Planning for growth without over buying on day one

Growth planning is not a forecast of how many sites you will own next year. It is a sorting job: every decision in front of you either unwinds with a support ticket, or costs you a scheduled weekend and some real risk. Commit early only to the second kind. Everything in the first kind can wait until an actual site needs it.

Sort each decision by what reversing it costs

Run every choice through one question: if I get this wrong, what does the fix look like six months from now? The answers cluster cleanly.

Why relocating a live site dominates the expensive list

A move is never one action. It is a chain of six or seven, each with a distinct failure mode, and the chain repeats for every site. Two sites is an evening. Thirty sites is a project with a rollback plan and someone watching logs afterwards. That asymmetry is the whole argument for choosing placement deliberately at the start: pick the machine you can still live with when traffic is a few times higher, rather than one you already suspect is marginal. When a genuine ceiling does arrive, a step up to a VPS is a planned migration you schedule, not an emergency you absorb.

The naming scheme is the sleeper

Naming feels free while you have four sites, then becomes the thing you touch most often. A scheme survives growth when it encodes only facts that cannot change. It forces a painful rename when it encodes anything volatile.

Survives:  cli41-us-01   cli41-us-02   cli41-eu-01
           (stable client code + region + counter)

Breaks:    acmeshop-2026-tier2
           (brand can be rebranded, year ages, tier changes on upgrade)

Fix the convention while a full rename is still an afternoon. At fifty entries it is a week of careful work with a real chance of orphaning a monitoring check nobody looks at until it matters.

A review cadence short enough to actually happen

1. Monthly, ten minutes. Reconcile the account against reality. Which domains went live, which quietly died, which addresses currently resolve to nothing. Reclaim the idle ones before the next renewal rather than after it.
2. Quarterly, half an hour. Look at peak load, not the monthly average, and compare it to the tier ceiling. If peaks touch that ceiling in two consecutive quarters, start planning the step up while you still get to pick the date.
3. Annually, at renewal. Review structure. Do your isolation groupings still describe the estate you actually run? Renewal is the one predictable moment when relocating a site is a project instead of a scramble.

Events that override the calendar

• A signed contract with a launch date inside the next two months.
• Traffic on one property doubling within a quarter.
• A client relationship ending, which converts a shared machine into a separation job.
• A software change that alters the load profile, such as a brochure site becoming a store.
• An address developing a delivery or reputation problem, which is a defensive move rather than a ranking matter.

The rule holds in one line: buy the irreversible things once and carefully, and buy the reversible things late and small. Counting what you need for the very first order is handled separately under sizing and selection, and the mechanics of an eventual relocation live in moving in.