Industry News Details

Jumping Into Artificial Intelligence: Five Considerations For Creating An AI Company Posted on : Jun 10 - 2019

The advice from Andrew Ng, former founder of Google Brain, former chief scientist at Baidu and founder and chief executive of Landing AI, on artificial intelligence (AI) is to just jump in. “It is a big journey, but jumping in is not hard,” says Ng. And indeed, in the past few years we've seen many folks jumping headfirst into AI.

My journey in AI and machine learning has spanned over a decade, in which time I’ve built 100-plus machine learning models and 50-plus models in production that have been deployed in actual products. On my journey, I’ve learned firsthand what it takes to build an AI-driven company. Below is a follow-on discussion based on Ng’s AI advice that explores what I believe are the key requirements of an AI company.

1. Make sure decision makers, engineers and product managers are all data-minded.

Data is king. Data has become a key business asset, and organizations should invest in, curate and harvest it like any other asset. Executives, decision makers, engineers and product managers at an AI-based company should all center their work around data. They don’t necessarily need to be data scientists, but at the very least they will need to consult with data architects and scientists.

More specifically, the full engineering team should understand data science and data engineering, and most will need to be able to build simple models. It’s not usually enough to sprinkle a few data scientists and magical models throughout the organization.

As for product managers in an AI organization, their roles should move from a front-end design model to a data-first model. This means that they will need to learn to query data and look at spreadsheets to recommend features or measure the accuracy of their models in conjunction with data engineers and data scientists. Typically, most product managers don’t have these kinds of data skills, but the availability of products like Data Studio, Presto and Big Query have made it easier for product managers to query and analyze their domain-specific data.

2. Compile and centralize your data. 

AI-driven companies should centralize data from disparate databases in a data warehouse (or data lake). Not all data needs to be there, but companies should centralize as much as possible with appropriate pseudonymization and controls. Teams need to ensure, however, that they avoid “data lake hype.” I don't believe that building massive data lakes is the proper approach for optimally utilizing AI. Instead, have a human look at small data to find initial answers, then apply AI to the larger dataset later.

Additionally, companies should augment their “inside” data with “outside” data. You can garner much more insight by adding outside data to what's happening inside the organization. For example, if you’re researching a person’s background, a Google search can return 100 useful insights about a person beyond just their resume. The same is true for your customer relationship management software (CRM) and supply chain data. Explore outside data to expand the limited perspective you can gain from your organization’s data. View More