In most companies, the phone system is the first piece of infrastructure installed and the last anyone thinks to replace. It works, nobody looks at it, and it keeps routing calls according to rules written by someone who left three years ago.
The problem is that the underlying network is disappearing. AT&T has set a target of retiring copper across most of its footprint by the end of 2029, and stopped accepting new orders and moves on copper services starting in October 2025. The question is no longer whether to modernize, but when and to what.
This guide covers the definitions, compares the three generations, prices the real cost, and gives you a decision grid by company size.
What a phone system actually is
A business phone system receives calls to your business numbers and decides where they go. It also handles outbound and internal calls.
In practice, it answers four questions on every inbound call:
- Who rings? A department, a person, a group, and in what order.
- For how long? Before moving to the next destination.
- What happens if nobody picks up? Voicemail, overflow to another group, mobile forwarding, callback.
- What does the caller hear meanwhile? Greeting, hold music, queue position.
Keep that framing, because it prevents most bad purchases: a phone system is not equipment, it is a routing plan. Two companies with identical hardware and different routing plans do not have remotely the same caller experience.
The vocabulary, in one pass
- PBX (Private Branch Exchange): the legacy on-premise switch, wired to the carrier network over analog lines or T1/PRI circuits.
- IP-PBX: same idea, but voice travels over your data network. The box still lives in your building.
- Cloud phone system, also sold as hosted PBX or UCaaS: the system runs on the provider’s servers. All that remains on site are apps and maybe a few IP phones.
- SIP trunk: the pipe connecting an on-premise IP-PBX to the public phone network. It is what replaced PRI circuits.
- Softphone: the app that turns a laptop or a mobile into a desk phone.
The three generations, compared
| Criterion | Analog PBX / PRI | On-premise IP-PBX | Cloud phone system |
|---|---|---|---|
| Where the system lives | Box in your building | Server in your building | Provider’s servers |
| Upfront cost | Long since amortized | $3,000-$25,000 by seat count | None |
| Recurring cost | Lines + maintenance | Maintenance + SIP trunk | $15-40 per user per month |
| Remote and hybrid work | Manual call forwarding | Possible, needs configuring | Native |
| Adding a user | Technician visit | Visit or license | Minutes, self-service |
| Updates | None | Scheduled, with downtime | Continuous |
| Site resilience | None — site down, phones down | None without costly redundancy | Calls fail over to mobile |
| CRM integration | No | Custom development | Native with most vendors |
| Remaining lifespan | Ending with copper retirement | 7-10 years per hardware cycle | Ongoing |
How a call actually flows
Follow one inbound call to a 30-person company’s main number.
Arrival on the main number
The system identifies which number was dialed. That is what lets you run different journeys for the main line, the support line, and a rep’s direct number.
Business-hours check
First decision, and the one most often misconfigured: are we open? The calendar has to include federal holidays and company closures. Outside hours, the call goes straight to the closed greeting or to an on-call path.
Greeting, then routing
Either the call rings a reception group directly, or it goes through a menu. That menu is where half the caller experience is decided — the rules are in our IVR guide.
Ring strategy
The system rings group members: all at once, in a fixed cascade, or round-robin. This single setting decides whether calls get absorbed or land on the same two people every time.
Queue or overflow
If everyone is busy, the caller waits with an announcement or overflows to another group. A working overflow is what separates a reachable company from one that rings into the void.
Exit path
Nobody answered: transcribed voicemail, mobile forwarding, or scheduled callback. There must always be an explicit exit. A call that rings forever is a lost call.
Each of those six steps is a setting. A badly configured system is almost never a technology problem: it is one of these six decisions that was never made, so the platform applied its default.
The six criteria that actually decide
Vendor comparison sheets list thirty features. Six criteria decide.
1. Share of non-desk staff. This is the structural one. If more than a third of your team works remotely, in the field, or across sites, an on-premise system condemns you to a patchwork of forwarding rules. Cloud is then not a preference, it is the only coherent architecture.
2. Number of distinct call journeys. One line and a receptionist, or fifteen DIDs across six departments? The second case demands an admin interface a non-engineer can read, or nobody will ever touch the configuration again.
3. CRM integration. If your sales or support team lives in a CRM, screen-pop on ring and automatic call logging change the day more than any telephony feature. Check that the integration is native and two-way, not a click-to-call link.
4. Internet quality. Voice tolerates low bandwidth but not jitter and packet loss. Before signing, check upstream bandwidth at peak and whether you can prioritize voice traffic.
