Over the past year, we’ve had the pleasure of welcoming a number of leading Machine Learning (ML) focused companies to the Notion family. These are startups that are tackling verticals ranging from autonomous vehicles (Five.AI) to behavioural monitoring within the enterprise (Status Today). What’s more, many of our existing portfolio companies are now using ML to power aspects of their products and delivering accelerating returns as they go. As a B2B fund with a strong history in SaaS, we keep an eye out for emerging technologies that have the potential to bring a business’s marginal cost down to near-zero and in turn, enable subscription-based models. Here are a few thoughts on why we are bullish about a new wave of AI startups and the lens through which we are analyzing these opportunities.

The current AI Summer, the third since 1950s, has manifested itself in three waves. The first is that of pure research-driven companies, set up anywhere from 3 to 6 years ago out of top academic institutions. While these had minimal commercialization efforts, they did succeed in capturing the best grey matter and made significant progress in a nascent field, usually resulting in an acquihire. Examples include DeepMind, which was acquired by Google in 2014, or more recently, Jürgen Schmidhuber’s Nnaissense. Then came startups that took core machine learning technology, usually in natural language or machine vision, and packaged them into APIs. Startups like Wit.AI, acquired by Facebook, and Alchemy API, acquired by IBM, have since been added to the API arsenals of large corporates.

This brings us to the third wave, that of applied ML companies going after a specific vertical. A number of factors explain this shift. First of all, players like IBM, Microsoft and Facebook have managed to commoditize some of the basic building blocks through acquisitions, hiring rockstars and leveraging their massive technology infrastructure. We have also seen AI enablers being more readily available, including training data sets available on the web, the adaptation of GPUs for deep learning, the drop in cost for cloud compute and storage, etc. There’s also a record number of ML PhDs coming out of top universities and joining or starting technology startups instead of professional services companies in order to commercialize this nascent technology. Given these volumes, differentiation is a must and verticalisation is the solution to this end. There’s data to support this: consider that in SaaS we see 38% of vertical companies with ACVs in excess of $100k, versus only 4% for horizontals. This has given us a range of companies going after very specific verticals ranging from credit scoring to optimizing CNC milling machines.

Third wave AI, with its focus on deep specialization, represents a dramatic shift from generalist approaches in both strategy and talent needs. Here are three factors critical to vertical ML companies:

Industry expertise: Unlike with horizontal software that requires more functional knowledge in sales, marketing or HR, vertical startups pushing a new service will require deep industry knowledge in order to build the right product and walk the C-suite through longer sell cycles.

Data: One of the advantages of going vertical is the prospect of capturing unique data sets that corporates or incumbents would struggle to get their hands on. Doing this early on in the right field can kick off a flywheel that reinforces your model by offering a better service that guarantees further data acquisitions.

Models: While a whole range of useful ML APIs are out there, building your core product around publicly available algorithms doesn’t do much for defensibility. While leveraging existing services might make sense for some, the focus for a third waive company should be on having top talent building a differentiated core product.

Paired with the right market, having deep industry knowledge, capturing unique data sets and forging state of the art models are great ways of building a competitive advantage. While being strong in all three isn’t necessarily a must, a combination of these and an understanding of where weaknesses lie and how to address them is critical.

Internet era technologies like the cloud, mobile, and optimization techniques have allowed companies to scale much more efficiently. By bringing marginal costs down, especially for early stage companies that have yet to scale, these technologies have enabled new industries to switch to SaaS models (e.g. Uber testing a subscription model), providing a much better and deeper experience for both users and companies. We are convinced that ML has the potential to continue this trend and make services an order of magnitude better, faster, and cheaper. Europe is punching well above its weight in this field, and we can’t wait to see what industries get challenged by third wave AI.

Thanks to Chrys Chrysanthou, Nathan Benaich, Ernest Oppetit, Nick le Fevre for the suggestions, edits and banter.

Third Wave.AI was originally published in Notion Capital on Medium, where people are continuing the conversation by highlighting and responding to this story.

Use case = How a user interacts with a product to achieve a specific goal

Being a tech investor I think about emerging technology all the time. I also get asked about the ‘next big thing’ on a regular basis. We all want to have a point of view about what happens next and the extraordinary potential of new technologies.

But I believe the more relevant question focuses on the use case. For me, as an investor focused on the enterprise market, I believe the strategic cornerstones of any business are as follows:

Customers — reach, retain & grow

Productivity — do more for less

People — the right people in the right places

Security — managing & minimizing risk

If you’re managing and improving each of these areas then your business will be successful. To what degree comes down to things like business model and market size but they will dictate the scale of the success.

