How to Install Your Own AI Detection Model in Ecsow

The English model is already installed. Five more languages are one download away: where to get them, which folder to point at, and when to restart.

The short version
The English model is already inside Ecsow. Five more languages are free downloads: Spanish, Arabic, Mandarin, Japanese, Korean, Vietnamese. The whole job is four moves — download, unzip, point AI Settings at the folder, restart. About ten minutes, most of it the download.

Ecsow’s AI-Enhanced Detection comes with a lightweight English speech model built-in. This is why the model fields may be left empty on a fresh installation.

We also offer free downloadable models for five other languages: Spanish, Arabic, Mandarin Chinese, Japanese, Korean and Vietnamese.

The whole setup only has four simple steps: download the model file, extract it, select the model folder inside AI settings, then restart Ecsow. Most of the waiting time is for the file download, and the process usually takes around 10 minutes.

You only need to install an extra language model if you are handling non‑English calls. If you haven’t turned on AI detection yet, please check our introductory setup guide first.

Step 1: Download the model

Go to the Ecsow download page and find the AI detection models. Six languages are listed.

LanguageModel typeSize
Spanishbase48 MB
Arabicbase114 MB
Chinese (Mandarin)base95 MB
Japanesetiny46 MB
Koreantiny47 MB
Vietnamesebase102 MB

Each link is a .tar.bz2 file. The “base” models hear more accurately than the “tiny” ones and use more CPU. Pick the language you actually call in — the model only understands one language at a time.

Step 2: Extract the archive to a local folder

ou can use 7‑Zip or Windows’ built‑in extraction tool. .tar.bz2 requires two extraction steps: first unpack the .bz2 file, then extract the .tar file inside it.

Place the extracted files in a simple path, for example: C:\EcsowModels\moonshine-base-es

Tip: Use a short folder path without spaces. This helps avoid loading errors.

Now check the folder. It must contain these three files:

  • encoder_model.ort
  • decoder_model_merged.ort
  • tokens.txt

If any file is missing, you may have opened the outer archive folder. Go deeper into the extracted directory and check again.

Step 3: Point Ecsow to the model folder

Open Tools > Options > Voice Activity Detection and click AI Settings. This is the panel you should see.

Ecsow AI-Enhanced Detection Settings with the Moonshine model folder field and its Browse button highlighted
AI Model settings

Find the setting labelled Moonshine model folder, click Browse, and select the folder you extracted in Step 2.

Ecsow will validate the folder automatically once selected. If you see Moonshine v2 model OK in the status bar at the bottom, the model loads correctly. Save your settings.

You do not need to manually select the language; Ecsow detects it automatically from the model files.

Step 4: Match the rules file to the language

After switching to a new language model, you must use the corresponding rules.tsv file.

The rules file (rules.tsv) is what turns transcribed speech into a decision — person, voicemail, screening assistant, phone menu. The default one matches the English model.

If you point the model at Spanish and leave the English rules in place, the two disagree. The model hears Spanish, the rules expect English, and the classification stops being trustworthy.

So when you change the model, change the rules file to match. Not sure which file fits your language? Ask support before you go live with it. It is a one-line fix once you have the right file.

Step 5: Save settings and restart Ecsow

Click OK, then close and restart Ecsow. The new model loads at startup, so a restart is what makes it take effect.

After the restart, the same status line you saw in step 3 is your confirmation. Then make a short test call and check the detection behaves the way the setup guide describes.

When something looks wrong

Symptom Likely cause
Status line does not say “model OK” Wrong folder, or one of the three files is missing. Recheck step 2.
Detections make no sense in the new language The rules file was not changed. Recheck step 4.
The machine got slower on busy hours A base model costs more CPU than tiny. Watch the task manager during your busiest hour, and lower ASR workers if calls start queuing.

None of these need a reinstall. Undo the field, restart, and the default English model comes back.

Quick answers

Do I need any of this for English calls?

No. The English model ships inside the installer. Leave the model field empty and it works.

Where do the model files come from?

The links on the download page point to the public release page of the sherpa-onnx speech project, which builds the Moonshine models Ecsow uses. Nothing is repackaged or resold.

Does the model send my audio anywhere?

No. The speech model runs on the machine running Ecsow. Nothing is uploaded and there is no per-minute charge, whichever model you use.

Which editions can load a custom model?

Both. Ecsow Auto Dialer and Ecsow Voice Broadcast share the same settings panel and the same model folder field.

