Connecting Starlink kits at scale: what actually breaks
Patterns that show up when fleets grow from dozens to hundreds of kits—discovery lag, billing reconciliation, and ops processes that worked at 20 sites.
What worked at small scale (and then didn’t)
| Practice at ~20 kits | What happens at ~200+ |
|---|---|
| Manual onboarding checklist | Drift; skipped steps; ghost inventory |
| “Check the portal when someone complains” | You learn about issues from customers |
| Spreadsheet for usage | Reconciliation fights; version chaos |
| One engineer who “knows all the sites” | Bus factor becomes a business risk |
Breakage pattern 1: inventory truth
Kits move. Companies rename sites. Installers swap hardware. Your system of record disagrees with the vendor’s.
Symptom: tickets against the wrong site; billing against the wrong customer.
Mitigation: single ownership for identity (kit ↔ customer ↔ site), change process for moves, periodic reconcile jobs—not hope.
Breakage pattern 2: billing and usage windows
Metered satellite services rarely align with “the calendar month your ERP likes.” When cycles, prorations, and top-ups enter the chat, naive monthly rollups lie.
Symptom: finance and ops argue using different totals that are both “correct” in their own frame.
Mitigation: agree on the window definition before you automate; store bounds with the data; never silently re-bucket history.
Breakage pattern 3: silent degradation
A path can be “up” while throughput collapses or latency spikes. Small fleets notice because someone is watching. Large fleets need systematic detection.
Symptom: “It felt slow for two weeks” with green lights everywhere.
Mitigation: track user-facing symptoms and capacity headroom, not only binary reachability.
Breakage pattern 4: process latency
Discovery and ticket routing that took an hour at small scale become multi-day backlogs. The network is fine; the operating system of the business is saturated.
Mitigation: automate the boring handoffs; reserve humans for exceptions.
A scale readiness checklist
- Can you list every kit → customer → site in one query?
- Can two people take vacation without freezing changes?
- Do you test failover and onboarding monthly?
- Do finance and ops share one definition of a usage period?
- Do you know your top ten noisiest alerts—and have you deleted any recently?