Hatcher+ Logo Hatcher+
Login

Hatcher+

Blog
AI vs. Human Reasoning in Venture

Three years ago, we started aggregating hundreds of thousands of venture deals and events, with the goal of using deep learning to create an investment model capable of generating predictable returns from venture.  Two years ago, we began using machine learning to analyse applications for funding across our network.  This year, we developed a similar, data-driven approach to determining which applications would best fit an impact-driven portfolio.  

Blog Image

The questions we often get asked by investors are: what role does human reasoning (still) play in executing the Hatcher+ investment model?  Can AI be used to replace human reasoning in the deal selection process?  Does AI have a role in the development and execution of startup businesses?  Is it conceivable that ideas will one day be created and executed by AI-powered "innovation algorithms" and process management tools, rather than by founders?

Quantitative Modelling for Venture
Quantitative investment modelling has been around for decades - and is now so widespread it had become the de facto method of trading public equities, and informs decisions inside even (especially?) the most complex asset classes.  Alex Foster estimates over 90% of trades (worth an estimated USD1 trillion annually) are initiated by quant-based algorithms. 

Within the past decade, some smart folks at 645 VenturesCorrelation Ventures, EQT, First Round Capital, SignalFire, and others, started building the first quantitative investment models specifically designed for venture.  To date, the finding most common to this modelling has been: size matters.  Because while some small portfolios do outperform, it turns out that large portfolios outperform consistently, and consistency is what makes for a robust investment model. 

The analysis engine we've spend three years building now contains over a half million events - the largest of its type in the world - and our core findings show that Dave McClure's initial hunch in starting 500 Startups was absolutely right: Larger portfolios benefit from power curve effects on a far greater and more consistent basis than smaller portfolios.  And you don't need a degree in advanced math to explain why: the power curve effects that kick in when several startups go 100x (or more) are, statistically, like rare butterflies: a larger net can help.   

Interestingly, respected independent industry researchers such as INSEAD's Claudia Zeisberger and Monisha Varadan and Loyal Ventures' Kamal Hassan have come to the same conclusion (shout out to Michael Jackson, who does a tremendous job of reviewing and aggregating this emerging data.)  And with over two thirds of AUM residing with those venture fund managers with 500 or more investments in their portfolios, LPs would seem to have been strongly in agreement with this thesis for some time.

Like all good quantitative models, our investment model is data-driven, and was built using a massive amount of data.  At a high level, we use it as the basis for constructing our portfolios.  The alternative portfolio construction approach most often traditionally used in venture - organic, resource constraint-based portfolio building - no longer appeals to us.  

Winner: AI

Intelligent Selection
As every venture investor or accelerator operator well knows, business plans and ideas vary widely in quality.  Some remove minutes from your life that would have been better spent perched naked on a cactus.  Others introduce you to a new problem set you didn't know existed, a new audience, a new way of looking at the world.  And then very occasionally, something comes along which is so audacious, so good, and so potentially valuable that you can't think about anything else for days.  

We've done some research into the selection process at work at over 160 accelerators and early-stage venture studios, and based on this research, we use a 1% Selection Ratio as our basis quality threshold when analysing deal flow.  Sometimes this skews higher, when working with a novel specialist accelerator invested in hardware, medical devices, or a similarly narrow field - but by and large, we've found that the 1% rule puts us in good company when it comes to how leading investors select. 

We've spent years training our AI and machine learning technologies to spot badly-written business plans, unfixable logic or founder gaps, and positive trends and characteristics common to those plans most commonly selected by the human assessors working in the venture studios, accelerators, and angel networks on our platform.   The goal of our AI has not been to exclude humans from the process - far from it.  Our goal remains this: Make the process more efficient by allowing the human assessors to more easily identify and isolate the top 1% of submissions. 