Should I pick the tiny or the base model?

If both exist for your language, tiny is lighter on CPU and base is more accurate. On an older machine with many lines, start with tiny. If your CPU sits mostly idle while calling, move up to base.

Answering Machine Detection: Old Way vs AI

Twelve blunt answers: why old detection fails, why one mistake costs thirty times more, and how to measure your own dialer in twenty minutes.

Twelve questions, in the order people ask them.

The short answer: old detection listens to how the call sounds, new detection reads what was said. Everything below is that difference, and why it costs money when you get it wrong.

What is answering machine detection, in one sentence?

Software that decides whether a person or a machine answered, before your agent starts talking.

Every dialer has some version of it. The cheap ones do one trick. The expensive ones stack all of them.

How does the old method decide?

It measures sound, in four ways:

Method How it decides Where it breaks
No detection Connects every answer Agents talk to machines all day
Beep detection Listens for the tone Many networks never send one
Greeting length Long greeting means machine Short greetings, long hellos
Silence pattern Waits for a gap in speech People who pause, hold music

Notice what all four have in common. None of them knows what the person said. They measure timing and tone, then guess.

So why does it get it wrong so often?

Because length and silence do not tell you who answered.

A person who picks up with “Hello, Dave speaking, who is this?” sounds exactly like a long greeting. A voicemail that says “Hi, leave a message” sounds exactly like a short hello. The old method has no way to separate those.

Everyone talks about a slider. What is that?

Because the old method measures one thing, it has one control. Turn it stricter and you hang up on more machines. Turn it looser and you hang up on fewer machines, and on more people.

Here is the part people miss. Both mistakes come from the same measurement. So the slider does not reduce errors. It moves them.

  Short answer Long answer
A person “Hello?” “Hello, Dave speaking, who is this?”
A machine “Hi, leave a message.” “You have reached the office. Please leave a message after the tone.”

Top right and bottom left are the problem cells. A person with a long, formal greeting looks like a machine. A short voicemail looks like a person.

No setting separates them, because the thing being measured is not the thing that differs. Length is not meaning. Words are meaning.

Which mistake is worse?

Hanging up on a real person, by a wide margin. Roughly thirty times worse.

Take 1,000 answered calls and an error rate of one in ten. That is 100 mistakes each way.

Hanging up on 100 real people10 agent-hours
Talking to 100 voicemails20 minutes

Why is it thirty times worse?

Because one costs a conversation and the other costs an opening.

A lost conversation is about three minutes of talk time. An agent produces roughly thirty minutes of talk per paid hour, so the other thirty minutes go to notes, callbacks and waiting. That means ten paid hours to produce five hours of conversation, which is why 100 lost calls cost about ten agent-hours.

A wasted opening is twelve seconds. One hundred of them is twenty minutes.

Ten agent-hours is roughly $40 in Manila and $180 in Ohio. The wasted openings cost about a dollar.

So most systems are tuned the wrong way round. They protect the agent’s patience because that mistake is loud and annoying, and they pay for the expensive mistake because it is silent.

This is the sentence worth remembering: the call you dropped looks exactly like a no-answer in your report. Same row, same zero. You will never see it.

What does the new method do instead?

It turns the first few seconds into text and reads it. That is the whole trick.

Once you have words, length stops mattering. Once you have words, categories that never existed become possible.

What answers Old verdict What it should be
A person human human
Voicemail machine voicemail
Screening assistant machine, so it hangs up a real person is behind it
“Please hold while I transfer you” machine, so it hangs up wait for the transfer
Phone menu machine phone menu

Two right, one right by accident, two wrong.

The two wrong ones are the interesting ones. A screening assistant is a guard standing in front of a live prospect, and the dialer hangs up on the guard. A transfer message is the phone system saying “hold on, I am fetching the person you called”, and the dialer hangs up on that too.

You cannot fix either with a slider, because the old method has no word for “screening assistant”. It only has human and machine.

Is it really just reading words?

Yes. Rules look for phrases — “leave a message”, “you have reached”, “who is calling”, “please hold” — and the first match wins.

There is no mystery in the logic. The hard part is doing it fast and locally, on the same PC that is running the calls.

Does it still get things wrong?

Yes, and here is the list I would want if I were buying.