5. Analog devices hanging off phone lines. Elevator phones, fire alarm dialers, fax machines, gate intercoms, credit card terminals. These are routinely forgotten and remembered on cutover day.
6. Exit terms. Contract length, termination conditions, and above all: can you take your numbers and your call history with you? A contract that makes leaving painful is a contract that will let a vendor raise prices for a decade.
What it costs
Cloud
The US market range is $15 to $40 per user per month. The spread is about what is bundled:
| Price tier | What is included | Who it fits |
|---|---|---|
| $15-20 | Full system: DIDs, auto attendant, queues, voicemail, softphone, unlimited domestic calling | Companies whose need is simply to be reachable |
| $20-30 | Adds native CRM integrations, call analytics, live supervision | Sales and support teams |
| $30-40+ | Contact center features: power dialing, always-on recording, AI conversation analysis, omnichannel | Call centers and heavy outbound teams |
Three line items to check individually: usage (unlimited domestic, or metered), extra numbers (often $1-$5 per DID per month past the first), and activation fees.
On-premise IP-PBX
Upfront runs $3,000 to $25,000 depending on seat count and redundancy, plus annual maintenance, SIP trunking, and a hardware refresh every seven to ten years. Over an eight-year life, a 30-seat company often lands near the total cost of a cloud subscription — but with the capital tied up, the site-failure risk, and single-vendor dependence on top.
Comparing on-premise to cloud on purchase price alone is comparing the price of a car to the price of a subscription, and forgetting fuel, insurance, and the mechanic.
On-premise keeps two solid justifications: a regulatory or sovereignty constraint that forbids external hosting, and an industrial site with genuinely bad connectivity. Outside those two, the math points to cloud.
Decision grid by company size
| Profile | Recommendation | The trap |
|---|---|---|
| 1-9 people, single location | Entry-tier cloud, softphone only, one or two numbers | Running the business on a personal mobile: no forwarding, no continuity when that person is out |
| 10-49 people, several departments | Cloud with auto attendant, queues, and analytics | Rebuilding the old PBX tree identically instead of simplifying it |
| 50-250 people, multiple sites | Multi-site cloud, CRM integration, supervision | Underestimating change management: the technical work is two days, the habits take two months |
| Heavy outbound sales team | Cloud built for outbound, with dialing and recording | Buying two tools that do not talk to each other: telephony on one side, dialer on the other |
| Strong sovereignty constraint | On-premise IP-PBX with SIP trunking | Forgetting the failover plan: what happens when the site goes down? |
If your primary problem is outbound volume rather than inbound reception, the selection criteria are different — we covered them in the sales phone tool guide.
Migrating without dropping the phones
Inventory what exists
Every number, including the ones nobody uses, and every device wired to a phone line. This is the step everyone rushes, and the source of every bad surprise.
Redraw the routing plan
Do not copy the old one. Start again from the six routing decisions, department by department. A tree inherited from 2012 always contains dead branches.
Start the port
Submit the LOA and a recent bill or CSR to the new provider. Absolute rule: never cancel the old service before the port completes, or the number is gone for good.
Configure and test in parallel
Run the new system on test numbers while the old one stays live. Have someone who did not design each path walk through it.
Cut over, then watch for two weeks
Keep a feedback channel open. Routing defects surface in use, not in testing.
The five most expensive mistakes
- Copying the old tree. A migration is the only realistic chance to simplify. Transfer it as-is and you transfer its flaws for another decade.
- A calendar without holidays. The system applies normal hours on July 4th, rings into an empty office, and nobody finds out for months.
- No exit path. Every branch must end somewhere: voicemail, forward, callback. Never an endless ring.
- Cancelling before the port. The number returns to the pool and is unrecoverable. It happens more often than you would think.
- Ignoring analog devices. The elevator phone and the alarm dialer do not announce that they are dead. You find out when you need them.
What to take away
The phone system moved from being equipment to being a service. That shift is not cosmetic: it moves the conversation out of the wiring closet and into the org chart. The real question is no longer which box to buy, but who should answer what, and what happens when nobody can.
That also makes the decision simpler than it looks. For the vast majority of US businesses in 2026, the answer is one line: a cloud phone system, billed per user, with a routing plan written down and reviewed once a year.
Everything else — the auto attendant, the queues, the analytics — is just the execution of that plan. If you inherited a system nobody understands any more, start with the greeting and the ring strategy: that is where calls, and therefore revenue, get lost.
