Ask how much bandwidth your organisation needs and someone will produce the formula: count the users, multiply by some Mbps per head, add 20% for luck. It fits on a whiteboard, it feels rigorous, and across the enterprises we audit it is reliably wrong in one of two expensive directions. Either the site is undersized and everything crawls at peak, or it’s oversized and finance has been paying for headroom nobody will ever touch.

At enterprise scale, both errors multiply. Ten oversized sites is a six-figure annual habit. Ten undersized ones is a productivity problem that gets misdiagnosed as an IT problem for years.

Users don’t consume bandwidth. Workloads do.

A 20-person architectural practice and a 20-person call centre have almost nothing in common on the wire. The architects push 10GB model files to the cloud and sit in video reviews half the day. The call centre needs a modest but immaculate stream of low-latency voice where a single congested moment is an abandoned customer. Head-count formulas treat these as identical businesses.

Sizing that holds up starts from the application mix, not the org chart. What actually rides this connection: cloud backups, video, voice, file sync, remote desktop, security cameras? When does each run, how much does it burst, and how badly does it tolerate delay? Do that mapping properly and interesting things fall out. That backup job demanding 200 Mbps runs at 2am and shouldn’t be sized into the daytime peak at all. The daytime number that matters might be barely half of what the formula said.

The three questions that actually size a connection

How much contention can you absorb? If speeds sagged 30% at peak, would your teams notice and cope, or would you lose revenue? The honest answer decides between shared infrastructure and uncontended fibre, and it’s different for a head office than for a distribution centre. Paying for uncontended fibre at a site that would happily cope on less is one of the most common oversights we unwind.

What does an hour of outage cost? Uptime percentages sound interchangeable and are anything but. 99% allows three and a half days of outage a year. 99.9% allows nearly nine hours. 99.95% allows about four. Somewhere in your organisation is a site where four hours offline is a shrug, and another where it’s a board paper. They should not be on the same class of service, and under most enterprise agreements we review, they are.

Where will you be in two years? Standing still, enterprise bandwidth demand grows 20-30% a year. Add a cloud migration or a serious video culture and it’s 50% or more. Connections should be sized for 18 to 24 months out, on contracts with an upgrade path that doesn’t require re-engineering the site. Otherwise today’s correctly sized link becomes next year’s bottleneck, locked in for another two years.

The part the carriers won’t do for you

None of this analysis comes from your provider, for the obvious reason: the carrier’s incentive is to sell you the bigger number. The formula suits them fine. What an enterprise actually needs is the workload map, site by site, matched against what each site is currently paying, and the mismatches priced. In our experience the same review that finds the undersized head office also finds three branch sites on enterprise-grade services running email and a point-of-sale terminal.

That review is squarely what a Telco Health Check covers. We map what your sites actually use, benchmark it against what you’re contracted for in both directions, and hand you the corrections. Sometimes that’s an upgrade. More often than the carriers would like, it’s not.