Decision guide
Build or buy a routing engine
Six questions that decide it, the three cases where building your own is the right answer, and the parts of a build that nobody costs up front.
Six questions
- 1Would a customer pay more because your assignment decision is different from your competitors’?Yes → build
- 2Do any rules have to be un-overridable — a licence, a certification, a jurisdiction — rather than a warning someone can click past?Yes → buy
- 3Will anyone ask, months later, why one specific job went to one specific person?Yes → buy
- 4Does the work have to be spread fairly across people, not only matched to the best one?Yes → buy
- 5Is the rule set small and stable — three or four tags and a round-robin, unchanged for a year?Yes → build
- 6Who owns this in year two, after the person who wrote it has moved teams?No answer → buy
Two or more “build” answers and a stable rule set: build it. Any single “buy” answer in rows 2 to 4 outranks the rest — those three are the ones teams rewrite from scratch the second time.
Build it yourself when
- Routing is the product.A marketplace’s matching is its moat. Do not rent your moat.
- The rules genuinely are simple.Round-robin over one skill tag is a week of work, and it stays a week.
- You cannot take a dependency.Air-gapped or sovereign deployments with no external service at all.
What a build actually costs
- The decision trail.Not a log line — a record of who was considered and who was blocked, and why, that answers a question asked six months later.
- Fairness.The part that gets rewritten three times, because “best fit” quietly means “one person does everything”.
- Nobody answered.Deadlines, escalation ladders, re-offers, and the backlog that builds while you decide what to do about it.
- A console.Someone in operations needs to see the queue and override it, at 07:00, without you.
Why we can afford to write the left-hand column honestly
The engine underneath Fivexer is MIT-licensed and public, so “buy” still ends with code your own team can run. That is what makes the case for building a real one rather than a rhetorical move — if our price stops working for you, the exit is a git clone.