The following is adapted from my new book, Real World AI.

You’ve found a way that AI can help your business. Maybe you want to automate the way help tickets are categorized, or improve search results, or increase efficiency in restocking.

You Want to Use AI: Do You Build, or Do You Buy?

Whatever you want to do, you face one big decision: Do you build the AI model yourself in-house, or do you purchase components from a third-party vendor and integrate them into your business?

It’s a critical choice to make, because it will have big impacts on the quality of the model, the cost, and the length of time to implementation.

It’s also a more complicated decision than you might think. Many people automatically assume that buying makes the most sense, but you can’t just buy a complete AI solution off the shelf. Many components go into a successful AI model, and there’s no one-stop-shop that can provide you with the entire system you need.

In order to implement AI successfully in your company, you must carefully strategize when (and what) to buy, and when to build yourself.

Is the AI Solution Related to Your Company’s Core Business Functions?

The first step in deciding whether to build or buy is to consider whether the AI solution is related to your company’s core business functions. If it is, then it often makes sense to build it in-house, as it would be a key competitive advantage. If not, it might make more sense to buy it, so that you’re not diverting valuable resources away from your core business functions.

For example, consider a fashion retailer that wants to use AI to create a better shopping experience. Specifically, they want to provide products relevant to a selected category—for example, when a consumer goes to the website and selects “blazer,” they want to show that consumer a bunch of blazers to choose from.

How do you determine the function?

The company’s core business function is fashion—that’s how they provide value to their customers. Based on that, it makes sense for them to create the training data for the AI model in-house, by taking lots of pictures of their inventory and annotating them by hand, deciding whether or not each picture was of something that could be considered a blazer.

A waste of time and money to always create technical infrastructure from scratch.

On the other hand, it would be a bad use of their time and money to create technical infrastructure from scratch. It has nothing to do with their core business functions, and it won’t provide them with an advantage over their competitors, nor enhance their brand. So they’re better off buying off-the-shelf solutions for most or all of the infrastructure.

How Quickly Do You Need a Solution?

Next, consider the role of time and urgency in your decision. There are many reasons that could drive a company to buy a component off the shelf, even if it does affect their core business, and time is one of the biggest.

The urgency options — building and buying, have timelines associated with them.

In general, it’s typically faster to buy and implement components than it is to build them from scratch. You might be able to buy and implement a component in half the time it would take you to build it yourself.

Shortening the time to market may be a priority. Often, companies will be tempted to take a shortcut to reduce their time to market if they see or anticipate competitors about to do the same. The opportunity costs of not solving the problem may be significant.

What Level of Quality Do You Need?

You’ll also have to examine the quality of buying versus building in-house.

Sometimes building in-house can result in a higher-quality model, because you can design everything for your very specific situation. However, this depends greatly on your company’s technical sophistication, resources, and expertise. Even if other considerations have you leaning toward building a component in-house, if you don’t have the capability to build it with sufficient quality, that option might be off the table.

Security is part of your company’s quality.

You should also consider security as you think about quality. You might think that buying a third-party product and integrating it deeply into your business has the potential to introduce security risks. But unless you have significant security expertise internally, you could just as easily introduce those risks by building insecure functionality.

How Much Do You Want to Spend?

Both building and buying require money and investment, so you’ll need to understand the budget you have in the context of the value your solution will provide to the company.

In some other cases, it may be prohibitively expensive to build a team to create infrastructure from scratch. When Yahoo! was making a similar decision, they were concerned that they wouldn’t be able to hire enough talent for a machine learning team to work on their core search functionality.

Costs must always be top of mind. What is your return on investment?

Facing pressure to stay competitive in the short term, they chose to stop investing in search as a core business. Of course, history has shown that Yahoo! lost that one to Google.

So while it’s often more costly to build something in-house, you have to think about the return on investment. If building in-house will drive greater strategic value down the line, it might be smarter to spend more now.

What Third-Party Components Are Available?

