CoBF equivalent for startups?

[LightWhale]

Hi guys,

 

I'm considering joining a ML startup as a founding member. This field of building young tech companies is new to me (my background is in theoretical science). During the last years when I was doing value investing on the side, CoBF and its members taught me much more about the practicalities of investing than any business school could have.

 

So I'm wondering of any of you knows a similar discussion board for startups?  Dealing with VC investors, finding HR & accounting vendors, buying insurance, issuing visas for foreign employees, and running cap tables are common themes that I'm about to bang my head into, and 10-15 pages of discussion about of each topic would probably save me endless days and mistakes.

 

Thanks!

 

 

 

 

 

[Jurgis]

http://www.cornerofberkshireandfairfax.ca/forum/general-discussion/start-up-ideas-where-to-look-for/msg363418/#msg363418

 

Yeah, I know the thread is about something a bit different, but Pelagic's pointers are pretty good for what you are looking for too. E.g. there are additional subreddit's closer to what you are looking for. YMMV.

 

Edit 2: you can also ask here, there are CoBFers involved in startups both on investing and on working-in-startup side. Although some of them don't post much. Personally I've done a bit on both sides (angel investing and working/running startups), but I have very little experience with "Dealing with VC investors, finding HR & accounting vendors, buying insurance, issuing visas for foreign employees, and running cap tables". I know a bit about ML startups. If you want to talk about anything, ask here or shoot me personal message for offline conversation. Good luck.

[KCLarkin]

Perhaps not exactly what you are looking for, but Hacker News might be a good place to start.

 

 

[LightWhale]

Thanks guys.

 

There's already a very solid idea of what the product should be. And funding is secured  ;) But seriously, we've already received seed funding guarantee for $5-6m. It's more about how to perform the everyday, non-technical practicalities that I anticipate to struggle with.

 

Jurgis, thanks for the offer, and I'll check out the subreddit forums you suggested.

 

 

[Pelagic]

Thanks guys.

 

There's already a very solid idea of what the product should be. And funding is secured  ;) But seriously, we've already received seed funding guarantee for $5-6m. It's more about how to perform the everyday, non-technical practicalities that I anticipate to struggle with.

 

Jurgis, thanks for the offer, and I'll check out the subreddit forums you suggested.

 

One thing I'll recommend to familiarize yourself with a lot of what you mentioned in your first post is Startup School by Y-Combinator. I think the summer session is coming up next week and you can still register. If not all of last year's sessions videos are online on YouTube. They go over a ton of different start up related topics so you can choose the lessons you want and watch them.

 

If you actually sign up for the course you'll have access to their online discussions and be placed in a group with other startups.

 

https://www.startupschool.org/

 

Congrats on the funding, that's a good chunk of change for a seed round.

[LightWhale]

That's a great tip, thanks. I'll have a look at their youtube channel. We got enough funding to have breathing room in case the world falls apart in the next 18 months, but we took it from investors with less infrastructure than traditional VC's. 

[LC]

Lightwhale, any thoughts on ML applications either generally or in specific cases? We find they are, in some popular use cases, essentially momentum strategies.

[LightWhale]

you mean ML apps in investing? I don't see a reason that correlations would zoom in only on momentum, but it's not my field.

 

Ultimately it would depend on the specific methods employed. ML is a fuzzy term, a suite of dissimilar methods of search ranging from generalised linear models to KNN, regression trees and deep neural networks, which are nonlinear. Plus the results will be heavily reliant on the input quality, where it is very hard to beat the big funds. 

 

Does it answer your Q? if you could be more specific, I'll try to help more.

[Jurgis]

you mean ML apps in investing? I don't see a reason that correlations would zoom in only on momentum, but it's not my field.

 

Ultimately it would depend on the specific methods employed. ML is a fuzzy term, a suite of dissimilar methods of search ranging from generalised linear models to KNN, regression trees and deep neural networks, which are nonlinear. Plus the results will be heavily reliant on the input quality, where it is very hard to beat the big funds. 

 

Does it answer your Q? if you could be more specific, I'll try to help more.

