Business internet tiers for voice quality, what speeds are enough

Suggested URL slug: business-internet-tiers-voice-quality-speeds-enough

Meta description: How much internet speed do business voice systems need? Practical per-call numbers, tiered sizing, tests, and a short runbook to validate call quality.

Last reviewed: September 30, 2025

Hearing a dropped call during a payment or an important client conversation is always loud in a quiet meeting room. The right internet tier prevents those interruptions without overpaying for unused capacity.

TL;DR: Reserve roughly 80–100 kbps per concurrent G.711 call and about 24–30 kbps per concurrent G.729 call, including protocol overhead. Match your tier to peak concurrent calls, not total users. Prioritize jitter <20 ms, packet loss <1%, and MOS ≥4.0 as pass/fail criteria. Run a 15-minute test during a busy hour that measures concurrent call count, jitter, loss, and MOS before you buy or cut over.

At-a-glance per-call sizing (first screen)

Codec	Reserve per concurrent call	Practical note
G.711 (uncompressed)	~80–100 kbps	Good voice quality, higher bandwidth; simple math for planning. See Cisco calculations.
G.729 (compressed)	~24–30 kbps	Lower bandwidth but slightly lower fidelity; useful for constrained WAN links.

Quick method note: per-call reserve includes RTP/UDP/IP and Ethernet headers and a small margin for silence suppression or extra signaling. For authoritative per-call math, see Cisco: Bandwidth, Latency, and QoS Considerations and a helpful calculator table at TechTarget: VoIP bandwidth calculations.

What is a business internet tier for voice, and why does it matter?

Internet tiers are the advertised speed tiers carriers sell: 10/1, 25/5, 100/20, and so on. For voice quality, the critical variables are not only raw Mbps but also latency, jitter, packet loss, and the link’s stability during your busiest hour. Two sites with the same nominal speed can behave very differently under load.

Think of voice sizing in two steps. First, plan capacity for peak concurrent calls using per-call reserve numbers above. Second, validate the link’s quality metrics under load. A single busy hour of poor jitter can make a high-speed plan unusable for voice.

Mini-story: a regional clinic upgraded from a 25/5 plan to 100/20 because call drops and echo occurred every afternoon. After measuring, we found simultaneous uploads from 10 staff backing up large files. A QoS policy and a small uplink upgrade fixed it without doubling the whole bill.

How to pick a tier: simple sizing framework

Start with concurrent active calls at peak. Multiply by codec reserve (G.711 ≈100 kbps, G.729 ≈30 kbps). Add 15–25% headroom for signaling, retransmits, and other traffic. Then check uplink capacity and concurrency limits for other critical apps like POS or video.

Example quick math: 10 concurrent G.711 calls → 10 × 0.1 Mbps = 1.0 Mbps. With 25% headroom → 1.25 Mbps. Practically, this fits in a 5 Mbps uplink with room for other uses. For 50 concurrent G.711 calls, plan at least 6–8 Mbps of dedicated voice-capable uplink plus QoS.

Tier matrix (practical guidance)

Scenario	Concurrent calls	Suggested minimum uplink	Notes
Home / remote desk	1–2	5 Mbps	G.711 works; prefer wired; enable DSCP if available.
Small office	3–8	10–25 Mbps	Reserve per-call math; prefer symmetric uplink where possible.
Medium office	9–30	25–100 Mbps	Consider dual-WAN or SD-WAN with voice steering for redundancy.
Contact center / heavy voice	30–200+	100 Mbps+ and dedicated voice paths	Use carrier SIP trunking with QoS and SBCs; consider private circuits for large sites.

Sources and authority: per-call bandwidth math is consistent with vendor guidance; see Cisco bandwidth reference and TechTarget’s calculator for header overhead assumptions.

How to test before you buy or cut over (short runbook)

1. Measure peak concurrent calls during a representative busy 15-minute window.
2. Run synthetic test calls from the site to a test number while generating typical data load (file uploads, POS traffic). Measure MOS, jitter, packet loss, latency. Tools: your PBX test call reports, or external SIP test services.
3. Pass/fail checks: MOS ≥ 4.0, jitter ≤ 20 ms at the phone port, packet loss < 1% on voice VLAN. If any fail, either raise uplink capacity, enforce QoS, or switch codec. (Our POV: fix observability first, then change codecs.)
4. Validate failover: simulate primary ISP loss and confirm calls either survive locally (survivable gateway) or failover to the secondary link with acceptable quality.
5. Document results and maintain a 30-day hypercare plan evaluating weekly trends.

