A contractor puts a trencher through a fibre duct three streets away. Nobody in your office knows that yet. What they know is that the ERP is spinning, the card machine says "no connection", the phones have gone quiet in a way that is not restful, and eleven people are now standing in the kitchen at half past nine on a Tuesday. The line came back at half past two. The invoice for that morning never gets written, because nobody invoices for a morning of standing around.
When every system you rely on moved into the cloud, the line into your building quietly became the most critical piece of equipment you own — and most businesses still buy exactly one of them.
What actually breaks
It is rarely the thing people picture. The five real causes, roughly in order of how often we see them: physical damage to the fibre — roadworks, cable theft, a trencher; an ISP-side fault that has nothing to do with your building; your own equipment, where the router or ONT has quietly died or wedged; power, because the network gear was never put on backup, which is the load-shedding conversation from the other direction; and congestion, which is not an outage at all but feels exactly like one at four in the afternoon.
The first thing to establish is who to phone. In South Africa the company that owns the fibre in the ground — the network operator — is usually not the company that bills you, and a fault in the street is theirs to fix on their timeline. Your ISP is your only route to them. Knowing that distinction before an outage saves an hour of phoning the wrong help desk while the office waits.
Two lines, or one line twice
The instinct is to buy a second connection. The trap is buying a second connection that fails at exactly the same moment as the first. Two ISPs reselling capacity on the same physical fibre into your building is not redundancy — it is two invoices for one point of failure. Real diversity means a different path: a second fibre from a different network operator if your area has one, or the far more common and practical answer for a small business, fibre plus a mobile connection. LTE or 5G comes into the building through the air rather than through the trench that just got cut, which is precisely the property you are paying for.
Redundancy is not about having two of something. It is about having two of something that cannot break for the same reason on the same day.
Failover that happens without a human
A backup line that somebody has to go and switch on is not failover, it is a spare. Somebody has to notice, find the router, know the password, and be in the building. By the time all four are true the morning is gone.
Automatic failover means a dual-WAN router that continuously tests whether the primary line is actually carrying traffic — not merely whether the cable is plugged in, since the most annoying faults are the ones where the link stays up and nothing passes — and moves traffic to the backup within seconds. Two things are worth knowing before you expect miracles. Sessions in flight generally do not survive the switch, so an active call drops and a large upload restarts. And unless you have arranged your own address space, your public IP changes on failover, which matters if anything out there is allowed in based on your IP: a VPN, a supplier portal, a bank. That list needs to include both addresses, and it is the item most often discovered mid-outage.
Not everything deserves the backup line
Mobile data is more expensive and slower than fibre, and the moment you fail over, eleven people's cloud backups, software updates and video calls all pile onto it at once. Without shaping, the backup line saturates in minutes and the business is no better off.
Decide the pecking order in advance and configure it once: the card machine and point-of-sale first, then voice, then the line-of-business system and email, then everything else. Cloud file sync and automatic updates should be paused entirely while you are on the backup path. This is a twenty-minute job on a decent router that turns a backup line from a gesture into a working plan.
What to ask an ISP before signing
Which network operator owns the fibre to this address? Is this a business or best-effort service, and what does the SLA actually commit to in hours? Is the connection uncontended or shared, and at what ratio? What is the mean time to repair for a street-level fault in this area? And who exactly do we phone at 07:00 on a Saturday — a number, not a portal.
What an SLA buys, and what it does not
A business-grade line costs more than the home-grade product with the same speed on the label, and the difference is not bandwidth. It is a repair-time commitment, a support queue that a business sits in rather than a consumer one, and usually a less contended service so that four in the afternoon feels the same as ten in the morning.
What an SLA does not buy is a promise that the line will not break. The compensation for missing it is almost always a credit against your monthly fee, which is trivial next to the cost of the outage it is compensating for. Buy the SLA for the response time it puts on paper, never as insurance against the loss.
Voice, and the calls that vanish
Cloud phones are the first thing anyone notices when a line wobbles, because a dropped call is more visible than a slow page. If the business runs on phones, put voice traffic at the top of the priority list, and check how quickly the handsets re-register after a failover — some systems take minutes to notice they have moved, and minutes of unreachable phones during an outage is exactly the impression you were trying to avoid.
The people who are not in the building
Half your team may work from home part of the week, which quietly moves the same single-point-of-failure problem into a dozen houses you do not control. You do not need to fund fibre for everyone. You do need a plan for the handful of people whose work cannot wait — usually a mobile router or a data allowance on a phone that can be tethered, agreed in advance rather than improvised on the day.
Test it, on purpose, monthly
Failover that has never been tested is a theory. Once a month, unplug the primary line and watch: does traffic move, how long does it take, what breaks, do the phones come back, can you still take a card payment? Ten minutes, on a quiet morning, with the team warned. Every business that has ever done this has found something — an address that was not whitelisted, a router that failed over and never failed back, a backup SIM that expired eight months ago. Better to find it on your terms than three streets from a trencher.