 

There was a whole session on ML in financial data analysis at ICML 2019 ( https://icml.cc/ExpoConferences/2019/schedule?talk_id=6 ) from Point 72 / Cubist ( you know who: https://www.point72.com/ ). As a good industry presenter though he mostly raised the questions rather than telling what they do/use/apply/etc. He did cover investing for periods longer than a day 8) but did not talk much about periods longer than couple months. So your buy-forever investments might still be safe from AI/ML. I would not try to compete with these guys in day trading or even month trading. Unless you work on illiquid securities that they don't touch.

 

He talked about the issue of knowing whether a strategy/algorithm is temporarily underperforming or lost alpha forever. I asked him about this for more detail and his answer was pretty reasonable: if underperformance is similar to what was seen in the past, then continue; if not, then likely pull the plug. Of course, this is on very general level.

 

There was a lot of recruitment too. Mostly from quant funds. Bloomberg was there (small booth with one rather bored guy), Citadel too. I can try to look up / remember other names if someone is very interested. As a true value investor I just took the swag.  8)

[LC]

you mean ML apps in investing? I don't see a reason that correlations would zoom in only on momentum, but it's not my field.

 

Ultimately it would depend on the specific methods employed. ML is a fuzzy term, a suite of dissimilar methods of search ranging from generalised linear models to KNN, regression trees and deep neural networks, which are nonlinear. Plus the results will be heavily reliant on the input quality, where it is very hard to beat the big funds. 

 

Does it answer your Q? if you could be more specific, I'll try to help more.

IMHO and in my experience ML techniques are better utilized in the consumer space, particularly consumer credit. These models provide some lift over traditional logistic approaches BUT I would argue it is still requires more time to manage stability concerns. Even forgetting regulator concerns, boosted trees, neural network variants and such are so difficult to understand you do not know whether this lift is attributed towards additional intuitive patterns being exploited, or random chance or some other factor which may not be relevant. Interpretability is a real concern.

 

On the trading side there is obviously much more opportunity, but this is why these models will IMHO have a more difficult time adding value here. One these markets are so picked-over it is ridiculous. Every intelligent statistician wants to design trading models. Two the majority of buyers are sophisticated. Compare the average derivative buyer (i.e. a highly educated bank/fund trader) to the average credit card buyer. Three, securitized financial products have more generally-accepted pricing mechanisms. (BS, Heston, etc.) What does the ML model add here? ML models excel at finding nonlinear patterns, and perhaps you can combine this with a traditional BS approach but you still run into the problem of interpretability.

 

Obviously I am speaking in hindsight here but it has become clear why ML techniques are having a difficult time on the trading side of the business. In this sense, I think a lot of the folks who jumped into this business based on the glamor of combining their ML experience either from a digital or consumer credit space towards a trading landscape, they have found themselves in front of something of a brick wall and are withdrawing their claws and turning heel so to speak.

 

Another area where ML approaches perform very well is in fraud use cases. Fraud techniques change rapidly (remember the old Nigerian Prince scam?) and fraudsters are adept at gaming basic systems. Easily you can see why the ML model excels here: it is highly adaptive and specializes in pattern seeking. As with all modelling you are not always looking for the best short-term fit but you are looking to match a particular methodology's strengths/weaknesses with your problem/use case.

 

One big drawback of ML models is managing overfitting. These models have traditionally shown very real stability problems out-of-sample. Regularized learning intervals is one method used to manage such risk, however this leads to "momentum" as I alluded, as the model is regularly updated to take into account and weigh more heavily the most recent data.

 

Curious what your thoughts are in terms of overfitting and managing stability in your particular use case, or prior use cases where you may have had experience.

[oddballstocks]

Startup School is a good idea.

 

Reality is this, get involved in your local community.  Most cities have a startup scene.  Retired execs flock to these meetups because they're looking for new and exciting things to do, plus companies to invest in.  There is a LOT of free advice floating around from people who have built multi-billion dollar companies.  But it's not on message boards.

 

The upside is most of these connections also love to get together for lunch or coffee and help you.  Again, not online, but 1000x more valuable than online. 

 

You guys have funding, but do you have revenue?  Have you proven you can commercialize the product?  Sales and a strong pipeline solves a lot of problems.

 

If there's one thing I've learned from running a company it's this: product doesn't matter much, being able to connect with people and meet their needs is what matters.  You could have a very inferior solution, but if you can network and sell you'll do a lot better than smart people with a better solution.

