VoIP is no longer a technology choice. With carriers retiring copper toward the end of the decade, it has become the default infrastructure for business telephony in the US. The useful question is no longer whether to move, but what to check so that it works well.
That is where projects diverge. A successful VoIP rollout is invisible: people make calls, it works, the bill goes down. A failed one produces three months of “the audio keeps cutting out” blamed on the technology, when the cause is almost always a misconfigured router or a connection shared with nightly backups.
This guide covers how it really works, the network prerequisites in numbers, the full cost breakdown, the rollout checklist, the E911 obligations, and how to diagnose quality problems.
How it works, in one page
VoIP splits speech into digital packets and carries them over the IP network like any other data. Three building blocks explain everything else:
- SIP is the signaling protocol. It sets up the call, handles ringing, transfer, hold and hangup. It is what “picks up the phone.”
- RTP carries the audio itself once the call is established. It often flows directly between the two endpoints, which is why quality depends so heavily on the local network.
- The codec compresses voice on the way out and rebuilds it on arrival. G.711, uncompressed, gives the best quality at roughly 100 kbps per call once network overhead is counted. Compressed codecs such as G.729 drop toward 30-40 kbps at a small cost in naturalness.
Network requirements, in numbers
This is the part projects skip, and the part that decides success.
Bandwidth needed
| Concurrent calls | Upstream bandwidth required (G.711) |
|---|---|
| 5 | about 0.5 Mbps |
| 10 | about 1 Mbps |
| 25 | about 2.5 Mbps |
| 50 | about 5 Mbps |
Those numbers surprise people by how small they are. A 50-person company never has 50 concurrent calls: real concurrency usually lands between 15% and 30% of headcount. Bandwidth is almost never the limiting factor.
The three numbers that actually matter
| Metric | What it measures | Acceptable | Symptom past it |
|---|---|---|---|
| Latency | One-way end-to-end delay | < 150 ms | Talking over each other, walkie-talkie feel |
| Jitter | Irregular packet arrival | < 30 ms | Metallic voice, warped syllables |
| Packet loss | Packets that never arrive | < 1% | Choppy audio, missing words |
The difference between a “fast” connection and a “good for voice” connection lives entirely in those three rows. A 1 Gbps fiber shared without prioritization with nightly backups and video calls can produce jitter where a smaller, well-configured circuit produces none.
The three settings that fix 90% of problems
- Prioritize voice traffic (QoS) on the router: tag voice packets and give them priority over everything else. This is the highest-return setting in the project.
- Wire the desk phones. Wi-Fi works, but it is the first suspect in any quality problem. For mobile softphone users, provision properly sized access points.
- Separate voice into its own VLAN once you pass roughly twenty seats. That isolates voice from the bursts of ordinary office traffic.
Real cost, line by line
The advertised per-user price is only part of the bill. Here is the full breakdown to ask for in a quote.
| Line item | Order of magnitude | The catch |
|---|---|---|
| Per-user subscription | $15-40 per month | Check what counts as a “user”: a person, a seat, or a concurrent line |
| Usage | Usually unlimited domestic | International and toll-free inbound are billed separately |
| Extra numbers | $1-5 per DID per month | Charged past the first, multiply by your DID count |
| Activation | $0 to a few hundred dollars | Sometimes per user, sometimes flat |
| Desk phones | $60-250 per phone, purchased | Optional: softphones cover most users |
| Analog adapters (ATA) | $40-150 per device | Needed for retained fax, gate, alarm |
| Number porting | Usually free | Confirm it is, and confirm the timeline |
| Regulatory fees and surcharges | Varies | Ask for an all-in monthly figure, not the headline rate |
The right reflex when comparing quotes: ask for total cost over 36 months, everything included, rather than the monthly per-user price.
Compared with a legacy installation — line charges, PBX maintenance, metered usage, hardware refresh every seven to ten years — moving to VoIP lowers the recurring cost in the large majority of cases, with flexibility on top.
The eight-point rollout checklist
Inventory numbers and devices
Every number, including unused ones, and everything plugged into a phone jack: fax, card terminal, elevator, alarm, gate, postage meter. The most rushed step and the source of every bad surprise.
