SD-WAN, LTE backup, and the unsexy network decisions running your restaurant

Most restaurant operators learn the importance of their network during the worst possible 90 seconds of a Saturday night. Up to that moment, a network is something somebody set up four years ago that nobody remembers ordering. After that moment, it's the entire reason a $4,800 dinner rush quietly walks out the door.

There's no glory in this part of the operation. There's also no place where small, correct decisions matter more. Below is the playbook we run when we onboard a multi-location restaurant, in plain English, with no vendor pitches.

Start with what the network actually has to do

Before any vendor conversation, write down what the network supports at the operation. For a typical 2-unit casual restaurant, the list looks something like this:

• POS terminals (4–8 per location, talking constantly to a cloud back end)
• Card payment terminals (PCI-scoped traffic — has to be isolated)
• KDS displays in the kitchen (mostly internal, occasional cloud sync)
• Online ordering ingestion from your domain and 2–4 marketplaces
• Office back-of-house: laptops, accounting, inventory tablets
• Guest Wi-Fi (separate, monitored, throttled)
• Music and signage, often cloud-streamed
• Cameras and DVR (heavy outbound bandwidth)
• VOIP phone system

Write that list down per location. Now you have the actual requirement, and any vendor proposal can be measured against it. Most can't be.

The PCI conversation, before any other one

Cardholder data has to be on a network segment that nothing else can reach. This is non-negotiable. Most older restaurants we audit are 'flat' — every device on one VLAN — which means the back-office laptop and the card terminal share a network. That's a PCI 4.0 violation and an insurance liability.

Step one of any network refresh: at minimum 4 VLANs (cardholder, back-of-house, guest, IoT). Most cloud-managed platforms (Meraki, UniFi, Cisco) make this clean. If you have a single $99 router from your ISP, you don't have a network — you have an obstacle to network.

What SD-WAN actually buys you

SD-WAN is a much-abused term. In a restaurant context it means three useful things and one over-sold thing.

What it actually buys you

• Multi-link aggregation: combine your fiber, cable, and LTE so traffic flows over whichever is healthiest in real time.
• Sub-second failover: when fiber drops, the LTE picks up before a TCP connection times out — payment terminals don't reconnect, they keep working.
• Centralized policy: change a firewall rule once, push to every location.

What it does not buy you

• Faster internet. SD-WAN is not a speed-up; it's a reliability-up.
• Magic security. SD-WAN tunnels are encrypted, but you still need segmentation, monitoring, and PCI scope discipline.

If you have one location and one ISP, SD-WAN is overkill. If you have three or more locations, the central-policy and link-aggregation cases start to pay for themselves.

LTE backup is no longer optional

ISP outages used to be rare enough that you could absorb them. They are not anymore — fiber cuts, neighborhood power events, and cellular-vs-cable contention have all gotten worse, not better. LTE backup is the cheapest insurance you can buy.

What 'good' looks like: a managed cellular router with a static IP from a major carrier (T-Mobile, Verizon, AT&T), bonded to your primary at the SD-WAN layer so failover is automatic. Cost: $80–140 per location per month, all-in. The first time a 90-minute outage doesn't take you offline, it pays for two years.

When to buy hardware vs lease it

We get asked this a lot. Our short answer: lease — usually as part of a managed services agreement. Reasons:

• Cloud-managed network gear has 5–7 year EOL cycles. Owning a thing that's about to be EOL is worse than not owning it.
• Firmware needs to be patched within 30 days of a CVE, and operators don't do this on their own. A managed agreement with patching SLA is a real value.
• When something dies, replacement is a phone call, not a procurement cycle.

Practical 30-day plan for a 3–10 unit operator

1. 01Audit each location's actual network device inventory. Take photos. Note model numbers and firmware ages.
2. 02Identify the locations with no LTE backup. Quote one. Install in the slowest one first.
3. 03Pull the PCI Self-Assessment Questionnaire for each location and check whether your current segmentation passes. If not, schedule remediation in the next quarter.
4. 04Get one written quote for cloud-managed network with monitoring, patching, and 24/7 escalation. Compare against your current 'when something breaks I call my brother-in-law' situation.
5. 05Decide which model you want for the next two years and migrate there with a vendor-agnostic partner who can run all of the above without owning the hardware.

None of this is glamorous. None of this is on the menu. All of this is what the menu runs on. Pay attention to it now, or pay for it on a Saturday night.