Machine learning · Marketing

Are you seriously considering incorporating neural networks into your SaaS startup?

Alexander Doak Tech lover, visionary, father of 8 and counting

April 28th, 2017

Even if the service offered isn't directly related to machine learning, it seems many are using it in their systems internally. Is this just a fad or a marketing gimmick? Or are these networks really worth deploying?

Mike Robinson

April 30th, 2017

Hi Alexander,


I notice your tag line says "father of 8". I have 7. Not often I meet a guy with more kids than me! ;-)


Anyway, to your question: IMHO, Neural Networks and AI are the future, and applications are endless. Obviously they play a role in self-driving cars, drones, robots, chatbots, and Natural Language Processing (NLP) like Alexa, Cortana, OK Google. etc. Does every startup need AI? No. There is a lot of hype around AI (and for good reason!), so investors are more eager to look at a company that uses AI. But it only makes sense if your startup is really doing something that NEEDS AI. If you are just trying to bolt on some kind of AI story in order to include that hot buzzword, then don't bother! Most investors aren't stupid. they can tell if your business really leverages AI or not. And not every business needs AI today.


So maybe ask yourself a different question: will AI really help my business and give me a significant competitive advantage somehow? If so, how?

James Lauzun Pharma Insider Creating AR Solutions for Big Pharm

April 28th, 2017

Very interesting question....


It's not a fad, and it's a marketing gimmick... Okay, let me give a bit for context.


Understand my background is in science- specifically chemistry. This makes me view machine learning a bit differently. When you really go and ask experts, they tell you that modern machine learning is about math; the best are essentially mathematicians. The big challenge with machine learning is how do you relate all these massive inputs/data together to create a clear picture- it's through algorithms, and algorithms are solved by mathematicians.


So to your question! Neural nets are fantastic when you have lots of complex data points about a high value data set. They'll tell you anything you want to know.


But if your solution doesn't create massive amounts of complex data, a neural net is probably overkill. So for example: if your company creates content of nearly any type, you probably don't need to invest into your own dedicated AI. There's products already existing on the market that would input all user's content consumption and would output individual recommendations. You're the lucky ones. Incorporate AI into your valuable value props as proven solutions emerge.


But if your value prop is entirely created by understanding massive data sets of movements, locations, ID tags, images, prices, etc and in relations that are predictable but nearly impossible to solve, then potentially your own neural net is worth developing.


You'll notice the end recommendation for both ends up having AI. It might be the future, but it isn't a must have for every company and technology- you don't necessary need to be an early adopter.

Sean Lorenz

April 28th, 2017

Having been part of and started several startups that tout neural networks (machine learning/deep learning/AI, etc.) as a competitive differentiator, I can tell you that it is both a marketing gimmick AND the best thing since sliced bread!


I'm halfway joking. The reality is that neural network algorithms have honestly not changed much since the 1980s. The difference is that we finally have the compute power and horde of data to solve supervised and unsupervised learning problems with brute force.


The question is: what is that problem you're trying to solve and does it fit the profile of what neural networks do best? Many startups say they're using deep learning but this almost always aspirational and not current implementation. How can I be so sure? Data. Most startups are trying to get customers, which means they don't have much data yet. Deep learning is only as good as the amount and quality of data you have. There's a reason Google freely opens its AI platform (TensorFlow) without worrying about competitors -- they have the data.


I think machine learning can be applied to just about any dataset, BUT you have to know which tool is best for the job and be confident that your business model shows a quick path to data access. If you're shouting "we have AI!" in the market, but don't have the data, people will jump ship quickly.