn't seem right? Right. That you that comes up in due diligence, right? If you dump the data room into our AI, it can come up and and and basically say, Hey, you know, they said this in the call, but I checked that and that doesn't match. And you know, like, not that founders are being they're not lying to us, but sometimes they're kind of stretching the truth a little bit. And sometimes they're out like I've, you know, every once in a while, like one percent of the time, they're out outright lying.
Martin Tobias (22:12.43)
Yeah. That's a capital T.
Brian Bell (22:24.482)
Like they said they had this this credential and they do not. You know? And that's the power of AI now is like we can dump data rooms in the cloud and chat GPT and build up a model and and see all this stuff at a glance. And man, I can't imagine what it was like for us, you know, five years ago having to do all this stuff manually. just crazy.
Martin Tobias (22:27.132)
Sure. Yeah, yeah. I mean I I
Martin Tobias (22:45.54)
Well, so what I think is what what I'm hearing you say that's that's new in the last couple of years that you can improve your information state even in an early thing like this, in a way that you couldn't even just three to five years ago. Because I I would think venture decisions, you know, when you started, when I started doing syndicates were much more related to the gut or a very limited amount of reference checks that you might have done or feelings about the market.
But now you can increase your information state quite a bit with AI, and hopefully that's gonna improve returns. I certainly think it will, but the funny thing about venture is that your initial decisions, you probably have an idea of your if your decisions six years ago were any good, but your decisions in the last 12 months, you have no idea yet whether they're any good. There's there's a time d delay.
Brian Bell (23:38.85)
Right. Yeah, it's like you're playing poker but you don't find out if you win the hand for five years, you know?
Martin Tobias (23:44.781)
Yeah. I w I was just talking to a another poker player. The you know, poker players have low information too. They have two cards, you have to decide if you're putting money in the pot, but you get an immediate feedback on your decision. In venture, you you might not have a feedback for ten years. You might have within eighteen months a feedback of would the next investor mark this up? Are they have they gotten enough traction? but that's not
sort of the final d this decision at all. so having done this strategy of writing, you know, smaller checks into a bigger pool of of Y C things, do you you you have written non Y C checks too. Do you think differently about non Y C or or
Brian Bell (24:32.183)
Yeah.
Yeah, and that and that's a that's a that's a benefit of starting from a pool like YC because you it's such high quality that when you see non-YC companies, you're kind of like your your brain is primed for YC level quality. And so now you're kind of you meet companies outside and you're like, gosh, you don't even know how bad you are. Like like some decks I'll see, my my AI will score at like a two point two, which is like an automatic pass, right?
Martin Tobias (24:52.842)
Yeah.
Brian Bell (25:00.046)
And there's a lot of I'd say the the range of AI scores for YC at the top end is like a four point two, a four point two or four point three is a very, very high, almost automatic yes, down to maybe a two. That's kind of the range of scores I get out of Y, and that's just YC. So you can imagine non-YC, the range of scores I get one point fives and one point sixes. And then if if if it's above like a three point five, I'll I'll tend to kind of take a closer look at it and meet.
Martin Tobias (25:20.054)
Yeah, I think.
Brian Bell (25:28.824)
But usually the AI's gotten so good and I've trained it now over years and years and years that I very, very infrequently disagree with it, right? You know. It yeah.
Martin Tobias (25:29.922)
It's fast.
Martin Tobias (25:39.244)
Mm-hmm. Yeah. But it's because you've been iteratively, you know, redesigning it and adding new new new things to it over time. So would you say if if there were somebody else just getting started and trying to decide make investment decisions in in in startup companies or other capital allocation decisions?
Would you say that's a framework that has worked for you is is paying attention to your algorithm and refining it with data, you know, over time so that over time the decision gets compoundingly better, hopefully.
Brian Bell (26:10.734)
I think so.
Brian Bell (26:15.694)
I think I d yeah, and and it goes back to this like old phrase that's floating around is, you know, AI is not gonna beat you. It's a person using AI that's gonna beat you, right? Or take your job. And I think the same thing goes for venture, right? I think AI is not gonna replace venture capitalists. Like we there's still just so much hum human intuition and human relationships in this business. But, you know, a a VC powered by AI is a very powerful thing.