  • It waits about three seconds before your agent speaks. You cannot classify speech you have not heard.
  • One-word answers are hard. A bare “Hello?” gives the model almost nothing.
  • Phone audio is narrowband. 8 kHz, compressed, often noisy. That limits the input, not the model.
  • Accents and rare languages are weaker, especially with the small default model.
  • People can sound like voicemail. A receptionist reading a script will do it.
  • It uses CPU. On a slow machine, calls queue and classification slows down.

None of that is a reason to keep the slider. All of it is a reason to test before you trust it.

How do I measure my own in twenty minutes?

Do not take a vendor’s accuracy number, mine included. Measure your list.

  1. Pick 50 numbers you can identify in advance: 20 that a person will answer, 20 that go to voicemail, 10 with a menu.
  2. Dial them with the detector on.
  3. Write down four counts: humans connected, voicemails caught, voicemails missed, humans wrongly dropped.
  4. Look at the last one first. If humans are dropped, the detector is costing you money however good the other three look.

What is bad about Ecsow’s version?

  • Three seconds of delay on every answered call, by default. Faster means less to read, and less to read means more guessing.
  • It eats CPU. Beep detection runs on almost nothing. This runs a speech model.
  • Six workers is the ceiling in the 32-bit build. Beyond that, calls queue.
  • Everything is local. No cloud fallback when the machine is busy, and no bigger model unless you install one.
  • You can edit the rules, which means you can also break them.

And the biggest one: better detection does not fix a bad list. If most of your numbers are wrong, you now reach voicemail more efficiently. That is not the same as selling more.

Should I turn it on?

If you run five agents and your list is reasonable, yes. The three seconds are annoying for a day and then nobody notices.

If the machine is old, or you dial thirty lines from one desktop, test it at peak first. Queueing calls is worse than a wrong guess.

And if you are still on a dialer whose only control is a slider, at least point the slider the other way. Protect the live person, not the agent’s patience. The wasted opening costs a dollar. The lost call costs a customer.

How to turn it on, and what each setting does is the companion piece. For why saved time matters at all, the talk-time arithmetic is the place to start.

Anything else?

Can it tell a Google screening assistant from voicemail?

Yes. Screening assistants ask who is calling and why, which is a different sentence from a voicemail greeting. Ecsow can play a short recorded reply and wait for the call to be put through.

Does it work on mobile numbers?

Yes, and mobile is where most of the value is. Mobile greetings are short, which was the hardest case for the old duration-based method.

What is the difference between AMD and AI detection?

AMD measures sound: greeting length, beeps, silences. AI detection transcribes the first seconds and reads the words. AMD has two outcomes. This has five.

How long before my agent can speak?

About three seconds by default. That is the cost of listening first. Shorten the initial listen and the agent speaks sooner, but it has read less, so it guesses more.

AI-Enhanced Detection: Human or Voicemail?

One checkbox, then three steps: what the five kinds of answer are, the five settings worth knowing, and what it still cannot do.

It listens to the first three seconds of an answered call, turns the speech into text on your own PC, and decides who picked up.

That is the whole idea. Below is the setup, in three steps, and the five numbers you may want to change.

1Turn it on

Open Tools > Options > Voice Activity Detection and tick Enable AI-Enhanced Detection.

That is the entire default setup. A speech model ships inside the installer, so you can leave the model and rules fields empty.

Ecsow Voice Activity Detection settings with the AI-Enhanced Detection checkbox and the AI Settings panel
The checkbox on the left starts it. The panel on the right changes the numbers.
2Know what it decides

Most dialers sort answers into two buckets: a person or a machine. This sorts five.

What answered What happens next
A person The agent is connected
Voicemail It waits for the greeting, leaves your message, hangs up
Screening assistant It plays a short reply, then waits
Transfer or hold message It waits for the transfer
Phone menu It hangs up, if you leave that setting on

The fourth row is the one people notice. “Please hold while I transfer you” used to sound like a machine, so the dialer dropped a warm call. Now it waits.

Why the old method struggled with rows three and four: this explains it in twelve questions.

3Prove it on your own list

Dial 50 numbers you can identify in advance: 20 that a person will answer, 20 that go to voicemail, 10 with a menu.

Write down four counts. People connected. Voicemails caught. Voicemails missed. People wrongly dropped.

Read the last one first. If it hangs up on real people, nothing else on the list matters.

Two fixes cover most of it. Dropping real people? Raise Initial silent listen to 4000 ms. Missing voicemails? Raise Voicemail max wait past 20 seconds.