[Jurgis]

Startup School is a good idea.

 

Reality is this, get involved in your local community.  Most cities have a startup scene.  Retired execs flock to these meetups because they're looking for new and exciting things to do, plus companies to invest in.  There is a LOT of free advice floating around from people who have built multi-billion dollar companies.  But it's not on message boards.

 

The upside is most of these connections also love to get together for lunch or coffee and help you.  Again, not online, but 1000x more valuable than online. 

 

You guys have funding, but do you have revenue?  Have you proven you can commercialize the product?  Sales and a strong pipeline solves a lot of problems.

 

If there's one thing I've learned from running a company it's this: product doesn't matter much, being able to connect with people and meet their needs is what matters.  You could have a very inferior solution, but if you can network and sell you'll do a lot better than smart people with a better solution.

 

QFT everything oddball said.

 

Re revenue: startups are in a bubble (oh no B-word!  8) ) too. Pretty much anybody who is anybody is looking at ~$6M pre-money with no revenue. And that's just at the angel level.

[LightWhale]

startups are in a bubble only if you think public markets, and FAANG stocks in particular, are in a bubble.

The need for greater funding comes from inflated labour costs. We compete with rising salaries in those public companies. Hiring a proper engineer for less than $200k (and % equity) is a real struggle. 

 

Re

I would not try to compete with these guys in day trading or even month trading. Unless you work on illiquid securities that they don't touch.

 

And with illiquid stocks the risk of overfitting is alarming.

 

 

You guys have funding, but do you have revenue?  Have you proven you can commercialize the product?  Sales and a strong pipeline solves a lot of problems.

 

re revenues, we haven't even started building the product. But we've gone through the PoC process with some potential S&P 500 customers. That's no reason to celebrate though, and chances are still skewed towards failure. Thanks for the community tip. Real-life connections are indispensable. I'm in touch with various entrepreneurs, but it's uncomfortable to always be on the asking side. 

 

 

IMHO and in my experience ML techniques are better utilized in the consumer space

 

It's probably obvious to say, but shallow learning might work better for trading than deep learning, because of the low signal-to-noise and the moving goalpost inherent in asset pricing problems. Plus, if the main factor is momentum, maybe it's because these models try to forecast changes in the risk premium? But I know far less than you as it sounds, so i'll try to stick to what I know :)

[Jurgis]

startups are in a bubble only if you think public markets, and FAANG stocks in particular, are in a bubble.

 

I disagree on FAANG. FAG stocks are mostly cheap based on a number of metrics. AN are not cheap, but not as much in a bubble as people claim.

You might be right if you mean that other tech IPOs with no earnings are in bubble. This might be the case.

 

The need for greater funding comes from inflated labour costs. We compete with rising salaries in those public companies. Hiring a proper engineer for less than $200k (and % equity) is a real struggle. 

 

Wait, they want 200K salary and % of equity from a startup?  ::)  ::)

 

And here I was ranting about entitled CEOs. (Well the CEOs are still more entitled.  8) )

 

I'll have to tell this to my friends and family who are working on a second startup for free...

 

When I worked at a startup, I got 1/X of that. And the founders took no salary at all. It was self-funded overall.

 

Anyway, glad you have the money, good luck.  8)

[LC]

Wait, they want 200K salary and % of equity from a startup?  ::)  ::)

They are absolutely worth it. Silicon valley investors are a dime a dozen and add little differentiated value. Quality engineers on the other hand are a rare commodity. If your friends/family are truly the latter they should be out on the west coast getting paid.

[Jurgis]

Wait, they want 200K salary and % of equity from a startup?  ::)  ::)

They are absolutely worth it. Silicon valley investors are a dime a dozen and add little differentiated value. Quality engineers on the other hand are a rare commodity.

 

Maybe they are worth it maybe not. I have a feeling I know people who you'd value more than 200K (well, at least I'd value them more than 200K) and who worked at startups for way less (plus equity of course). Though maybe you're superselective and I'm wrong.  ;)

 

But hey, if that's the situation that's the situation. I'm sure it's nice to work at a startup with full salary. It's like a lottery where you cannot lose: you get paid and you might hit the jackpot.