And so if you're getting started and you're thinking about allocating capital and doing syndicates or making angel investments, you absolutely should be using AI and you should be trying to work with it on every decision because it's going to teach you, right? It has 150, 160 IQ, right? It's probably smarter than you. And so you can plug stuff into it and say, what do you think about this investment? Right. And then start developing your own framework for things that you like and don't like and things that have been successful and unsuccessful. And you can build up your own.
Martin Tobias (26:53.534)
That's
Brian Bell (27:08.598)
algorithm over time.
Martin Tobias (27:11.878)
I think that is good at advice and I I haven't heard that from other in investors yet is that you need to build up your own algorithm, adding your own takes against the AI and and the data that you're doing to to align with what you think is related to returns and and you will be judged by those returns by your LPs in the future. But you know, when it when I started using AI, you know, about three years ago for for for
investment analysis, one of the things that frustrated me is that it told me to invest in fucking everything. It was terrible about it. And and and it wants to tell you what it thinks it you want to hear. So you I had to actually put in a bunch of adversarial stuff there. And and maybe you could share a little bit with some of the gremlins or some of the problems, you know, cause you know, of or challenges that you've had with
Brian Bell (27:47.95)
Yeah. Yeah, totally. I remember that. Yeah.
Brian Bell (28:06.798)
Absolutely.
Martin Tobias (28:09.36)
With getting AI to you know conform with the way you want to do investments versus the way it might want to do investments. I actually found in the beginning it to be hard to to to have it understand my thought process because it had it was very sycopantic even in wanting to do every deal.
Brian Bell (28:29.315)
Yeah.
Yeah, and especially this was a problem with earlier versions of Jat GPT, definitely four. was a sycophantic. You remember when they, you know, sunset that model, people were just all up in arms because they they missed their sycophantic friend, right?
Martin Tobias (28:46.469)
Yeah, yeah, it was just patting them on the back and gr positive reinforcing the you know, loop on whatever the fuck they said. That's a great idea, Martin. You're so smart.
Brian Bell (28:51.374)
Yeah, you're so smart, you know, like, gosh, that's a good idea. Yeah, you should like, you know, go yeah, you should have a snow shoveling business in in in LA. Yeah, and so we've we we've actually developed
Martin Tobias (29:02.565)
So so how did you design adversarial things in your in in your process? Yeah. Yeah market isn't.
Brian Bell (29:08.824)
Yeah, Mark Andreessen actually has a really awesome prompt. I don't know if you've seen it that I caught I kind of copied and paraphrased, but it's it's something around truth and accuracy and don't tell me what you think I want to hear. truth above all else. so I kind of copy that as a custom instruction into my AI. And that helps a lot because now it's it it it almost painfully tells me what I don't want to hear. in a way that I'm like, you're right. Like, oof, I really wanted to make this investment. But
Yeah, these three or four things are missing. and so it can be really painful as an investor because sometimes you want to be validated because you've already made the the the decision and the AI is telling you no, or vice versa. You're like, I don't think this is a good company. And and the AI is like, no, like look at these like five things that you're missing. And so a lot of that is it's almost like a thought partner. Genevieve our friend Genevieve thinks thinks of it that way. it's a thought partner, and it really is. it's this, it's this, it's this.
Martin Tobias (29:46.193)
And that's not
Martin Tobias (29:52.417)
And that's like now then all that easy.
Martin Tobias (30:00.399)
Yeah.
Brian Bell (30:06.94)
it's almost like it's it's the Team Ignite investment committee, right? I, you know, plug everything into the AI that's been trained on, you know, tens of thousands of deals and we have a conversation and try to come up with a consensus on what we both believe. And sometimes I override the AI and sometimes it overrides me. And it's it's a partnership in a way.
Martin Tobias (30:15.215)
So talk to me a little bit as we get to the end here about some of the decisions that you've made, you know, the checks that you've written. You know, you you you you use a combination of AI and your own pattern matching to to write the checks.
but not all the checks work out. So if you look at the bottom twenty percent, the ones that didn't work out, you know, do you have a process of going through those as to why they didn't work out and and re changing your decision process? Or is that just expected? I mean, do you di would you still write the check again to the bottom 20% that failed, given the information that you do at the time? Or, you know, how how do you think about the ones that didn't work out despite you having
The high conviction at the beginning.