You could call it "anti-selection".  Whatever we call it, it's become an increasingly necessary part of the selection process - based on volume alone.  Some of our early-stage deal origination partners receive over 1,000 business plans for every cohort they run - and some communities receive in excess of 10,000 submissions for cohorts that will eventually feature just 10-20 startups/founders.  Allowing these selection teams to potentially focus on the top 5%/10%/20% of plans allows them to spend more time looking at quality submissions, and talking with quality founders.  That's a good use of AI.       

[Dan suggests users of our tools should think of our Hatcher+ Score as "one voice at the table", rather than a recommendation.  I agree - but as the tools continue to improve, we may have to revisit and revise this characterisation...]   

Winner: Humans (the advantage goes to those using AI)  

Impact Readiness
Impact investing - and the scoring of the likely impact of a company or technology - is an area where there has been a massive amount of research going on.  Julie Muraco of the Triple Bottom Line Group recently told us on a call that she sees a new impact scoring system proposed "every two weeks".  On the other hand, several large impact investor groups have recently confirmed to us how difficult it has become for the groups to assess the likely impact of an investment and allocate funds, using the existing tools available.

How is our Impact Readiness score different?  For one thing, we're focused on much-earlier stage companies than most people looking at this space.  And for another, we recognise that impact is not a simple on/off (digital) concept: depending on how you view it, there are at least five "buckets" into which any impact investment might fit.  Add to that 17 United Nations Sustainable Development Goals, and multiple differing investor mandates, and you start to get a sense of the scale of the problem: and why AI might just be the most useful way to solve it.  

Right now, rather than looking to determine the level of potential impact - something difficult to do at later stages, let along the early stages where we invest - we are focused on the determining the best "fit" from the founder-investor perspective.  The question we're seeking to answer is essentially an extension of the question we ask on behalf of our venture studio and accelerator partners: is this deal - and team - a fit with the investor mandate? 

One more thing: One thing we're in the early stages of adding is AI-assisted video analysis - so we can help you get a better view on founder personalities and presentation styles (something COVID is making increasingly difficult), and also potentially spot falsehoods, including fake submissions, testimonials, references, etc - something that we've found is quite important to reputation-sensitive impact funds (if you don't think fake references and manufactured hype is going to become a larger problem at some point, then you probably should check out the testimonials that can be generated for $55 on Cameo.)

Winner: Humans (useful tools are emerging, but still nascent)

AI as an Innovation Engine
I tackled this final question in a recent blog (Can a Robot Write a Business Plan?) - and suffice to say that while AI can do a pretty good job of mimicking the process of writing an executive summary, or the brushwork of famous Chinese artists, or the intricate movements required to pilot a fighter and down an enemy plane, the creation and execution of truly innovative ideas by non-humans is still decades away - especially when it comes to the execution part. 

[Unlike an A-380, a startup starts with no controls.  No processes.  And only a vague flight plan - often to an unknown destination market of undetermined size, that is only reachable via a change of course (otherwise known as a 'pivot'), several refuellings, and sometimes, a change of pilot.  We're still some ways away from an AI-powered daily standup meeting - and that's probably a good thing!]    

This prediction might not age well - but I think founders and innovators will be safe from meaningful competition for decades.  Because while ideas might be possible before then, the complexities involved in both inventing and subsequently executing a business plan involve permutations and physical interactions on a scale that goes far beyond even impressive recent achievements like Alpha Go.  Can AI help us innovate?  Of course - and it already is assisting us massively in finance, health, and everything and anything involving physics, chemistry, visual and audio design, audience analysis, marketing, you name it. 

Winner:  Core innovation?  Deep tech?  Ideas?  For now (and for the foreseeable future), we're going to keep investing in humans, via humans.     

Sharp
John Sharp
Hatcher+

John is a serial entrepreneur and investor, and the Founding Partner of Hatcher+, a next-generation, data-driven venture firm that utilises a massive global database in combination with AI and machine learning-based technologies to identify early-stage opportunities in partnership with leading accelerators and investors worldwide.