 

Of course, there's the cash burn but that's what the leeching SV VCs are for, yeh? They'll fund another round if you need more cash. 8)

[LightWhale]

It's a matter of opportunity cost.

How much do you think such engineers make at Google/Facebook/Amazon?

 

[Jurgis]

It's a matter of opportunity cost.

How much do you think such engineers make at Google/Facebook/Amazon?

 

You did not say what "such" is. So either let's talk exact qualifications or you tell me how much they earn at G/F/A? 8)

 

Although I'll say this: if a run of the mill, not superstar engineer earns more than 200K at G/F/A, they are overpaying.  8) If you are talking about someone of the caliber of Jeff Dean or Demis Hassabis before their full ascension but with demonstrated work/qualities/brilliance, then sure they likely would get more than 200K.  8)

 

Edit: 3 year old data: https://www.quora.com/What-is-the-salary-of-fresh-PhD-machine-learning-at-Google-Facebook-Apple-in-2016

https://www.quora.com/What-is-the-salary-of-senior-researcher-at-Google-Facebook-Apple-in-2016

Glassdoor/Paysa have more up to date numbers, but these are IMO more iffy. They seem to be in similar ballpark as Quora's though.

[oddballstocks]

I know engineers at Google/Amazon.  I know a few at Google, they're smart, but not earth shattering smart, I'm sure they're making over $200k.  I know plenty of people who are technically much better who earn less.

 

I worked with a guy who went to Amazon.  He was terrible, absolutely terrible, couldn't code his way out of a wet box.  Studied like crazy, went to a top MBA school, was hired at Amazon.  He played the political game well and is now in charge of a massive division that most probably interact with.  A friend kept in loose touch, said nothing had changed about him, he didn't get smarter, just found an environment where politics was valued over output and he keeps moving up.  This was the type of guy who's show up at 5am if the boss arrive at 5:30am and stay to 11pm mindlessly doing nothing, but sitting there to show he was "committed".

 

Also know engineers at Uber doing the self driving stuff.  Same thing, smart, better than average, but not mind numbing smart.

 

Mind numbing smart.... I have met a few working at startups for almost nothing.  People who are inventing new types of materials, or doing wild things with AI.  I have a feeling if you told them they'd be provided food and have a place to sleep they'd work on this stuff for free.

[Jurgis]

I know engineers at Google/Amazon.  I know a few at Google, they're smart, but not earth shattering smart, I'm sure they're making over $200k.  I know plenty of people who are technically much better who earn less.

 

I worked with a guy who went to Amazon.  He was terrible, absolutely terrible, couldn't code his way out of a wet box.  Studied like crazy, went to a top MBA school, was hired at Amazon.  He played the political game well and is now in charge of a massive division that most probably interact with.  A friend kept in loose touch, said nothing had changed about him, he didn't get smarter, just found an environment where politics was valued over output and he keeps moving up.  This was the type of guy who's show up at 5am if the boss arrive at 5:30am and stay to 11pm mindlessly doing nothing, but sitting there to show he was "committed".

 

Also know engineers at Uber doing the self driving stuff.  Same thing, smart, better than average, but not mind numbing smart.

 

Mind numbing smart.... I have met a few working at startups for almost nothing.  People who are inventing new types of materials, or doing wild things with AI.  I have a feeling if you told them they'd be provided food and have a place to sleep they'd work on this stuff for free.

 

I'm not sure why I post anything. Whatever I try to say, oddball says it better.  8)

 

Another QFT.

[LightWhale]

If you know mind numbing smart guys who do AI, are worth $0.5M in annual salary, and would truly work for free, why don't you send them over? I'm serious. But until that materialises, forgive me for remaining doubtful.

[Jurgis]

If you know mind numbing smart guys who do AI, are worth $0.5M in annual salary, and would truly work for free, why don't you send them over? I'm serious.

 

That's not how it works and I believe you know that that's not how it works.

 

You have to build your network, you have to sell your vision, you and your startup should be attractive to such people. It's not a supergeeks-on-demand service, especially since we pretty much know zero about you or your startup.