Brian Bell (31:14.68)
Yeah, this this is what's fun about being an investor. Because, you know, we were talking before we started recording, is it's fun to think about how you make decisions, right? And there's this really good book, The Heath Brothers. Yeah, it's called Decisive. I read it like 10 years ago. Really great book on making decisions. and you know, you want to you know, widen your options and prepare to be wrong. And there's just like this framework of of how you make decisions. And one
Martin Tobias (31:27.541)
I haven't seen that one.
Brian Bell (31:44.001)
I love actually going back through the decision making process on successes and failures. So when I get a big, you know, hey, we just raised this massive, you know, 650 million series B, I'm like, wow, this is this is, I mean, I got in at a five cap, right? You know, so I'm sitting on a hundred X. Let's go back and like think about what we saw, what the AI thought back then and you know, how much better it's gotten. And I'll run through, I'll run through the algorithm, say, hey, imagine it's 2023 or 2022.
And here's all the information you have. run it through the the AI again, you know, score this whole thing. And and we'll kind of refine it. And I'm like, okay, well, you actually scored this a three point eight, but this turned out to be really successful. This won a hundred X for us, at least on a gross basis. so let's think about what you know, how we can improve and invest in more companies like this. And the inverse of that is what you just said, like this the shutdowns. and I actually I've developed eight gates in the model.
Martin Tobias (32:23.892)
Yeah, well.
Martin Tobias (32:29.492)
I broke it.
Martin Tobias (32:40.948)
Yeah.
it's that we can go.
Brian Bell (32:44.518)
there's eight gates that we've developed over the years from failures where like if this is not in there, if all these gates aren't at least at least like complete passes or partial passes, we won't invest. and these gates are kind of there to sort of protect the downside a bit. but there there's always like this like push and pull between it's it it's hard because even though I do a hundred investments a year, that's I'm only investing in about one percent of what I see.
Martin Tobias (32:50.384)
Yeah, multi-table active part of the test. there's a lot of like the whole thing. I'm gonna understand that I think it actually I think
Brian Bell (33:12.484)
about 50 Y I'm only investing in 50 YC companies a year out of eight hundred that go through YC. So it's still a really hard decision. There's eight hundred really amazing companies there. They all think they're gonna be, you know, the next stripe or whatever. And our job as capital allo allocators is try really tease out what what is that one or two percent that you're seeing that are really, really promising. yeah, and it it's fun. It's fun to kind of go back and learn, you know, take your lumps. And you know what I've what I've realized as an investor, you know,
Martin Tobias (33:20.17)
And then I'll be correct.
Martin Tobias (33:30.987)
Okay.
Okay. Yeah, it's fine, it's fine.
Brian Bell (33:42.519)
allocating other people's capital now for you know five or six years is man, I really didn't know what was doing back, you know, when I started. I I look at some of the investments I was making back then. I had no clue. You know, pre AI, pre, you know, all the lumps. There's this, there's this phrase in real estate that it takes a hundred investments, you know, to to make a real estate investor. In VC, they say it takes 10 million to make a VC. I think that's really true. I think you got to go deploy into like a hundred startups
Martin Tobias (33:50.436)
In the beginning. Yeah. Yeah.
Brian Bell (34:12.11)
Before you really know what you're doing.
Martin Tobias (34:14.602)
I would I would agree with that. So to wrap up here, like what is w you know, the two or three decision frameworks that you would give listeners to be better at low information decisions that you found worked for you and you've learned over the last couple of years?
Brian Bell (34:35.022)
I think is, you know, use AI, start training your own model and algorithm and weighted scorecard. try to go learn from history and and try to tease out, you know, because it's it's not just, you know, if you just do consensus investing, everything's gonna get bit up. I mean, you probably see this, right? When when there's a super hot round and everybody wants in, it's gonna be you're gonna be investing at a hundred, two hundred X forward revenue, right?
Martin Tobias (34:37.546)
Here's a guy.
Martin Tobias (35:00.848)
Yeah, think.
Brian Bell (35:01.612)
Maybe, maybe like infinite forward revenue. Some some companies now are getting huge rounds with no revenue. and so a a lot of what we do as early stage investors is try find that that sweet spot of kind of this sure bet that nobody knows about.
Martin Tobias (35:18.866)
So the counter be be f be counterintuitive and be right. So what is the one counter rate, what is the one mistake that you might have made in the past or that you see people making capital allocation decisions or early stage things still making today that you would tell people to avoid?
