A B2B commerce platform must decide whether to build its own logistics network or partner. What should drive the decision beyond cost?

The factors that actually decide it

• Is this differentiating? If reliable delivery to small-town retailers is the reason customers choose you and no partner can provide it, the capability is the product. If partners can, then building it is duplicating a commodity at enormous capital cost.
• Does the partner's economics work at your scale and mix? Third-party logistics is priced for average parcels on dense routes. A business whose parcels are heavy, whose routes are sparse, or whose delivery windows are unusual may find no partner willing to serve it at a viable price — which forces the build regardless of preference.
• What does control buy? Data, service-level guarantees, the ability to change the process, and the ability to promise something competitors cannot. Those are real and they are what justifies the capital.
• What does it cost when it goes wrong? An owned network's fixed costs persist through a downturn; a partner's variable costs do not. This is the single largest risk of building and it is usually under-weighted during growth.

The hybrid that is usually correct

Own the dense, differentiating routes and partner for the sparse tail. Dense routes have the volume to justify fixed cost and are where service quality is a competitive advantage; the tail has neither.

That also gives a negotiating position and a fallback, which a pure-partner model lacks and a pure-owned model does not need.

The architectural consequence

Either way, the system must be able to route across multiple providers including your own, with a common abstraction, per-provider capability declarations, and per-provider reliability tracking. Building that abstraction first preserves the option; hard-coding a single provider — including your own network — forecloses it.

The most expensive version of this decision is making it irreversibly, either by building a network that cannot be wound down or by embedding one partner's model so deeply that switching is a rewrite.