Blog: Actionable advice and learnings: Profitable AI — build or buy it?
What do Sundar Pichai, CEO of Google, Mark Zuckerberg, CEO of Facebook, and Jeff Bezos, CEO of Amazon, all have in common? Besides the obvious ones like rolling in money and being in charge some of the world’s biggest companies, they’ve all joined in on the choir praising AI. And when they preach, the world listens and jumps on board. And the hype is not for nothing: How we like to think about it, without appearing like crazy fortune-tellers, is to compare AI with the rise of e-commerce. When e-commerce grew in popularity, it spread like wildfire and many jumped on the train by pouring resources into building their own web shop from the ground up. Some naturally did so successfully, but for many it was too expensive and difficult to build and maintain the shop with no support from experts and third parties. Not very long after simple frameworks appeared that could be bought cheaply and implemented easily, enabling new e-commerce businesses to focus on their core competencies without also having to write a lot of code. In other words, web shop frameworks now function as an underlying layer of e-commerce businesses and most probably don’t really think about how it exactly works, but it’s a must-have for selling products in today’s digital world.
We believe this will be the case for AI as well: In a near future it’s going to be implemented as a layer underlying almost every business in one way or another.
Now back to the bald founder of Amazon, who luckily agrees with us (or maybe it’s the other way around). His words were: “Machine learning and AI is a horizontal enabling layer. It will empower and improve every business, every government organization, every philanthropy — basically there’s no institution in the world that cannot be improved with machine learning.”
We made the global hype local by addressing AI in our latest StartupTalk at PreSeed Academy. And the choice of topic wasn’t incidental as some of the startups and founders in the PreSeed Ventures portfolio are advancing in the AI space. The guys throwing out AI puns and learnings were seasoned entrepreneurs whose babies are not just their startups, but also the AI technology that is at the core of their businesses and careers. Michael Green, co-founder and CEO of Desupervised, showed up alongside the AI-nerdy duo from Forecast, Dennis Kayser and Jacob de Lichtenberg. Oh, and our very own tech-guru, Investment Manager Anders Bach Waagstein, appeared for an investor perspective on the subject. We’ve compromised the wise words from the AI wizards into the following learnings and considerations.
Avoid the hype trap
With the big choir preaching and the media being all over it, it’s hard not to think that your startup should integrate AI to the business in one way or another. And that’s a healthy thought. With the expected development mentioned above it’s reasonable to have an idea of how to potentially implement it. But be realistic about it: Don’t think that you can grab an ounce of AI, throw it at a problem and now your product is 10x better and all your challenges have vanished. Or that just because you’ve included two capital letters in your pitch deck, every investor will hunt you down begging you to take their money. In fact, many founders are trying to build startups revolving around AI without really knowing nor understanding the complexity of it. Moreover, most AI developed (around 66%) never sees the light of day and never gets deployed. Even if it does, it doesn’t necessarily work the way you’d hope.
Despite the hype, rather than being a promise of success, integrating AI in your business is linked to high risk, which is why you should understand and respect the technological complexity before going full Terminator.
What do you need it for?
Before directing your entire focus on how you should implement AI in your business, it’s vital to understand exactly why and what you need it for. Only when you’ve figured that out, you can move on to the next step in deciding whether you need to build an in-house AI team or if you’re better off leveraging existing tools on the market. It basically boils down to either building or buying your AI — or combining the two.
Moneymaking AI automates personalization
So, you’ve avoided the hype trap and you’ve started at the end, asking what the pain you cure is, but there is a few more things you need to have in place to turn AI into your own moneymaking machine. It kind of boils down to user experience and engagement. There’s especially to boxes you need to check; 1. Make sure your product becomes better over time and 2. Personalization. At the very beginning you ‘put a lot of data’ into your AI and start training it, so it’s going to work for the customers. You could almost say that you need to keep going back to the beginning. Every single time a customer uses the product, it becomes better. The use of the product ‘automatically’ adds data, that trains the AI and ultimately it should improve the customer experience. A big part of a better customer experience is personalization, the product shouldn’t only become better in general, it should become better because it customizes the product to help the specific customers personal needs. The great thing about a product that gives the customer a better and more personalized experience the more the customer uses it, is that it becomes sticky and profitable.
Strategic approaches to AI
There’s not only path to applying your AI, so let’s have a look at different strategic approaches. We’ll stay basic by highlighting three fundamental approaches to the build or buy question.
First, there’s the IT centric approach. In Machine Learning as a Service, an AI provider is responsible for using and deploying your machine learning algorithms as well as the data used. There’s plenty of providers to choose, including the companies from the abovementioned choir.
This approach is fast to get up and running and it’s relatively easy to implement. You have certain inputs and a desire to get certain outputs. Most likely, you don’t know what’s going on, but maybe you don’t really care. The downside is that you have to pay for everything. Training of the model, data storage, output, etc. It can quickly become expensive, if you don’t control the models that you run and don’t keep tab on the value it brings. It require you to think closely about how you really use it.
Second, there’s the integrated structure. It provides you with a bigger ownership of your AI and combines MLaaS with customized machine learning platforms. With bigger ownership comes more responsibility and this approach requires a lot more from you as you’re an active part of the development and technology. Also, you’ll be needing data scientists who are hard to attract. However, this approach allows you high flexibility and you can tweak the technology, make it more bespoke and continuously improve it for your specific business. Therefore, you need to find people that understand the algorithms, but you don’t need them to reinvent the wheel. You can use what’s already in the market and capitalize on that. As mentioned, attracting and retaining these talents is extremely hard as they’re in great demand and can require a high salary — even higher outside of the small Danish monarchy.
Lastly, there’s the specialized approach. This is all about building. And not just building anything, but something groundbreaking that is better than what’s already out there. This approach is only relevant if AI is core to your product and the betterment of it. It requires a highly specialized team, with roles including data architects, data engineers, data analysts, business analysts, data scientists and data journalists. The team required for this approach is extremely hard to attract and retain and the approach is only possible if you have that AI dream team. Even if you manage to get the best line-up and create an AI solution that effectively solves a real problem, it’s not necessarily a winner. You’ll still have to have people use and buy your product and in a rather new industry, which AI is, it often comes down to adaptation. The market and potential customers may not be ready or interested in an AI solution yet.
Your business model decides your approach
What it comes down to is basically what you need AI for in tandem with your business model. Is it a ‘need-to-have’ or a ‘nice-to-have’? If it’s core to your business and your competitive edge, then building AI in-house seems like the appropriate choice. But if that’s the case, you also need to ask yourself what talents and competencies you’re able to attract. If you’re not able to build an AI dream team anyway, then destiny might have directed you back to the buy option.
Talking of building a dream team, our next StartupTalk will concern how to continuously attract and retain talent in a startup while keeping the team motivated. Be there or be square.