So it stands to reason that the most important thing about an enterprise IT product is that it has a material and measurable impact on at least one of these areas. Ideally I want to hear from the person within the business who is responsible for one of these areas that the product is driving material improvement and that they have the data to back it up.

When I get asked about emerging technologies like artificial intelligence, virtual reality or the Internet of Things, I try to relate it back to use case. However exciting a new technology might be, it has to have a clear and present use case. In the enterprise world it needs to be an enabler of making a business more successful.

It sounds simple but in all the excitement it’s amazing how often this gets overlooked.

The humble use case was originally published in Notion Capital on Medium, where people are continuing the conversation by highlighting and responding to this story.

Product market fit (or ‘PMF’) is one of those terms in the tech world that is constantly talked about and written about. Sometimes when a term becomes so universal it loses its impact. We stop thinking about exactly what it means and what we should do about it. I think there is a need for more definition and action around the subject.

I was talking on a panel at Saastock recently about why more European SaaS companies don’t reach scale and global success. I thought a good deal about this question and the single most important reason I could come up with was start-ups spend too much pre PMF product and then don’t spend enough post PMF.

This starts with a definition. Most definitions (including some of my own) are too vague and leave too much wiggle room. For PMF to be useful it needs to be defined in black and white terms. It also needs to be a high bar.

PMF starts with having clarity in your value proposition, your pricing and your target customer segment. Once these are in place you need to demonstrate clear evidence that they are aligned. You do this by having customers responding to you value proposition, agreeing to pay as per the pricing schedule and signing a contract to set it in stone. Once you have a representative group of customers doing this within your defined segment then the general thinking is that you have reached product market fit.

But the more I think about it, I don’t think this is enough. It is too open to interpretation. The bar is not set high enough. I want to imagine a really really high bar covered in sharp barbed wire and wired up to 500 volts of electricity. What does that look like?

A crowd

I want to see a representative group of customers. One or two customers may just be lucky or the result of pre-existing relationships. You want to see enough customers to demonstrate a trend and also so you can cross-reference to identify recurring themes. As a general rule of thumb I believe the number has to be in double figures.

Can’t live without you

I want to see a level of dependency being established with the customers. The product should address a real need not a curiosity. You want to hear customers already finding it hard to imagine a world without your product and wondering how they survived before you came along. There should be a clear and direct link between the product and the success of their business.

I don’t want to hear the customer talking in a more promiscuous way. That they are taking a look across the market, looking at the product in a more experimental way and are not yet ready to settle down to a long term commitment.

Time tells you allot

Given that this will be an early stage business, customers are unlikely to have been using the product for any length of time. But even a few months is so valuable as compared to customers who are running some kind of pilot or in their first couple of months. Time gives you dots to connect.

What you want to see is customer satisfaction and dependence trending up month on month. In the early stages, the best proxy for this is user adoption and user engagement. If more people are using the product and, of those people, using it increasingly over time then that is a pretty clear sign of growing dependence. To identify these kinds of usage trends I think the customer needs to have been using the product for a least three months and ideally six.

Commercials

The commercial commitment also tells you allot. Has the customer signed a contract and for how long? Have they signed up to a pricing schedule that is at or close to the standard pricing being proposed by the company? An annual contract reflecting standard list pricing is what you want to see. Anything else compromises the quality of these customer relationships.

Know your market

The customer segment needs to be clearly defined. But at the same time it needs to be big enough to be able to generate significant value in itself and also act as a beachhead into adjacent markets over time. You could argue that a barber shop may have product market fit in many respects. But its market of say men within a ten minutes walk is simply not big enough to generate meaningful value and it does not provide access into adjacent markets.

So here’s my real PMF checklist:

[x] At least 10 customers paying at or close to standard pricing and on annual contracts

[x] Majority of customers telling you there is a level of dependency on the product; that there is a direct link between the product and the success of their business

[x] Majority of customers have been using the product for at least 6 months and there is clear evidence of growing user adoption and engagement

[x] A market segment that is clearly defined and big enough to build meaningful value

Having all this evidence in place is a high bar that I think represents ‘gold plated’ product market fit; anything less could be considered an imitation.

I see start-ups time and time again raise money and then spend it far too freely before PMF is in place. This generally results in inefficient spend and performance that falls below plan. This in turn leads to major challenges in raising the next round of funding and the company going into a kind of holding pattern that it might never recover from.

The pre PMF stage requires allot of discipline. You should be running a lean organization and the founding team should be doing most of the work themselves. The longer you make the cash last the longer you have to get to PMF. You need to be very clear about the market you are going after, be listening to those customers and constantly iterating the product based on their feedback. You need to be flexible and prepared to pivot. It all takes time and discipline.

