Whether your business needs a backup internet connection is arithmetic — this article is the other half: making the backup actually work, automatically, without anyone crawling behind a rack mid-outage. Because the gap between “we have a second connection” and “outages don’t stop us anymore” is entirely in the engineering details, and they’re the details installs most often skip. Here’s failover done properly.
The pieces: two paths and a brain
The second path. The pragmatic default at small-business scale is cellular failover — a business LTE/5G modem or router with a data plan, giving a path that shares almost nothing with your wired line (different medium, different infrastructure — the shared-failure-mode check that matters). The step up is a second wired circuit from a different carrier, ideally entering the building differently — more capacity, more cost, and worth verifying it’s genuinely a different path, since two “competitors” reselling the same underlying fiber fail together. Size the backup honestly for essential traffic — cloud apps, email, phones, payments — not for a full-speed ordinary day; failover mode can be a slightly slower day, and that’s fine.
The brain: dual-WAN on the firewall. The switch has to be automatic, which is the firewall’s job — business-grade units support dual-WAN natively: monitoring the primary continuously and shifting traffic to the backup within seconds of failure, then back when the primary recovers. Configuration details that separate working setups from theoretical ones: health checks that test real reachability (pinging beyond the ISP’s own equipment — the classic misconfiguration detects only the dead modem, not the dead circuit behind a live modem) · priority rules for failover mode — phones and business-critical apps first-class, the update downloads and streaming politely throttled, so the smaller pipe carries what matters · and a notification when failover occurs — routed to a human, because running silently on backup for three days, data plan draining, primary unfixed, is the second-most-common failover story after “it didn’t switch.”
The dependencies that quietly break failover
Three details, each discovered the hard way somewhere every week: power — the failover gear, firewall, and switches belong on the UPS, because regional outages take power and internet together, and a backup connection that dies with the wall socket protected nothing · phones — VoIP should ride the failover with priority (that’s much of the point — clients calling a dead line draw conclusions), which means confirming the phone system’s traffic is in the priority rules, not just assuming · and inbound expectations — if anything reaches into your network (a hosted service, vendor connections), failover changes your public address; cloud-first businesses mostly don’t care, but the exceptions should be known before the day, not diagnosed during it.
The quarterly test: pull the plug
Failover has the same epistemology as backups: untested is a hope. The test is gloriously simple — on a scheduled quiet moment, unplug the primary WAN and watch: does traffic shift within seconds? Do the phones stay up? Does the notification fire? Does failback work when you reconnect? Ten minutes, quarterly, logged in one line — and the first test almost always finds something: the health check watching the wrong target, the phone traffic missing from priority, the cellular plan that lapsed. Finding it on a Tuesday you chose is the entire product.
The bottom line
Failover is a system, not a purchase: a genuinely independent second path, a firewall configured to switch automatically on real reachability, priority rules that fit essential traffic through the smaller pipe, everything on protected power, phones confirmed aboard, a notification when it engages — and a quarterly pull-the-plug test proving the whole chain. Built that way, the next outage in the neighborhood becomes a notification on someone’s phone and a slightly slower afternoon — which, measured against what the outage used to cost, is the best trade in infrastructure. For the local weather this system will actually face, the ice-storm preparation checklist shows exactly where failover earns its keep.