// Delivering on the Promise of Artificial Intelligence · Learning With Data

Delivering on the Promise of Artificial Intelligence

Oct 5, 2019 16:27 · 1172 words · 6 minute read AI startup

Photo by [Alex Knight](https://unsplash.com/@agkdesign?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) on [Unsplash](https://unsplash.com/s/photos/robots?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) Photo by Alex Knight on Unsplash

You keep hearing about Artificial Intelligence (AI). Nvidia’s CEO says, “AI is going to eat software.” Mark Cuban exclaims, “Artificial Intelligence, deep learning, machine learning — whatever you’re doing if you don’t understand it — learn it. Because otherwise, you’re going to be a dinosaur within 3 years.”

And yet, at the same time, you read articles about companies failing to execute with AI. Some even claim that 85% of efforts fail.

Being part of a startup, you don’t have the capital nor the time to have a potentially expensive failed experiment with AI, but you also can’t afford to be left behind. So where do you start?

I talk to startup founders often and am asked this question all of the time. I would like to propose 3 steps that will help you maximize your chances of successfully executing on AI. You might be surprised that none of them require state-of-the-art AI.

Always Start with the Data

Data is the lifeblood of AI. Think about all the companies you hear about when it comes to AI: Google, Facebook, Amazon, Netflix. They all have a lot of data and it is growing every day. Not only do they have a lot of data, but they also have a lot of proprietary data. No one else has all the sales data of Amazon, no other company can access the consumer data behind all of the Netflix shows.

Large amounts of proprietary data is an enormous benefit.

Fortunately, almost every company generates some type of proprietary data. If you are in e-commerce, you have data on which consumers are buying what products. If you are a SaaS company, you have data on how your clients are interacting with your product.

Your first step on your AI journey should be sitting down and answering the following questions:

  • What data assets does my company generate?
  • How proprietary are they?
  • How much data do I have now?
  • How fast are my data growing?
  • Where are the data stored? And how can they be accessed?

If after this analysis you find you have a significant amount of proprietary data that are growing fast and can be reasonably accessed, you are ready for the next step.

If you are having trouble identifying proprietary data, then try to think about how you can change that. Are you not leveraging Google Analytics for your website? Are you using a Facebook Pixel for your ads? Find ways to make your company generate data for you.

If no one at your company has any idea where the data end up or how to extract it, you need to first answer those questions. Maybe do some Google searching to find out how to extract Google Analytics data. Or hire a freelance engineer to help you.

Photo by [NeONBRAND](https://unsplash.com/@neonbrand?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) on [Unsplash](https://unsplash.com/s/photos/business-value?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) Photo by NeONBRAND on Unsplash

Identify Impact

Now that you feel comfortable you have some data assets you can leverage, you need to identify your first problem to solve with AI. My number one recommendation during this step is to

Ignore the hype of AI and focus on creating value from your data

For this step, I would identify AI as any machine aided process that uses data to deliver value in a scalable way. This is perhaps an overly broad definition of AI, but it should allow you to not care whether what you are doing is “state-of-the-art”, but instead focus on creating value from your data. Wikipedia defines AI as “intelligence demonstrated by machines.” I believe leveraging technology (machines) to use data to create impact (intelligence) definitely falls into that definition. Scalability is critical. It doesn’t count if you can’t scale your solution with technology.

Here are some items to consider when selecting a problem to solve:

  • How impactful would it be if I could solve this problem?
  • How hard does this problem seem?
  • Do we currently have anyone who could implement a solution using data and technology?
  • Does it seem reasonable that given some data a human could solve this problem?

I would answer those questions for 3–5 of the most impactful projects you can think of. You are looking for an impactful problem, that already has data, and doesn’t seem too challenging. Prioritize easy problems to try and secure a quick, first win. It’s a huge bonus if you think you already have an employee who could try and solve it. If not, try to find an employee who is passionate about solving it and not afraid to learn new skills. There is a tremendous amount of resources on the internet for learning about data analytics and artificial intelligence. Otherwise, look to hiring a freelancer on a platform such as Upwork. Just be sure to provide him or her with significant clarity and easy access to well-described data. This is usually a preferable first step, as opposed to a new hire, in case you find out you were not as ready as you thought.

Have a Scientific Mindset

Solving problems with data is really a scientific process. It tends to consist of developing a hypothesis, testing that hypothesis, and then learning something. And repeat.

Don’t expect your first hypothesis to be successful

From the step above, you might have developed the hypothesis that you can use sales data to identify customers who would be good targes to upsell. You’ve looked at some of the data and are pretty confident you see correlations between product purchases. So you repurposed and employee to solve the problem, implement your solution, and your upsell rate doesn’t change. Well, I guess you have just failed at AI — time to move on.

No. You tested a hypothesis and that hypothesis failed. Now you need to dig in, learn why, develop a new hypothesis, and test it. Perhaps the correlation you saw wasn’t causal; maybe there was a bug in your code; perhaps your test just needed to run longer. You need to develop a hypothesis as to why it failed. That will then guide your next move.

This process of testing hypotheses, learning, and repeating takes time. It’s a large reason why it is preferable to pick a problem that appears to have a quick win. That will hopefully allow you to not have to fail too many times before your first taste of success.

Photo by [Clark Tibbs](https://unsplash.com/@clarktibbs?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) on [Unsplash](https://unsplash.com/s/photos/success?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText) Photo by Clark Tibbs on Unsplash

Now — Go Do It

There you have it. Take the time to think about what data assets you have, analyze potential areas of impact, and think like a scientist. If you start with these three steps you will be in a solid place to start delivering value to your startup with data and AI. These steps also don’t require you to have a strong background or understanding of AI — you just need to think strategically about your problems and your data. Once you have that understanding, you then bring someone on (or learn yourself) to execute on the AI. Just make sure you keep thinking like a scientist while you do!

LinkedIn Share tweet Share