A platform must decide whether to build its own routing and dispatch optimisation or buy a commercial service. What decides it?

The primary test

Would you accept your closest competitor using the identical implementation?

If yes, the capability is not differentiating and buying is usually right — identity, payments, notifications, email delivery, observability. Building these means spending engineering on a problem others have solved and that no customer chooses you for.

If no, building may be justified — but only if you can actually build it better, which is a separate and harder claim.

Why routing and dispatch is the hard case

For a delivery marketplace it is arguably the core capability: matching orders to shoppers under changing supply and demand, with quality directly affecting delivery times, cost per order and both sides' experience. A generic routing service optimises distance; the business needs to optimise for a blend of time, cost, shopper earnings, order batching and customer promise — objectives no vendor knows.

That is the real signal: when the objective function is specific to your business model, generic solutions optimise the wrong thing.

The secondary considerations

• Speed to first value. Buying gets something working now, which is worth a great deal early. A common path is buy first, build later once the requirements are understood — and understanding them is itself the main output of the buying period.
• Ceiling. A bought solution has limits. Reaching them means migration, so the question is whether you will reach them and when.
• Data. If the capability improves with your data, buying may mean the vendor's model improves from your data while yours does not. For an optimisation capability this can be decisive.
• Total cost including operations. Building means operating, maintaining and staffing it permanently, not just writing it.
• Dependency risk, which is real and usually overstated relative to the cost of building.

The pattern that works

Buy the surrounding infrastructure; build the objective function. Use commercial mapping, traffic and geocoding data; build the optimisation and dispatch logic on top. This concentrates engineering on the part that is differentiating and buys the parts that are not.

The failure mode in each direction: building everything spreads a limited team across problems others have solved, while buying everything means the differentiating capability is identical to competitors' and is improved on a vendor's roadmap rather than yours.