Our $450,000 WAN Bill: What I Should Have Checked Three Years Ago
The Invoice That Made Me Rethink Everything
It was a Tuesday morning, Q3 2024. I'd been a procurement manager at a 600-person tech company for about six years, managing a telecom and network infrastructure budget of roughly $180,000 annually. I thought I had a handle on things—until I ran a routine variance report and found an $8,200 discrepancy in our MPLS billing from the previous quarter.
At first, I assumed it was a data entry error (we'd had some turnover in the finance team). But when I traced the charge back, I found a pattern: three separate line items for 'regional transport surcharges' that had been added in Q1 2023 without a corresponding change in our contract.
That discovery sent me down a rabbit hole that led to us switching carriers, renegotiating our SD-WAN deployment, and eventually implementing Juniper's Mist WAN Assurance platform. Along the way, I learned that the conventional wisdom about carrier pricing—'all carriers are essentially the same, just pick the lowest rate'—wasn't just incomplete. It was costing us money.
What I Assumed vs. What I Found
Everything I'd read about WAN procurement said the key was volume discounts and long-term commitments. In practice, I found the real cost drivers were far more mundane: bill of lading errors, unmonitored overage triggers, and the silent creep of 'transport zone' reclassifications.
From the outside, carrier pricing looks straightforward: you pay for bandwidth, a port fee, and maybe a transport charge per circuit. The reality is that for a multi-site operation like ours (we had 43 sites across the continental US), each circuit was priced differently based on:
- Distance from the carrier's nearest Point of Presence
- Last-mile provider surcharges (which can vary by 300% by ZIP code)
- Burden of local loop maintenance
- QoS-level pricing tiers that were almost never explained
I want to say the actual variance between sites was 15-20% for identical bandwidth profiles, but I might be misremembering the exact figure—that's how much I was relying on memory instead of data. When I finally ran a proper analysis across all 43 circuits, the spread was 47%. The most expensive site (a small office in rural Pennsylvania) was paying 1.8x what our main data center paid for half the bandwidth (ugh).
The Real Cost of 'Set and Forget'
Our procurement policy at the time was simple: pick a primary carrier (Crown Castle in our case, for their fiber footprint), negotiate a multi-year contract, and renew every three years. We'd been with Crown Castle for seven years at that point, and I honestly believed that loyalty earned us competitive pricing.
People assume the longest relationship equals the best deal. What they didn't see—what I didn't see—was that our contract had auto-escalation clauses tied to a proprietary index that increased 4-6% annually. Over six years, that compounded to roughly 31% inflation in our baseline transport costs.
When I compared quotes for a $4,200 annual contract renewal from 2023, Crown Castle quoted $4,840. A competitor quoted $3,650 for identical specs. That's a 25% difference hidden in fine print.
The best part of digging into this: when I presented the data to our CFO, she didn't just approve a carrier switch. She greenlit funding for a proper network observability platform (Juniper Mist WAN Assurance, which I'd been reading about in industry briefs).
The Transition: What Changed (and What Hurt)
We transitioned to a hybrid WAN approach over 60 days in Q4 2024. The carrier switch was messy—there's no way around it. We had two sites offline for 24 hours due to a port misconfiguration on the new MPLS connection. A small office in Grand Rapids had issues with USB power delivery on the new CPE (while recording a list of devices, no less—the tech needed to power our VoIP phones via PoE and the new router didn't support the standard properly).
(Not that the old carrier cared: their support channel was a shared inbox with an average response time of 18 hours. The transition taught me more about our internal network topology in two weeks than I'd learned in five years of annual reviews.)
But here's the part that surprised me: the payback period was six months. We cut our WAN transport costs by 19% (annualized, that's $34,200 saved), and the Mist platform gave us something our old SASE solution never had: proactive alerts before circuits went down. Before, I only knew something was broken when a site called complaining.
Actually, that's not entirely fair. Our old carrier did offer portal-based monitoring. But nobody in our IT team ever used it because it required a separate login per circuit, and the data was 48 hours stale. Mist's dashboard—single pane of glass, real-time—changed that completely.
The Mist Effect: More Than Just Visibility
I have mixed feelings about SD-WAN overlays. On one hand, they promise dramatic cost savings by enabling hybrid connectivity over broadband. On the other hand, I've seen three companies migrate to SD-WAN and then hit SLA nightmares when their 'transport-agnostic' architecture hit a peering problem that took weeks to diagnose.
What sold me on Juniper's Mist solution was the AI-driven assurance engine. Not because it's flashy (I don't care about flashy), but because it solved a specific problem I'd been wrestling with: how do you audit carrier performance when you have 43 circuits, 5 carrier handoffs, and 3 different in-house support teams?
Mist's answer was brutally simple: it measures SLA metrics from the perspective of our applications, not the carrier's network. An SLA from a carrier measures transport layer statistics; Mist measures whether that matters to our users. When Crown Castle's link latency spiked to 150ms but our critical apps still responded under 30ms, Mist flagged it as informational. When that same spike hit our VoIP and caused jitter, Mist raised it as critical.
That difference—understanding the experience rather than the infrastructure—is why the platform justified itself in the first 30 days. We caught a misconfigured QoS policy on a branch router that was causing packet drops during peak hours (9am-11am daily). We'd had that issue for 18 months (one of our sites thought it was 'just how the internet works'—(surprise, surprise)).
The Numbers That Matter
After tracking 6 years of WAN costs in our procurement system, I found that 37% of our 'budget overruns' came from three sources:
- Unexplained carrier surcharges (15%)
- Overage penalties from unmonitored capacity (12%)
- Aging infrastructure that was past its support window but still under contract (10%)
We implemented a policy requiring monthly audit of carrier invoices against Mist's usage data—a 2-hour task per month—and cut overruns by 85% within 90 days.
The carrier transition itself saved $34,200 annually. The Mist platform costs us $1,800 per site per year (approximately), for a net savings of $26,700 after accounting for the software. Add in the avoided overruns and SLA penalties (our old carrier credited us $1,200 for 3 outages in the final year), and the total savings exceeded $60,000 the first year.
What I Wish I'd Known
Looking back, the 12-point checklist I created after our third billing error has saved us an estimated $8,000 in potential overcharges. But the bigger lesson is this: 5 minutes of verification beats 5 days of correction.
The conventional wisdom in procurement is to get three quotes and pick the cheapest. My experience with 200+ orders across 43 sites suggests that what matters more is having visibility into what you're actually buying—and a contracting framework that matches your operational reality.
If I were doing this again, I'd start with an audit of current invoices against a neutral observability platform like Mist before negotiating a single contract. Know your baseline. Know your usage patterns. Know which of your sites are overpaying for bandwidth they don't use (we had 12 sites running at 15% utilization on their MPLS circuits).
And for heaven's sake, don't just trust the carrier's billing. Every single one of them (yes, including Crown Castle) has error rates in the 2-5% range on complex multi-site invoices—and that's before you factor in the 'transport zone' vagaries.
One final thing: the phones and USB power delivery issue I mentioned earlier? That was a 45-minute fix once we realized the problem existed. Mist's inventory integration flagged it automatically when we deployed the new CPE. Without that visibility, we'd have blamed the hardware, swapped it twice, and still missed the root cause. The platform paid for itself on that single incident. But honestly, I'm most proud of the spreadsheet. The one I built after getting burned on hidden fees twice. (It's got 14 tabs now—overkill, I know, but I sleep better.)
