A Model That Fits: Making Sure Your Company is Ready for Machine Learning

Machine LearningRecent buzz-worthy articles on machine learning and artificial intelligence have piqued a mainstream interest in machine learning. NASA is finding new planets using Google’s AI, Apple (and competitors) are locked in a rat race to bring neural networks to LiDAR in self driving cars, and even insurance companies are using ML to fuel chatbot conversations about auto claims. Although your company’s objectives might be smaller scale or less revolutionary, adopting machine learning into your business process can drive results even for niche industries.

While services like Google Cloud Platform’s Machine Learning Engine and their open source Tensorflow library have helped to make machine learning more accessible for everyone (heck, even an elementary school student can understand their teachable example), developing ML software is still an undertaking. Before you jump to hire a machine learning engineer and think of names for your robots, it’s important to know which conditions need to be ripe for a useful machine learning model to grow.


Articulate a Clearly-Defined Business Objective for Machine Learning

Like most of Google Cloud Platform’s services, Google ML allows you to start small and scale later, but there is a time and monetary cost to investing in machine learning. In order to foster support within your organization, you must be able to articulate a valuable business problem that machine learning could help solve. If you’re unsure where to start, you can explore 10 common use cases to generate ideas for your company.

And don’t worry, you won’t need to start from scratch. Google has been developing AI algorithms for a decade for their consumer services, and now you can leverage these image, video, speech, text and translation APIs to incorporate with your business process.

Just because machine learning is a hot topic doesn’t mean that it’s a right fit for every business process. For processes that deal with clear, predefined set of rules, where there is no ambiguity in results, a simple automation task might suffice. Ultimately, machine learning is most applicable to prediction scenarios with a complex array of inputs.

Get All of Your Data into One Place

A machine learning model can only work with the data it’s exposed to. If your data is in silos across spreadsheets, offline databases, or somewhere in your email inbox, it’s a higher priority to focus on data consolidation first. And this is no small decision – Google BigTable, for instance, is perfect for data-intensive training sets typical of TensorFlow models (Google documentation recommends BigTable for sets >300GB). If you don’t have this much data or it’s already stored in Datastore or BigQuery, these services are also compatible with Google ML and its APIs. Once you get data relevant to your business problem into a storage service, then you can shift focus to machine learning models.

In order to make sure your data is ready for a machine learning model, ask yourself: Can I query this data to answer relevant business questions? Is there any crucial ingredient missing? For instance, we cannot create a useful model to describe the price of a used car without knowing its mileage, even if we have other important data like year, make and accident history.

Once you’ve wrangled together your data, only then can you start to understand which type of machine learning model you will be working with. Does your business problem require supervised learning (good for classification and regression models), unsupervised learning (good for pattern recognition from an array of input variables), or reinforcement learning (good for a models with a desired positive outcome and known inputs)? Often, this distinction will arise naturally from understanding your input data and a clearly defined business problem.

Do You Have Enough Data to Train the Model?

It’s important to start simple. Although several companies attempt to bring machine learning into their analytics, only one in 20 has done so extensively. Common pitfalls include starting with too complex of a problem or insufficiently training a model to provide useful output.

For instance, we cannot create a model to predict whether our site visitors will buy novelty, holiday-themed T-shirts if we don’t have training data from visitor behavior last holiday season. While we can attempt to use a pre-trained model, or visitor behavior for a similar clothing website, we’ll have a less error-prone model if we have authentic training data. For supervised learning models, it’s critical to have real results that help guide and iterate the accuracy of the model.

Getting Started in Google Cloud Platform

Now more than ever before, ML models are attainable to small and medium-sized businesses courtesy of services like Google ML and open source libraries such as Tensorflow. Google advocates that it’s not necessary to have a PhD anymore to understand deep learning, largely due to these platforms that abstract machine learning concepts from the complex mathematical operations that drive them. Not only are there a multitude of quickstart tutorials in the Google documentation, there are also several stellar resources for programmers looking to become versed in machine learning.

Google has made it clear that they are an AI-forward company, and many others are following suit. While the water is warm, and barriers to entry are lower than ever before, be sure to come up with a deliberate, thoughtful strategy to incorporate machine learning into your business process.


For more information on Google Cloud Platform, click here for a free consultation today!


[Request a Google Cloud Platform Consultation]


Simon Margolis
Director | Cloud Platform

Simon Margolis

Leave a comment

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>