Build Versus Buy: Should You Build or Buy a Predictive Analytics Solution?
By Melissa Todisco, October 11, 2019
I recently was at the Strata Data Conference where the conversation was focused on how to deliver on the promise of AI (or rather how to apply machine learning to deliver business value).
I walked around the Expo Hall where there were 100s of software and data management solutions, yet listening to the talks and speaking with attendees many were focused on building and deploying their own models.
Every company is faced with this choice:
Do you build or buy a predictive analytics solution to meet business needs?
So how do you go about evaluating the trade-offs about whether to build or to buy?
Understand the Problem.
Having a thorough understanding of the requirements for a predictive modeling solution will help you evaluate whether your in-house talent has the skills or capacity to deliver and deploy a solution. If solving the problem in-house has the potential to leverage or create proprietary IP, it might make sense to build it in-house. For example, Netflix’s recommendation engine is a core component to its product. It wouldn’t make sense to outsource that type of predictive solution. On the other hand, if you are looking to create an automated customer service chatbot or a predictive planning and budgeting solution there are many software companies who are experts at addressing this need.
Evaluate Your Culture.
Building a solution in-house will take time and will likely include a lot of learning and failures. Depending on the business need, it is important to evaluate whether your culture will support this process of learning and the amount of time it will take to deploy a working solution.
Evaluate The Costs.
For software, the costs are more obvious. In addition to licensing and implementation fees, it is also important to consider whether additional capacity may be needed from existing vendor contracts. For example, will you need more computing resources or will you need to expand your database?
While building it yourself doesn’t have a fixed budget associated with it, here are some indirect costs to be evaluated:
Employee Time: Depending on the complexity of the project, you may need multiple employees to allocate their time to building and deploying the solution.
Ongoing Maintenance: After the model is deployed, your team will need to continue to maintain it. While the maintenance may not be as intense as development, it is possible that needs may change creating the need for additional updates.
Opportunity cost: While your team is working on this initiative, it means they can’t be working on other initiatives.
Time to utility: Building custom software is a time consuming process (we know!). The more complex the solution, the longer it will take to build and deploy it.
Weigh Scale vs. Specificity
For projects with specific requirements, it may be easier for an in-house team to build it. For example, if you have a specific way in which you categorize and track data about employee performance, it may be easier to build a retention model working closely with the team that is the closest to that data (if they have the predictive skills, of course). Whereas if you need a highly flexible solution designed to meet the needs of various stakeholders, you may consider looking for a software that has developed a process for scaling across an organization and a variety of use cases.
There are a variety of factors to weigh when building or buying a predictive analytics solution and there is no right answer. It depends on the specific problem you are looking to solve and the unique dynamics of your organization.
Here is a brief summary of the pros and cons we see for each:
We also created this decision tree to help you decide whether to build or buy a predictive analytics solution: