An automated call system places calls from a contact list without a person dialing each number, then either plays a recorded message, runs an IVR menu, or hands a live answer to an agent. For customer follow-ups it turns a job that eats an afternoon into one that runs in the background. The catch is that most of the work sits in call pacing, answering machine detection, and consent, not in the recording itself.
We build and support ICTBroadcast, an Asterisk-based platform that runs voice, SMS, fax, and email campaigns from one place, so what follows is written from watching these campaigns run rather than from a feature list.
What automated phone system software actually does
Strip away the marketing and there are four moving parts.
The dialer decides which numbers to call and when. The media layer plays audio, records replies, and collects keypresses. The detection layer works out whether a human or a machine picked up. And the disposition layer writes down what happened so the next attempt is smarter than the last one.
Most buying mistakes come from judging the first part and ignoring the third. Detection is where campaigns quietly fall apart, and it never shows up in a demo.
The dial modes, and which one fits a follow-up
Broadcast mode calls a list and plays a message with no agent involved. Appointment reminders, payment-due notices, and delivery windows fit here. It’s the cheapest per contact by a wide margin.
Preview mode shows an agent the record first and lets them press dial. Slow, and right for high-value accounts where a bad opening line costs you the relationship.
Progressive mode dials one number per free agent. Predictive mode dials several per agent and connects only the live answers, which is faster and carries the abandonment risk we get into below. If you’re doing post-purchase follow-ups on a warm list, progressive is usually the better trade: answer rates on warm lists run high enough that predictive over-dials and starts dropping calls.
Our two explainers go deeper on the mechanics: what an auto dialer is and how a predictive dialer decides its pacing.
Answering machine detection is the part that decides your results
Roughly two thirds of outbound attempts to consumer mobiles reach voicemail. What your system does in the first two seconds of that decides whether the campaign works.
Detection listens to the shape of the answer. A human says “hello” and stops. A machine keeps talking, and the greeting has a different length and rhythm. The algorithm guesses, and it is a guess. Tune it aggressively and it starts hanging up on real people who answered slowly, which is the worst outcome because those are your engaged customers. Tune it loosely and you dump agent time into voicemail.
The number to watch is not detection accuracy in isolation. It’s false positives on humans. We’d rather leave a few extra machine messages than cut off a customer who picked up while carrying groceries.
One practical setting most teams miss: give the system permission to wait for the beep and drop a recorded message instead of hanging up. A voicemail with a callback number recovers a chunk of the contacts you’d otherwise lose, and it costs nothing extra once the call is already connected.
Consent and calling rules come before anything else
In the United States, calling a mobile with an automated system needs prior express consent, and marketing content needs it in writing. Follow-ups on an existing transaction sit in a friendlier category than cold marketing, but the line moves depending on what your message says. A delivery notification is not a promotion. A delivery notification with a discount code attached is.
The mechanical rules your software has to enforce, whatever your legal position:
- Scrub against the federal Do Not Call registry, plus state lists, and re-scrub at least every 31 days
- Keep a permanent internal DNC that applies across every campaign, not per campaign
- Respect 8am to 9pm in the called party’s time zone, worked out from their number and not from where your server sits
- Keep abandoned calls under 3% per campaign over 30 days if you dial predictively
- Log consent against the contact record so you can produce it two years later
Since August 2026 the EU AI Act also requires a spoken disclosure when the voice on the call is synthetic, and the FCC has signalled it’s heading the same way. We wrote up what the AI disclosure rules change inside a dialer. If your follow-ups use a synthetic voice, that disclosure belongs in the first sentence, not in a footer.
There’s a second reason to care about all this beyond fines. Carriers watch complaint rates, and a number with a bad reputation gets labelled before your call ever rings. Our piece on why calls that pass STIR/SHAKEN still show Spam Likely covers what actually drives that label.
Retry logic separates a good campaign from a busy one
Most teams set three attempts and call it done. The attempts land at the same time of day, hit the same voicemail, and burn the list.
What works better is varying the window. If attempt one was a weekday morning, put attempt two in an early evening slot and attempt three on a Saturday late morning. Cap total attempts per contact rather than per campaign, so a customer in three campaigns doesn’t get nine calls in a week. And treat a voicemail drop as a soft attempt rather than a hard one, because you didn’t actually reach anyone.
A payment reminder campaign we watched went from 31% contact rate to 44% by doing nothing except spreading three attempts across three different day parts. Same list, same recording, same software.
