Some things I learnt from a16z
With the holidays approaching it’s a good time to relax and look back at what you’ve achieved throughout the year. There may have been some bittersweet moments but the key point to remember is that you’ve been able to move forward, even just a bit. It’s times like this that I get to take a step back and look at WORKING ON rather than WORKING IN my business.
With that in mind I’ve been lucky to stumble upon some great market insight videos from Andreessen Horowitz (or a16z as their also known - for reasons why check out this link: https://www.techopedia.com/definition/28701/a16z). For those of you who aren’t aware, they’re a large venture capital fund in the USA with some well known companies they’ve supported. This portfolio includes Facebook, Uber, AirBnB, Asana, Box, Buzzfeed, Okta and Slack (just to name a few). So it’s safe to say that they know a thing or two about industries (because wouldn’t you if you were going to throw a lot of money at an unknown investment).
Anyway, the videos they have on their website (www.a16z.com) and YouTube (https://www.youtube.com/channel/UC9cn0TuPq4dnbTY-CBsm8XA) have been quite interesting for me and I hope they might be for you too. They cover topics like what’s working and not in the Artificial Intelligence space, a primer on Quantum Computing and the various types of marketplaces and what works to be successful there.
Below are my summaries from watching a few of these (a good trick is to play them at 1.5x speed to get more info faster).
GOING TO MARKETS WHEN NO MARKET EXISTS
This video hits close to home for me because it was the eternal struggle I felt at QMG where we had to educate the market about the services and products we offered. The education about how our economic model gave unparalleled market insights and was in turn valuable for the investment process was something that each client needed to learn and the pace at which this understanding was transferred differed from client to client.
This video goes through the different types of marketplaces for new versus mature products and the types of staff you need, sales processes you can expect and is a good way to look at how to go to market when you have a new idea.
Some key points from this video include:
There is an assumption that sorting out distribution is the key but there is no greater driver of impact on future valuation than pricing.
When we think that customers are not buying because the price is too high we just haven’t sold them on the value of the product.
Normally, prices are worked out based on how you want to deliver the product and what type of services are offered with it. The higher the price the more direct the sales approach.
Also, when talking about pricing it is important to
keep things simple
not bundle with more mature products (as peoples buying decisions will be based on that product not yours)
start at the higher end when pricing so that when the market matures you can offer more discounts
Mature vs Early Market Sales
Early market sales vs mature market sales require different types of thinking of how to approach customers. It’s more important to focus on relationships when the technology is being created.
The type of people you want to work in this early market space are known by different names but typically they are hunters who are resourceful in creating their own sales tools and have a passion for the technology. It’s not surprising at all that very early on there is hardly a break even point, that comes as the market matures.
Typical Early Sales Engagement Model
Customers well trained to talk about sales early but its so important to educate
I’ve experienced what we thought was a sure sale of a product only to learn that the potential customer was asked by their buying manager whether our product represented to them a “nice to have” or a “must-have”. Because we just breezed through the sales process as they seemed to sign off on everything fairly quickly, we failed to ensure that we educated the customer well enough and we ended up not being seen as completely integral to the process. Learn from this too.
To PS or not to PS
This is about professional services and the choice of whether to go down this path or not.
Might not be great from an investment point of view (as products are higher margin) but it’s absolutely necessary from an early market perspective so need to beware of this.
Building out a partner network in an early market area is tough to do. It only really works well when there’s a “pull” market. This is because in early markets there’s too much evangelism required.
However, it’s worth investing here for when market does mature but it won’t drive your numbers.
Common mistakes include hiring the wrong type of sales leaders (i.e. someone not used to being in that hunter mindset and willing to own the product sales lifecycle). Additionally, the assumption that the creation of early feature commitments to certain customers will be right for other customers later as you may end up cannibalising future business but also you cannot focus too broadly when well qualified opportunities might be staring right at you.
AI: What’s Working, What’s Not?
From the outset, Frank Chen of Andreessen Horowitz gets us straight to the punchline and that is that companies all around the world (especially large ones) are doing a lot with AI so it’s time to think about what we (as smaller companies) can do with AI.
Some lessons from the talk
AI will likely get into all types of software coming out in future just like databases did.
Databases very good at storing information; AI is very good at understanding things like seeing pictures, understanding language and forecasting.
Registrations for AI conferences like (NIPS) have increased faster in more recent years.
CEO’s and world leaders are talking more about AI as well whether in earnings calls or political interviews.
The next part of the video shows what’s been working in terms of AI in products. The examples shown were Pinterest (which recognises items in an image and finds similar ones in other images), Airware (gathers data from drones to conduct safety checks at mining sites). Language understanding is also another facet of AI that has been shown to work with services like Siri, Alexa, Cortana. There’s also Everlaw which translates and transcribes (as well as clusters) legal documents and has been working well within the legal industry. In terms of predictions, there are apps like Cardiogram which can use non-healthcare devices like the Apple Watch to predict whether you’re at risk of heart attack by using deep neural networks and reinforcement learning.
Furthermore, there’s also a bunch of companies providing infrastructure to the market which is akin to those who provided the picks and shovels in the early gold rush days. For example, there’s databricks which is helping to provide capacity to do machine learning to those building AI models. In that space there is a lot of data and processing that’s needed. Helping data scientists just focus on their modelling as opposed to having to focus on developer operations. Taking that further there are companies like SigOpt which help with the paramaterisation of AI models and can help companies optimally tune these models.
What’s not working
Next, the video focuses on what’s not working in the AI space. There are 3 stages of AI and what works in one area may not work in another. The example here was that the algorithms that work for Pinterest compare images won’t help Everlaw sort documents and those are also not helpful for self-driving cars. The next stage from this narrow focus is general AI where the AI is smart enough to solve generic problems. All of this comes well before having to worry about super AI, a type which one day can think like humans.
Right now, there is a lot of Narrow AI work being done but there isn’t a consistent research agenda to get us to General AI.
AI Frontier has build an AI report card to help grade the various types of narrow AI products out there and not all do well. There is still a gap even in narrow AI between human and computer performance when doing image recognition or answering what we see as simple questions. Getting to learning things on the fly will take time but there are a number of tests including the Turing test and the others shown below.
Whilst some people are afraid the future of AI leads to Skynet (from the Terminator movies), it’s more likely we’ll get to Jiminy Cricket. The latter was the children’s book character that would give Pinocchio advice. Just like apps like Waze will give you advice on what side street to turn down, the future of AI will be about Jiminy Cricket like moments as well. The key here is that innovations like this make humans smarter and more effective.
AI eating all the jobs
This is something that gets brought up as each new technology comes out. The examples given were that the looms were supposed to take all the textile jobs, the tractors all the farming jobs, the semi-conductor taking all the jobs etc. But it turns out the economy created new jobs for people and that’s because we have an infinite appetite for new types of services etc. All of the jobs below didn’t exist before recently.
It’s counter-intuitive because we think that with automation come job losses but if we look at the rise of ATM’s we actually see bank teller jobs go up. This is because ATM’s gave banks an ability to reach out to rural communities and hire staff there whereas without automation and technology they couldn’t afford to put all the manpower required into opening branches in those remote areas.
So what does this mean for us?
AI can help with existing products and services but we have to pick where we want to deploy it. Is it in new or existing products or is it to assist with corporate objectives. Not all AI research has been applied to software and products so there’s still a long pipeline of products and services that can be optimised with AI. People need to be trained as well but they don’t need to go back to university. It’s now quite easy to study online through sites like Udacity.
The key for us is to find where AI can help in our organisations and it potentially starts with simple questions like what can and should be automated.