A team needs a new ML capability. Should it develop the capability internally, buy a service, or combine the two? Start with a clearly defined problem, then investigate what the market already offers. A familiar problem may have an established vendor solution, but availability alone does not make that solution a good fit. section 3.0 section 3.2.1
Work through the decision at the level of system components:
- Is this capability central to competitive advantage? Third-party infrastructure can be sensible even when it is critical to operations; businesses whose infrastructure distinguishes them may invest in building it. section 3.2.1
- Does a vendor meet the actual need at an acceptable price? Check the required capability and metrics, not just the vendor’s broad product category. A price that works for one company may not work for another. section 3.2.1 section 3.4.2
- What would building or adopting open source require? Compare vendor costs with internal hiring, infrastructure, maintenance, problem-solving, available time, and team capacity. Open source is not automatically free to operate. section 3.2.1 section 3.5
- Could an external product satisfy only part of the need? Sketch the system in a few clear blocks before deciding which components to build and which to obtain. section 3.5
Search for a photo marketplace
A marketplace with millions of images needs to answer customers’ text queries with relevant photos. It would be easy to assume a company of that scale must build its own search engine. First, survey search-as-a-service providers. Many may be unsuitable because text-to-image search is not the usual offering, but some may warrant a closer look. Also inspect how other search systems approach related problems. The goal is not to reproduce a much larger search engine; it is to find an approach that meets this marketplace’s needs. section 3.4.2
The decision is not necessarily all-or-nothing. Look for products that satisfy the whole requirement or a useful portion of it, then decide what remains to be built. Revisit the choice when the proposed level of innovation demands investment that may not pay off. section 3.5