But once real PMF has been reached that’s the time to double down and be aggressive. PMF means you should be able to scale efficiently. The limiting factors become how much money you have raised and how large your market is. You are dealing with a different set of ‘scale up’ challenges like how to hire the right people and bring them up to speed, how to enter new markets and how to support an ever growing number of customers. This is the time to raise more money than you think you need and be bold in the knowledge that you have solid foundations in place.

Are you sure you have Product Market Fit? was originally published in Notion Capital on Medium, where people are continuing the conversation by highlighting and responding to this story.

At Notion we have been keeping an eye on the automotive industry for a while now as we believe it’s a market that, as it transitions from selling cars to offering mobility as a service, will cause broader tectonic shifts into our overall way of life, and not just in terms of transportation. We started our venture in this massive transition by joining FIVE.AI’s journey earlier this year.

So when the rumors started a couple of weeks ago that Apple might have taken an interest in McLaren we decided to take a look at why McLaren might make an interesting acquisition target, and what this means for tech companies trying to partner with more traditional auto manufacturers. And to be clear, we’re still not sure if we are looking at an acquisition, strategic investment, a partnership, or just unfounded gossip.

The British automotive company was founded in 1965 by Bruce McLaren as a motor racing team. It has since then grown into a conglomerate with $475M in revenues and has diversified into other precision manufacturing areas. The group is now broken down into:

McLaren Applied Technologies

McLaren Automotive

McLaren Racing

Absolute Taste (A thriving luxury catering business…)

McLaren Technology Centre

While a few of these might represent an interesting acquisition target, the one to focus on is McLaren Applied Technologies (previous McLaren Electronic Systems); in my view the crown jewel. The Applied Technology team which creates control and data systems has applied its research across healthcare and energy amongst other sectors but their true expertise lies of course in automotive and especially in the highest levels of motorsport. Their high-rate real time telemetry systems were first used for sports cars at Le Mans in 1991 and in F1 in 1993. They now provide the powertrain control and principal data logging tools for all competitors in those championships which are capable of dealing with large volumes of time-series data generated from various sensors in real time (including F1, Nascar, IndyCar etc). Some of their flagship products, like vTAG and ATLAS, focus on the viewing, analysis, recording and management of data primarily in the field of body dynamics (body roll, damper height, body flex etc). The team also developed an extensive and impressive collection of sensors. All of this allows teams to analyze data in real time, run simulations, and optimize their vehicles accordingly.

It is easy to understand why Apple might be interested in a team that has developed such high quality hardware and software technologies that have successfully been embedded into racing cars, performing while being subjected to the most extreme conditions. Especially the modeling and simulation expertise on the team would be a highly sought after talent pool. This would allow anyone managing an autonomous vehicle fleet, the ability to track, analyze, and control virtually all aspects of vehicles in motion, in real time. This is a set of capabilities which Apple (rumoured to be building its own vehicle), and most pure tech players have yet to develop, and acquiring/partnering with someone like McLaren could help them significantly speed up their product roadmap. The traditional vehicle OEMs would have developed in-house expertise for some of these capabilities over the years, but for someone building a vehicle themselves, Apple will need to have this in house as well to complete another piece of its full-stack puzzle.

While working at Living PlanIT I had the pleasure of engaging extensively with the Mclaren Applied Technologies team as one of our partners and seen their impressive operations and quality of their work. (I also had the pleasure to also visit their amazing technology centre in Woking on several occasions — a must visit for F1 fans). So no wonder they’re a target for any serious player in the autonomous vehicle space.

As this point, it’s hard to tell if these rumors are substantiated, nothing but gossips, or serious acquisition talks that fall through without ever going public. What is clear is that pure tech players will struggle to build a car on their own. We’ve seen Google and Apple push back their release dates and cycle through a few project leaders. Up until now, analysts have focused on traditional OEM’s lead in terms of manufacturing and infrastructure as opposed to tech player’s owning the state of the art in terms of machine learning. Players have tried to compensate with investments (Tesla Giga-Factory) and acquisitions (GM-Cruise), but one area that has been ignored is high performance data capture. Regular winners at LeMan and F1 have used, and enriched, systems like those produced by McLaren Applied Technology. Just a GM’s acquisition of Cruise triggered an applied ML for automotive acquisition race, an Apple-McLaren announcement could do the same for high performance data capture systems.

It’s about the high performance data capture (and a little about the auto manufacturing… was originally published in Notion Capital on Medium, where people are continuing the conversation by highlighting and responding to this story.