Where do Unicorns come from?

Boardwave are on a mission. Over the next 10 years they want Europe to become a global superpower in software.

The UK, as third largest tech economy in the world, are marketing themselves as Unicorn Kingdom. Unicorn is the term used in the venture capital industry to describe a startup company with a value of over $1 billion.

As the chart below shows, Europe attracts as much money in early stage ventures as USA, but as round size increases and business grows, Europe falls away.

Looking at 2020 McKinsey data, when it comes to the proportion of software and software-enabled companies among the region’s 100 most valuable companies, Europe (7%) significantly lags USA (39%) and Asia (25%).

USA (specifically the San Francisco Bay area) tend to have a more collaborative ‘move in packs’ spirit when it comes to scaling software businesses. Paypal Alumni have gone on to found a mind-blowing number of successful tech companies. Boardwave is looking to directly address the apparent lack of a quid-pro-quo mentality within Europe. Not easy with language, cultural and geographic barriers.

Relevance of AI to scaling software businesses

Before AI came along we had good old deterministic software. Fixed inputs, fixed outputs. Happy days. Predictable revenue (happy CFO) and minimal risk (happy General Counsel).

In 2019-22 I observed most of my startup/scale-up clients claiming that AI was a ‘core’ part of their solution and/or integral to the scaling potential. Music to the ears of investors. The translation - this business is defensible and can scale in a low-opex manner. Part of my job back then was to lead teams of consultants, data scientists and ML engineers to get under the hood of a business and smell whether this narrative was BS. Some examples of simple conclusions I encouraged the team to think about:

1 - This business is genuinely defensible due to the proprietary data that they have gathered, but the sophistication of the algorithm/s is irrelevant.

2 - This application of AI (say, a propensity model driving a prediction) could be easily replicated by a competitor in 6-12 months.

3 - There is no AI. This is a deterministic (rules-based) business. The commercials of the business are sound but they will need to turn this into an AI solution in the next 12 months for accuracy, scaling and defensibility.

The art of spotting a great software investment is understanding the relevance of proprietary data and algorithms to competitive advantage and ability to scale.

How important is Generative AI if you want to become a Unicorn?

A company such as LinkedIn (became a unicorn in 2011), was a pioneer of AI software. AI software is probabilistic which means it draws inferences about the world based on the data that it sees. This introduces a lot more complexity and risk (operational, legal, ethics, explainability - this is often grouped into a term called ML-Ops) for engineering teams. Companies such as LinkedIn cracked this long ago. The majority of software businesses still struggle significantly with this.

Then Generative AI came along. It’s on all CEO’s agendas. Why else would Elon Musk fly in to Luton? In fact, Generative AI has become so important, so quickly, that FTSE 100 CEOs are disgruntled at being invited before being uninvited. Even 12 months ago, the idea any FTSE 100 CEO would have FOMO on an AI conference would have been laughable!

Generative AI presents new and very different challenges to ML-Ops (increasingly referred to as FM-Ops and LLM-Ops). The thorny ethical and legal issues, which need to be resolved to help LLM-Ops cross the chasm from tech blogs to the day-to-day realities of commercial leaders, are being debated by Elon et al at Bletchley Park today.

So you can think of the aspiring commercial leaders in attendance today as having a head-start in the Unicorn race. Need further convincing? Anthropic and Databricks are already in the top 10. The fact that Anthropic were founded in 2021 and are valued north of $20bn helps us to understand why Rishi Sunak is - going live with Elon.

Welcome to Unicorn Kingdom. Turn left for light-touch-regulation rollercoaster, turn right for ethical epcot. Alternatively, head straight on for a talk from the head zookeeper on how the UK can help you defy the odds in actually becoming a Unicorn.

How relevant was AI to European Startups in Q4 2021?

So what’s the best data source for understanding promising European startups? Startup Heatmap List tracks over 2,000 participants of Europe’s best early-stage accelerators every year. This seems like a decent dataset to take a look at.

This is a spreadsheet of 585 startups with the following 22 columns of information: Accelerator Program, City of Accelerator, Country of Accelerator, Typical funding by accelerator, Name of Startup, Industry of Startup, Description of Startup, Verticals of Startup, HQ City of Startup, HQ Country of Startup, Origin City of Founder, Origin Country of Founder, Name of Founder, Founder has tech skills, Founder has founded before, Founder has over 10 years managing experience, Accelerated in year, Website of Startup, Year Founded, Number of Employees, Gender of Founder, Founder LinkedIn profile.

Industry of Startup - 6% ‘Artificial Intelligence.’ 1.5% ‘Data & Analytics’. 13% ‘Information Technology.’ 1% ‘Software.’
Verticals of Startup - 13% showing as ‘Big Data.’ Annoyingly, ‘AI’ is not a drop-down.
Founder has tech skills - 312 / 585 = 53% showing as ‘Yes'.