Note: MOS, jitter, and loss thresholds are practical targets many integrators use; they map to user experience better than raw Mbps. If you want reference tables, see TechTarget: VoIP bandwidth calculations and vendor QoS guidance from Cisco.

Pros and cons: SD-WAN / dual-WAN vs single business tier

Pick single high-quality symmetric fiber when you can. But many Canadian and regional sites cannot get fiber quickly. Dual-WAN (fiber + LTE) or SD-WAN gives resiliency and the ability to route voice on the best path in real time.

• Single premium circuit (fiber): simpler, predictable latency, easier QoS enforcement. Risk: single point of failure unless you add backup.
• Dual-WAN with QoS and failover: cheaper resilience. Trade-off: more complex routing and sometimes asymmetric latency causing short glitches.
• SD-WAN with application steering: improves path selection and observability. Trade-off: requires configuration and monitoring discipline.

Our POV: select the approach that matches your failure tolerance. If you run card payments or emergency dispatch, design for graceful failure first, then optimize for cost.

Application to common personas and ICPs (practical vignettes)

Clinic manager (small single site): If you average two simultaneous calls and run electronic medical records uploads at midday, pick 25/5 or 50/10 and enforce uplink QoS for voice. Test during peak appointment hours. Confirm NG911 routing and address accuracy with your voice provider.

Retail manager (multi-site, seasonal peaks): Use per-site math for concurrent calls at peak. For seasonal spikes, consider temporary voice routing rules or a short-term uplink bump. For card terminals, validate a 30-minute busy-window transaction test.

IT manager for distributed sales teams: For many remote users, require wired connections or minimum home uplink (5–10 Mbps) and mandate headset quality. Use softphone codecs and centralized logging to detect regional jitter patterns.

Objections and common pitfalls

“We have 100 Mbps — why are calls still bad?” Because raw downlink numbers alone don’t guarantee low jitter, low loss, or stable uplink bandwidth. Voice is very sensitive to bursty upload traffic. Solve with QoS, policing, and an uplink sized for peak concurrent calls.

“Can we just change the codec to fix issues?” Codec changes reduce bandwidth but can degrade quality. Our sequence: measure, control network behavior (QoS, limit bulk uploads), then consider codec as a resource optimization step. Never swap codecs without testing MOS on real calls.

“Is silence suppression safe?” It reduces average bandwidth but adds complexity for network planning and can frustrate agents who expect continuous media. Use it only when bandwidth is constrained and after testing.

How our company solves this

Outcome: Reliable, measured call quality for your sites in Alberta and BC. How we do it: we size bandwidth to peak concurrent calls, enforce end-to-end QoS, and run a pilot that measures MOS, jitter, and loss. If you want us to size bandwidth and test jitter at your sites, we can run it this week.

FAQ

How many Mbps do I need per phone?

Plan using concurrent active calls. Reserve ~100 kbps per G.711 call and ~30 kbps per G.729 call, then add 15–25% headroom for signaling and other traffic.

Will Wi-Fi voice work the same as wired?

Not automatically. Wi-Fi needs AP capacity for voice, correct airtime settings, and WMM/802.11e QoS. For critical desks, prefer wired where you can.

Can LTE or 5G be used as a backup?

Yes. LTE/5G make good secondary paths. Ensure your SIP provider and SBC support the failover design and test for voice quality under failover conditions.

What cities do you operate in and test in?

We commonly design and test for Edmonton, Calgary, Red Deer, and Cranbrook. Local routing and carrier availability can change results; always validate in the target site.

Sources

• Cisco: Bandwidth consumption for VoIP calls
• TechTarget: VoIP bandwidth calculations
• ATCVoIP: VoIP network requirements and best practices

Optional images and alt text suggestions

• Per-call bandwidth table image — alt: “Table showing per-call bandwidth reserves for G.711 and G.729 codecs.”
• Internet tier matrix diagram — alt: “Matrix comparing small office, medium office, and contact center tiers with suggested uplink ranges.”
• Runbook flowchart — alt: “Five-step runbook for measuring and validating call quality before cutover.”

“,