Qualify the network at each site
Upstream bandwidth at peak, latency, jitter, packet loss. A full-day test beats a single reading at 2 pm on a Tuesday.
Size concurrency
How many simultaneous calls at peak? Look at your current system’s history rather than estimating. Most companies over-provision by a factor of three.
Write the routing plan
Groups, hours including holidays, overflow, exit paths. Do not copy the old one — this is the only moment you will get to simplify it. Detail in our business phone system guide.
Configure the network
Voice prioritization, voice VLAN if relevant, SIP ALG disabled, required flows opened. Do this before testing, or you will diagnose the wrong problem.
Register E911 locations
Direct 911 dialing with no prefix, on-site notification when 911 is dialed, and a dispatchable location per site and per floor. Kari’s Law and RAY BAUM’S Act make this mandatory, not optional.
Port, then test path by path
Submit the LOA with a recent bill or CSR. Test every routing branch with someone who did not design it, including night, weekend and holiday scenarios by forcing the date.
Cut over and watch for two weeks
Open a dedicated feedback channel. Routing defects surface in use. Re-read the call analytics at day 15: they show what nobody reports.
Diagnosing a quality problem
When “the audio keeps cutting out,” the reflex is to call the provider. In practice the cause is almost always local. Here is the symptom-to-cause table.
| Symptom | Most likely cause | What to check |
|---|---|---|
| Choppy audio, missing words | Packet loss | No voice prioritization, saturated Wi-Fi, congested link |
| Metallic or robotic voice | High jitter | Unprioritized competing traffic, undersized VPN |
| Talking over each other | High latency | Unusual network routing, VPN taking a detour |
| Echo | Local audio loop | Headset or speakerphone, volume too high, mis-set handset |
| One-way audio | Media stream blocked | Firewall, NAT misconfiguration, SIP ALG enabled |
| Calls always drop at the same duration | Session timeout too short | Router settings, NAT table timers |
Security and compliance
VoIP inherits the risks of the internet. Three measures cover the essentials, in this order of importance.
1. Cap outbound calling. This is the priority. Toll fraud means compromising an account to place mass calls to premium-rate destinations, typically overnight or over a weekend so nobody notices before Monday. Set three caps on day one: an allow-list of permitted international destinations, a maximum number of calls per hour per account, and a spend alert.
2. Encrypt. SIP over TLS for signaling, SRTP for media. Standard among serious providers, but not always on by default — ask explicitly.
3. Unique credentials per account. Never a shared password across handsets, never a generic account. A compromised extension must be disconnectable without cutting off the company.
On the regulatory side, two items deserve attention. E911: Kari’s Law requires direct 911 dialing without a prefix and an on-site notification when 911 is called; RAY BAUM’S Act requires a dispatchable location to be transmitted. Because VoIP numbers are tied to a registered address rather than a detected position, those registrations must be maintained. And call recording: consent requirements vary by state — some require only one party’s consent, others require all parties’ — so announce recording and set a justified retention period.
The five most costly mistakes
- Migrating without qualifying the network. You will find the problem in production, under pressure, and it will be blamed on VoIP.
- Leaving SIP ALG enabled. The feature meant to help breaks signaling on a large share of consumer routers.
- Forgetting outbound caps. Uncapped toll fraud runs into thousands of dollars over a single weekend.
- Copying the old routing plan. You inherit the flaws of an organization that no longer exists.
- Ignoring analog devices. Elevators and alarms do not report their own failure — the detail is in our guide to POTS line retirement.
What to take away
VoIP is a mature technology: the problems attributed to it are, overwhelmingly, local network and configuration problems. Three settings — voice prioritization, wired desk phones, SIP ALG disabled — eliminate most quality incidents.
The rest of the project is inventory and decisions: which numbers, which devices, which routing plan. The technical part settles in days.
And if your primary use is high-volume outbound rather than inbound, the selection criteria change entirely — dialing, caller ID management, CRM integration. We covered those in the sales prospecting tools guide.
