Machine learning solutions are here, but it comes down to building versus buying.
Machine learning is changing the landscape of business. It’s allowing us to make better decisions, understand our target market, and offer comprehensively better experiences for our prospects, customers, and employees. The problem is that it’s a mystery to many of our shareholders and corporate managers. Directives come down from meetings to incorporate machine learning, but there are a lot of choices and a lot of directions. The biggest question of them all? To build your solution or to buy it.
There are lots of good reasons to choose either strategy, so let’s take a look at the pros and cons to help you decide based on your organization’s unique situation. Here’s what you need to know to get started on the path to effective machine learning.
Building A Machine Learning Solution
Your first option is building something from the ground up. A unique, in-house machine learning solution does provide a level of customization you won’t get anywhere else. But before you build, you’ll need to know a few things.
Building an internal solution gives you complete control over your machine learning path. If your company’s mission is unique, and machine learning capability will be an integral part of your pipeline from here on out, a customized solution could be the answer.
Organizations with the budget also use an in-house solution to provide the necessary security for sensitive data. This sometimes includes organizations in fields where burdensome compliance regulations can hamper solutions that aren’t built with the specific organization’s mission in mind. The resulting system is heavily secure and considers exclusive access to sensitive or proprietary data.
– Resource and personnel heavy
– Training intensive
If you aren’t an AI company, building and training an AI solution – even with open-source materials – can cost you big time. Building an AI solution can run in the millions, from the software to the team required to develop and maintain it.
And it’s not just money. It can cost a significant amount in resources and personnel. Many experts cite the need for at least three to five data engineers per data scientist needed to build the algorithms plus other vital team members from data analysts to software engineers. These experts aren’t cheap, not to mention the extra staff needed to support a larger organization.
Finally, it’s training intensive. You’ll need massive amounts of data with a streamlined pipeline before you can even get started, much less begin delivery of insights.
Buying a Machine Learning Solution
If you aren’t sure you have (or want to find) the resources to build your machine learning platform, buying an out of the box solution could help you realize your goals. All the questions you may have about how to build, train, and deploy a machine learning solution have probably been answered already by a company with a product that fits just right.
– Quick deployment
– Monitored performance
– Low startup cost
If your machine learning initiative isn’t unique to your particular business or the central tenet of your business mission, using an out of the box solution could be just the thing to get your machine learning initiative off the ground. You can take advantage of a company’s solution that’s already built and trained without the massive cost of building in-house. Plus, their expertise ensures your models keep running.
It’s tempting to consider building because that seems more respected, but in reality, few companies have the time or resources to build – and more importantly, maintain – a solution. When you think about long term deployment, it makes sense for most companies to use an out of the box solution.
– Fewer customization options
– Potential security issues
An out of the box solution can be challenging to deploy if your machine learning initiative is unique to your field (or at all). Buying a solution works best when machine learning informs a common initiative like marketing. If you’re the only company currently researching an obscure protein’s role in tissue regeneration, that may prove more difficult for out of the box solutions.
In some cases, security can be an issue as well. For truly sensitive data, having a third party with access to data, even encrypted data, can provide too many security and compliance concerns to be worth it. For most companies, data isn’t quite that sensitive. However, for those with the most complex data concerns, building a solution may be the only way to help alleviate those concerns.
Building or Buying – How to Choose
And now for the practical part – how do you choose which path to take? Building a solution commands a great deal of respect, and those bragging rights can sometimes obscure what the right answer is for your unique situation.
To get past the glory of jumping in with both feet to building – and soon discovering you’ve bitten off more than you can chew – here’s your path to get there.
Choice 1: Long Term or Short Term Project?
For short term projects, such as using machine learning to make marketing decisions, the length of each is an essential factor. If you’re building for a short term solution, you may be able to put together a solution using open-source models.
For long term projects, you may need the support and continual updates of a dedicated, out of the box solution. It requires plenty of resources to build the infrastructure for long term solutions because technology moves quickly, and security needs to evolve. Using an out of the box solution puts continual updates and dedicated security evolution at your disposal without removing your team members from their primary tasks.
But what if it takes three times as long to build a solution for my short term project? Is it worth it to commit resources, even open-source ones, to the solution for a project that takes a few weeks? What if there are more iterations – does that mean it’s long term or short term?
And what if my long term project doesn’t fit any solution out there? We have proprietary data or exclusive access to data for our long term, ongoing project, and there’s no solution out there for us? What then?
Glad you asked. This gray area leads us to our next question.
Choice 2: Large Scale or Small Scale Project?
A small scale project could allow you the cohesion of building a solution in house regardless of the actual time period the project will happen. For smaller-scale solutions, it may still be worth it to put the time in at the beginning to get a cohesive model that works for your specific needs. Small scale projects may not be quite as time-intensive as
For larger-scale projects, on a long or short term scale, the cost of building involves a great deal more resources for cohesion. The moving parts require so much skill and expertise to maintain over the broader picture that unless you’re already a company working in AI solutions (think FAANG or AI-specific startups), you may not have the team you need and the resources required to handle it.
Okay, okay, but there’s still a gray area. Here’s a third, decisive question.
Choice 3: Common Solution or a Solution Only You Need?
Now we get to the bones of it. Your machine learning initiative is most likely just like quite a few others, and that’s great news. If it’s solving a problem that other companies also have, there’s a solution already out there for you.
Most companies aren’t dealing with a scope or field exclusive only to them. Netflix, for example, needs a recommendation engine on a scale unlike any other. They have to build in-house. However, if Netflix needs to build an algorithm that provides a basic insight into the popularity of specific offerings based on geolocations, even Netflix might buy their solution.
Building Versus Buying – The Real Deal
For the vast majority of businesses, it’s going to make sense to buy a solution because of the sheer resource load needed to build and maintain something in-house. The technology changes quickly, with security needs moving right along with it.
Buying an out-of-the-box solution makes sense in all but the most narrow cases. If you have a completely unique use case, scale and scope like no other organization, or access to extremely privileged data – plus the financial resources to hire your team and dedicate time to build and train – then an in-house solution could work.
But there’s a good chance you don’t fit any of those cases, so the clearest and safest choice will be to buy. If nothing else, trying an out-of-the-box solution first allows you to pivot in-house later while abandoning the more expensive in-house solution for a cheaper out-of-the-box solution really hurts. When you buy, you’ll deploy cheaper and faster, get your people back on their primary missions, and have security support from day one.
By: Elizabeth Wallace
You might also be interested in Build or Buy Data Science Solutions.