Inconclusive. The number is clearly higher than 6%. However, very hard to gauge what the true number is. At this point I decided to upload the 585 website URLs to GPT-4 (web browsing) and ask for help. I was on the edge of my seat when it responded ‘I have started investigating the usage of AI or ML in the startups listed. I will continue investigating the remaining startups and compile the findings into a summary spreadsheet for you.’

Then this happened …

As an aside, working with GPT-4 right now feels like working with a bright and enthusiastic intern who not only spectacularly under-delivers but also needs a check-in every 5 minutes. Or perhaps a retired 90-year old omniscient University professor who falls asleep with alarming regularity.

How does this compare to European Startups in Q2 2023?

Analysis from a Q2 2023 spreadsheet in an almost identical format from Startup Heatmap:

Industry of Startup - 2% ‘Artificial Intelligence.’ 1% ‘Data & Analytics’. 8% ‘Information Technology'. ‘22% Software & Technology.’
Verticals of Startup - 8% showing as ‘Artificial Intelligence.’
Founder has tech skills - 312 / 585 = 49% showing as ‘Yes'.’

Can GPT-4 help us with this 21 vs 23 analysis?

In short, MASSIVELY! See the end of this article for a summary of the conversation (with GPT Advanced Data Analysis) that helped me get to the below answers:

2021 AI Startups - 39% (228/585)
- 2021 Generative AI Startups - 1.7% (10/585)
2023 AI Startups - 60% (376/628)
- 2023 Generative AI Startups - 6% (37/628)

Although the analysis was a little crude (the most obvious limitation being the likely inflated numbers driven from searching ‘AI’ that probably picked up words such as ‘Blockchain) and the dataset limited, with a little help from GPT-4, I was able to identify that AI (and Generative AI) as a percentage of all startups appears to be on the rise and therefore increasingly relevant when on the lookout for the next European Software unicorn. Do not forget that accelerators only take the top 1%-3% of entrepreneurs.

Can GPT-4 replace the role of an analyst right now?

No.

Just how good is GPT-4 then?

A number of reasons why the technology is tantalising …

Speed and efficiency - it helped me get to a ‘good enough’ answer in less than an hour. Nothing mind-blowing right now, we are talking 2x/3x levels of efficiency improvement. No reason this won’t become 10x/20x if we imagine this technology in 12 months. Everybody should be experimenting with it.
Proactivity - it is incredibly willing to proactively make suggestions once it understands the goal. More often that not, these are not stupid suggestions. Part companionship, part cognitive inspiration.
Explainability - it is less black-box than you might imagine. Interpretability and reasoning will be crucial to the adoption of new analytical workflows in the future.
Grunt work - GPT-4 naturally and effortlessly cleans / merges / suggests data column titles. This data engineering task is high-value and a really important pre-requisite for any analytical task or downstream data science modelling.
Functionality - right now functionality is not joined up. You can’t upload a file and then ask for information about that file available on the Internet (you need to switch chat windows). Once these tasks are joined up as one use-case (or ad-hoc through plugins) - the value of the technology will be evident and adoption exponential. Not convinced? See how close I was (below) to getting the answer I was looking for in searching for more information on all of these startups on LinkedIn. The third screenshot tells me to do the work manually!

So, can GPT-4 predict the next European software Unicorn?

Why don’t we ask it …

We have a winner … BEMYCAR.
What’s really interesting is that the language for this startup is actually written in Spanish. Here’s the full translation which GPT-4 did automatically for me. Another time-saving task I didn’t initially even realise it had done.
Bemycar is a hyperverticalised, scalable, diversifiable and internationalisable technology company with the main mission of ensuring that companies in the sector sell more and offer a better digital experience to their potential clients through our automatic enriched lead qualification process supported by our chatbot. and in our community of more than 3,500 car owners, willing to help leads with their questions about any vehicle. With the possibility of integrating our technology into the client's CRM, we are able to qualify, enrich and classify 100% of your leads, eliminating the commercial effort involved and guaranteeing the best user experience. In addition, we are specialized in the automotive industry and our service works in real time, 24/7, allowing salespeople to focus on what they do best: sell. Thus, through any of our flexible monthly plans and applying our technology, we get the different companies in the industry to manage qualified leads enriched with real purchase interest, filtered and classified in order of conversion probability, instead of raw or poorly qualified leads. and of low quality.
A bit of fun? Yes.
Is BEMYCAR truly a Generative AI Startup? Looking at their website, yes. Looking at their LinkedIn, on the way.
Ability to understand factors you care about (in this case Generative AI) and conduct further logical analysis? Yes.
Ability to give explainable answers based on reasoning? Yes.
Ability to consider external market trends even though it was in ‘analysis’ mode not ‘Bing’? Yes.
Ability to understand the relevance of the memory of the conversation? Yes.
Ready to be used by investment analysts to take some bets? No. Largely because it heavily indexes on keywords you suggest as the answer.
- But … I bet companies like Prosus have already built and tested this type of customised solution.

Disclaimer - this is an answer generated form GPT-4 and should not in any way be interpreted as any form of investment advice / recommendation from AI Strategy.