Where to put the follow-up in your process
The follow-ups worth automating share a shape: they’re time-bound, the message is the same for everyone, and a reply is useful but not required.
Good fits are appointment reminders 24 to 48 hours ahead, renewal notices ahead of a lapse, post-delivery satisfaction checks, payment reminders before a due date, and no-show recovery on the same day. Two of those pay for the software on their own. Appointment reminders cut no-shows sharply in clinics and service businesses, and a renewal notice three weeks before a lapse catches customers while cancelling is still effort rather than a decision already made.
Poor fits are anything where the answer needs judgement, anything carrying bad news, and any first contact with a cold list. An automated call cannot handle a customer who wants to argue, and putting one in front of them creates a complaint you’ll pay for twice.
The pattern we’d push most teams toward is a two-stage one. Send an automated voice call, and route anyone who presses 1 straight to an agent. You get the reach of broadcasting with a human on the calls that need one, and your agents only ever talk to people who asked to talk.
What to check before you buy
Feature lists all look identical. These are the questions that actually separate platforms:
- Can you tune answering machine detection, or is it a switch? A single on/off toggle means you’ll be stuck with whatever their default costs you.
- Does the abandonment cap apply per campaign over a rolling window? Anything simpler will not survive an audit.
- Are dispositions writable back to your CRM? If dispositions live only in the dialer, nobody outside the call floor sees them and the data dies there.
- Does pricing scale with agents, minutes, or neither? Per-seat billing punishes exactly the seasonal ramp most teams need.
- Can you bring your own SIP trunk? Bundled minutes usually cost multiples of wholesale rates.
- Where do recordings live? Someone will eventually ask you to produce or delete one.
On the last two, self-hosting changes the maths. ICTBroadcast is licensed per installation rather than per seat, runs on your own server, and works with any standard SIP carrier, so a team that doubles for a busy quarter pays for the extra minutes and nothing else. The REST API ships with the Service Provider Edition at no extra cost, so pushing contacts in from a CRM and pulling dispositions back is part of the product rather than an add-on. Full detail sits on the packages page.
For teams that want the follow-up call to hold a short conversation rather than read a script, ICTBroadcast integrates with Google Dialogflow, so the agent you build there answers on the call and hands off to a person when it hits something it can’t handle.
A realistic first campaign
If you’re starting from nothing, this order works:
- Pick one follow-up type with a clear time trigger. Appointment reminders are the usual starting point because the value is easy to measure.
- Export a small list, 200 to 500 contacts, and scrub it before you dial anything.
- Record the message yourself and keep it under 20 seconds. Say who you are in the first five words.
- Run it in broadcast mode with voicemail drop enabled and press-1 transfer to an agent.
- Read the disposition report the next morning and fix the message before you scale the list.
Most of what you learn arrives in that first report. Contact rate tells you whether the list is any good, the press-1 rate tells you whether the message is, and the complaint count tells you whether you should be calling these people at all.
FAQ
Is automated calling legal for customer follow-ups? In most cases yes, when you have consent and the call relates to an existing transaction. Marketing content raises the bar to written consent. The rules that never bend are DNC scrubbing, calling hours, and honouring opt-outs immediately.
How is this different from a predictive dialer? A predictive dialer is one mode inside an automated call system, the one that dials ahead of your agents. Broadcast campaigns need no agents at all. Most follow-up programmes use broadcast with an agent transfer rather than predictive dialing.
Can it send SMS as well as calls? Yes, and mixing the two usually beats either alone. A common pattern is a voice call first, then an SMS with the same information to anyone who didn’t answer. ICTBroadcast runs voice, SMS, fax, and email campaigns from the same contact list.
What contact rate should I expect? On a warm list of existing customers, 35% to 50% live answers across three attempts is normal. Cold consumer lists run far lower. If a vendor quotes you numbers well above that range, ask which list they measured.
Do I need my own phone system? No. You need a SIP trunk from a carrier and somewhere to run the software. Many teams run it on a single cloud server and add capacity only when concurrent call volume demands it.
How many calls can one server handle? That depends on your hardware and codec, but a modest server handles a few hundred concurrent channels comfortably. Capacity is rarely the constraint; carrier limits and calling hours usually bind first.
If you want to see the campaign builder before committing to anything, the live demo is open, and the campaign setup walkthrough covers the same steps listed above with screenshots.