 

In the best case you should have had eye on the people who are good/great way before (co?)founding the startup and reached out to them now that you have the startup. You possibly should have participated in local tech scene (which might be small or huge depending on your location - and you probably need to be smart/selective/etc. if you're in SV where everyone's in tech haha).

 

Is this easy? No. A lot of startup founders don't do it. But those who do it, IMO have a huge advantage.

 

If you hire using standard big-company methods, then yes, you are at disadvantage especially in tech upturn/tight labor market.

 

In general: don't sell salary and equity; don't sell the foosball and beer (if anyone still sells that lol); sell the vision, sell the product, sell the team. Yes, there are bunch of people who will go and work for less if the the vision/product/team is very attractive. But it's gonna be harder if you have to cold sell them through ads/hiring agencies/etc.

 

You could post your sales/hiring pitch here. Not sure that would yield something, but it probably won't hurt. Unless you are very much into stealth mode and you don't want to do that. Then you could send it privately to whoever.

 

Good luck.

[Jurgis]

Alternatively, you can hire outside SV (if that's where you are located). But that comes with it's own set of issues, so possibly not worth discussing unless you are considering it.

[oddballstocks]

Completely agree with Jurgis.  These guys travel in circles, move from one company to the next, everyone knows everyone.

 

The best jobs are networked, the best developers are networked.

 

I'm part of the local tech scene here in Pittsburgh.  I'm on a private mailing list.  I get emails from CEO's of cutting edge companies probably once a week saying "Someone I know, vouch for and highly recommend is looking for a job, here is their info."  No one knows they're on the market, their employer probably doesn't and it isn't public.  Heck most of the jobs these people fall into probably didn't exist before someone said "hire x they're awesome."

 

The same is true when a company disbands, I'll see "let's try to find them jobs" and a bunch of people are placed quickly.

 

This isn't rocket science, this is how 99% of the business world works for quality talent.  For average talent there are job postings.

 

Jurgis is also right about the vision and job.  You could offer a lot of people, myself included $500k a year to sit in a drab office, and constantly change Excel cells from blue to red, and back to blue and I wouldn't take the job.  Maybe I'd suck it up for a month or something, but I just couldn't.  When you say "we'll pay $500k for top talent" you attract a different type of talent.

 

What I've found is this.  For top talent pay is secondary, if the pay is good enough they don't care about it.  If they are looking for $150k and you offer $155k or $160k the difference just doesn't matter.  It's the projects, the vision, the team that matters.  Some will work for almost nothing to work on a high performing team.

 

The fact that you haven't stumbled into this sort of tells me you haven't brushed up against this world at all.

 

You're in a tough position too because you have money.  When you have no money but an awesome idea you have to sell like crazy to get people on board.  In your case you have money so you're trying to substitute that for passion.  The lack of resources at startups fuels innovation because people need to be creative.

 

I'd recommend coming to Pittsburgh and interviewing people from CMU, some excellent talent that will work for less than half a million.  I mean Uber hired away all of CMU's top of the industry robotics researchers for less than $500k apiece, I'd guess they're mid $100ks.  You can also buy a really nice house here for $400k or so.

 

Here's an example.  This is a local company building self flying helicopters.  Top team, if you're looking for earth shattering AI this is it, you're moving in three dimensions, will be used in military theater.  The firm has received significant investment from the military and Boeing.  This is their glassdoor: https://www.glassdoor.com/Salary/Near-Earth-Autonomy-Salaries-E1481640.htm

 

They're hiring senior engineers for $111k.  A robotics software engineer is making $79k.  I met a former CEO from a military contractor, her company built missile guidance systems, satellites, stuff she said was much cooler that she couldn't talk about.  She spoke glowingly about them, said from what she knew it was a top team and they are going to solve this and should expect a flood of money at some point in the future.

 

That's a big problem, high profile, awesome backing, and look at the salaries.

 

I don't know what you're building, but it probably isn't as cool as self flying helicopters.  Maybe it is, but likely not.  If they're able to get a top shelf team for what everyone says on glass door as "average or below average salaries" then my guess is your problem isn't talent, it's the sales pitch.

 

[Jurgis]

Man, I am moving to Pittsburgh tomorrow today.

 

...

 

...

 

Never mind.

 

But oddball had me there for a while.  8) ./bow