You also need to consider what third-party components are available — you can’t buy something if it doesn’t exist. There are many pieces of major infrastructure you’ll need to set up to enable your eventual success with AI, and you might be surprised by how many third-party solutions are available to help.

The first piece of infrastructure is your data.

You’ll need to have a lot of data to feed your model, as well as a way to clean it, move it, organize it, and store it. Unless you have extremely specific needs, there are many open source and commercial products that can handle the mechanics of moving data from here to there.

You’ll also need infrastructure that enables you to annotate all your data.

In some cases, your annotations will be the key differentiator that allows your model to provide business value (as in the fashion retailer example), which might convince you to build this infrastructure yourself to protect your IP. But there are also many commercial companies, such as Appen, who have security solutions in place to protect your data, as well as the processes and knowledge to help you annotate your data most effectively.

Next, you’ll need a platform to orchestrate training, testing, and hosting your models.

All of the major cloud platforms—Amazon, Google, Microsoft—provide machine learning platforms that can automatically train, test, tune, and deploy models. There are also full life cycle open source solutions, like Kubeflow, as well as point solutions that can be integrated together or with components you build yourself. And commercial vendors like Databricks can build more sophisticated custom solutions.

Mix and Match for the Perfect Solution

All of these considerations will play into the ultimate decision of whether you build or buy. Most projects end up being a combination of both, mixing and matching different components.

Conclusion

By considering your core business functions; the desired time frame, quality, and cost; and the available third-party components, you’ll be able to find the right solution for your company. Other things being equal, you should try to build components that are key to your company’s core business and buy the rest.

For more advice on building or buying an AI solution, you can find Real World AI on Amazon.

Wilson Pang joined Appen in November 2018 as CTO and is responsible for the company’s products and technology. Wilson has over nineteen years’ experience in software engineering and data science. Prior to joining Appen, Wilson was chief data officer of Ctrip in China, the second-largest online travel agency company in the world, where he led data engineers, analysts, data product managers, and scientists to improve user experience and increase operational efficiency that grew the business. Before that, he was senior director of engineering at eBay in California and provided leadership in various domains, including data service and solutions, search science, marketing technology, and billing systems. He worked as an architect at IBM prior to eBay, building technology solutions for various clients. Wilson obtained his master’s and bachelor’s degrees in electrical engineering from Zhejiang University in China.

Alyssa Rochwerger is a customer-driven product leader dedicated to building products that solve hard problems for real people. She delights in bringing products to market that make a positive impact for customers. Her experience in scaling products from concept to large-scale ROI has been proven at both startups and large enterprises alike. She has held numerous product leadership roles for machine learning organizations. She served as VP of product for Figure Eight (acquired by Appen), VP of AI and data at Appen, and director of product at IBM Watson. She recently left the space to pursue her dream of using technology to improve healthcare. Currently, she serves as director of product at Blue Shield of California, where she is happily surrounded by lots of data, many hard problems, and nothing but opportunities to make a positive impact. She is thrilled to pursue the mission of providing access to high-quality, affordable healthcare that is worthy of our families and friends. Alyssa was born and raised in San Francisco, California, and holds a BA in American studies from Trinity College. When she is not geeking out on data and technology, she can be found hiking, cooking, and dining at “off the beaten path” restaurants with her family.

Image Credit: markus spiske; pexels

Wilson Pang

Wilson Pang

Wilson Pang joined Appen in November 2018 as CTO and is responsible for the company’s products and technology. Wilson has over nineteen years’ experience in software engineering and data science. Prior to joining Appen, Wilson was chief data officer of Ctrip in China, the second-largest online travel agency company in the world, where he led data engineers, analysts, data product managers, and scientists to improve user experience and increase operational efficiency that grew the business. Before that, he was senior director of engineering at eBay in California and provided leadership in various domains, including data service and solutions, search science, marketing technology, and billing systems. He worked as an architect at IBM prior to eBay, building technology solutions for various clients. Wilson obtained his master’s and bachelor’s degrees in electrical engineering from Zhejiang University in China.