▮ Build Or Buy?
When settings up an ML infrastructure, at one extreme, a company can outsource everything except data movement. At the other extreme, a company can build everything and maintain all the required infrastructure.
However, most companies are in neither of these extremes, which means they need to decide whether to buy or build ML infrastructure. There are several factors when making this decision, so I’d like to share the 3 most common.
Reference: Designing Machine Learning Systems
▮ The 3 Factors
1. Company Stage
If your company has just started out leveraging AI, using vendor solutions might be better because of the velocity of implementation and its relatively low costs. It will require large amounts of effort if starting from scratch such as hiring more engineers and building the infrastructure.
However, as the project scales, using vendors will become quickly costly. In this case, it might be better to invest in your solution.
2. Belief in Competitive Advantage
Using an outsourced ML model means that your competitors can easily replicate it as well.
If the company wants to be the BEST at a certain solution, it is better to manage it in-house. If not, outsourcing may be the optimal option.
The majority of non-tech companies, where ML infrastructure is not their primary focus, tend to bias toward buying and preferring managed services.
On the other hand, companies where technology is their primary competitive advantage, tend to bias toward not buying.
It is important to find the right solution while understanding what kind of bias the customer might have.
3. Maturity of the Available Tools
In some cases, there may be no tools available at the time which meets specific company requirement.
This frequently happened in the early days of ML adoption.
