For each piece of the platform you can build it (write your own code), assemble it from open-source components you run yourself, or buy it as a product or managed service. Engineers tend to underrate buying, because a subscription is visible on an invoice while their own time is not. The questions below help.
| Question | Points toward building | Points toward buying |
|---|---|---|
| Is it your competitive edge? | Yes: recommendations, pricing logic | No: copying Salesforce to a warehouse |
| Who will run it at night? | A team with on-call capacity | Nobody: keep the team small |
| How will usage grow? | Fast: per-row pricing would explode | Slowly or predictably |
| How easy is leaving? | Lock-in is unacceptable | Open formats keep an exit open |
Ingestion is the classic case. Copying data from a common SaaS application or database into a warehouse is the same problem for every company, so managed connectors (Fivetran 16,237 ) or open-source ones (Airbyte 128,715 , dlt 876,044 ) usually beat custom code (Ingestion Tools). A custom connector looks like a two-day job; keeping it working through API changes, rate limits and schema drift never ends. By contrast, the logic that turns BookNest's orders into its margin figures is unique to BookNest and belongs in its own dbt 37,942 models, whatever tool runs them.
Two rules of thumb help: buy commodity, build differentiators, and keep an exit by holding data in open formats (Parquet 129 , Iceberg 129 ), so that a price rise or an acquisition (Fivetran and dbt Labs Merger) is a negotiation, not a crisis.