Brian Bell (35:38.285)
Yeah, I mean, that's probably gate five in our model, you know. and there's lots of different ways to tackle that question, but is there something really differentiated? Is there a differentiated wedge, a path to control like a control point, a path to a system of action, a system of record? Is is it have we actually identified the kind of business that it actually is and how it grows and what the particulars of that business?
Is it which trajectory is this company on? Is it distribution first or research first or or something else? And so really trying to understand like how how that business is going to compound. because really we're not in the business of investing in single million digit companies. We want a hundred hundreds of millions of revenue companies in the future.
Martin Tobias (36:24.08)
So the so the mistake that people make is investing in features instead of platforms, or what is the mistake that that investors can get sucked into?
Brian Bell (36:34.414)
That's that's a that is something to really look at twice when you're making an investment is is this a feature in somebody else's platform? Right. That's another that's another feature in our model is what's the bundling risk of the incumbents, right? Because sometimes you have a really nice feature, but it's not a platform. And I I'm I'm I'm sure that's why some people probably passed on Google. It's like, n nice demo, kids. And I think about that a lot.
Martin Tobias (36:48.655)
Lingerous.
Brian Bell (37:00.194)
You know, but but Yahoo, Yahoo has you know, is a billion dollar company and they're just gonna build it, they'll just kinda integrate that feature into their catalog of the internet.
Martin Tobias (37:00.272)
Well I I think about that a
Martin Tobias (37:09.284)
I think about that a lot and sometimes I s e even something that looks like a feature, if I believe I could c th that the there is a path to make it a platform, I would still invest. That's exactly what I did in DocuSign. DocuSign was actually very hard to raise money for in the beginning because people are like, that looks like a feature. Signing a contract on your phone. Adobe could copy that in five seconds. Why is is why is this a company?
And what ended up happening is they turned it into a platform with signing order and emails and they integrated to things and by but it took a lot of work to turn it from a feature into a platform and they got rewarded richly for turning it into a platform. So some things that might look like features can turn into platforms. So but then the mistake I I see a lot of people making is things that look like features and then they stay features.
Brian Bell (38:01.334)
Yeah, and that and that g gets into the founder quality vision question. You know, one of my favorite questions, like, what's your vision for the future? You know, five or ten years out, you've built this thing and it's really successful. What does the future look like? And you kind of wanna you wanna listen for that sort of language. the the vision's way beyond the future. the vision's way beyond this like initial wedge. they have this aspiration to build this this this massive company that
makes a huge dent in the universe, right, to use a cliche.
Martin Tobias (38:32.512)
Ab absolutely. All right, Brian, this has been very good for me. I learned a few things.
Brian Bell (38:37.544)
and one one thing I wanted to mention, because it's so hard to make decisions in startups, it's also really hard to invest in GPs like us. And so I'm I have a book coming out in September called Yeah, it's called LP, The Insider's Guide to V Investing in Venture Capital Funds. And I wrote it because I I I was an LP once and I didn't I didn't know how to evaluate venture funds. I actually invested in in a couple of really bad funds. I've interviewed a lot of LPs, really great GPs like yourself. And so I I literally wrote a book.
Martin Tobias (38:45.923)
Yep.
great.
Brian Bell (39:07.516)
it comes out September fifteenth at wherever you can buy books, Amazon and just wanted to mention that here.
Martin Tobias (39:14.322)
absolutely I would love to to read that. I had on earlier you can listen to the episode with Aram Atar. Well actually it's coming out out next week. He runs he he's an LP and he has this whole framework on how you pick GPs and he can explain that. I'd be interested how it compares to to to your thought. I'm an LP in a bunch of funds too. and and and I love being both a GP and LP. So you've got a book coming out. Where else can people find you if they're interested to learn more about Team Ignite? Twitter, LinkedIn, website.
Brian Bell (39:44.342)
Yeah, and you can find our on episode two fifty two of my podcast as well. and then you know, so yeah, we're the Ignite podcast, Team Ignite Ventures. so you can find us at teamignite.vc. We just rebuilt our website with AI. So go take a look at it. It has lots of information there.
Martin Tobias (39:48.854)
Okay.
Martin Tobias (40:00.482)
Okay. Well thank you for your time, Brian. Let's do more deals this year.
Brian Bell (40:04.121)
Yeah, likewise. Thanks, Martin.