The five numbers

Setting Default What it does
Detect language en Name the language you are calling. auto works, but it checks several languages on every call and costs extra CPU.
Initial silent listen 3000 ms How long it listens before your agent speaks
Voicemail end silence 1000 ms How much silence means the greeting finished
Voicemail max wait 20000 ms The longest greeting it will sit through
ASR workers 2 How many calls it can read at the same time

Two checkboxes sit underneath. On timeout leave voicemail decides what happens when it cannot tell: leave a message or hang up. Hangup IVR line automatically ends calls that land in a phone menu. Both are on by default, and both are right for outbound sales.

One more field worth knowing: Screening reply WAV. Google’s call screening asks who is calling and why, and this is the four-to-six-second recording it hears. Keep it honest and keep it short.

What it cannot do

  • It waits about three seconds. It cannot read speech it has not heard yet.
  • One-word answers are hard. A bare “Hello?” gives it very little to work with.
  • Phone audio is narrow. 8 kHz and compressed, which limits every model.
  • Accents and rare languages are weaker, especially with the small default model.
  • It cannot fix a bad list. Wrong numbers stay wrong.

It also uses CPU, and the checkbox says so. Each worker reads one call at a time, so two workers on five lines means three calls wait their turn. Start at two, raise it by two, and watch the task manager during your busiest hour rather than a quiet morning.

If calls queue at peak, add workers or dial fewer lines. No setting makes a slow machine fast.

The feature is in both editions. If you already run Ecsow it is one checkbox away, and the trial dials 20 contacts per campaign if you want to test it first. Licences start at $10 a month for one line.

Quick answers

Does it send my audio anywhere?

No. The speech model runs on the machine running Ecsow. Nothing is uploaded, and there is no per-minute charge.

How many calls can it read at once?

The worker setting goes from 1 to 6 in the 32-bit build. Each worker handles one call, and extra calls queue briefly. Start at two.

Does it work in other languages?

Yes. Set the language you are calling, such as en, es or fr. Naming it is faster and more accurate than letting the software guess. The auto option works, but it costs extra CPU.

Which editions have it?

Both Ecsow Auto Dialer and Ecsow Voice Broadcast. Same checkbox, same settings panel.

Voice Broadcast: Setup, Cost and Operating Limits

A working manual: three checks before you send, the legal pre-flight, a ten-step procedure, the six settings that matter, and what to do when calls fail.

Operating manual
Voice Broadcast
Applies to: Ecsow Voice Broadcast edition, Windows.
Scope: sending one recorded message to a list, what it costs, the legal checks before you start, and what to do when it goes wrong.
Not covered: live agent calling, inbound IVR, or anything with a conversation in it.

1. Check that this is the right tool

A broadcast plays one recording to everyone who answers. No agent, no listening, no reply. Everything below follows from that.

Run these three checks before you spend anything. If any answer is no, stop here and use a dialer with agents on it.

JobBroadcast
Appointment reminderyes
Payment reminder before collectionsyes
Delivery or installation windowyes
Outage, closure, schedule changeyes
Any job where someone may ask a questionno
Cold list, or a list you boughtno

The pattern: the person already knows who you are, and the only open question is what they should do next. Cold lists answer at two to eight in a hundred, and the other ninety-two remember your number for the wrong reason.

2. Legal pre-flight

Do this before section 3, not after. These are operating conditions, like voltage. I am not a lawyer, and the rules below are the US ones. India, the Philippines, Nigeria, the UAE and the EU each set their own.

2.1 Consent. A recorded call to a mobile needs prior consent. If the message advertises anything, the consent has to be in writing. Keep the record of how you got it.

2.2 Identification. Every message states who is calling in the opening seconds, and gives a callback number. This is 47 U.S.C. 227(d)(3)(A), and it is also the only way people know it is not a scam.

2.3 Opt-out. Marketing messages offer a keypress opt-out, and you honour it immediately. Keep the list.

2.4 Calling window. 8am to 9pm at the recipient’s local time. Not your time.

2.5 Do Not Call. Load the registry numbers you are required to suppress into the campaign’s own DNC list. The software will respect a list you give it. It does not query a national registry for you.

2.6 The customer exemption is narrower than you think. “They bought from us” does not cover a recorded marketing call to a landline.

2.7 The penalty. Under 47 U.S.C. 227(b)(3) it is $500 per call, up to $1,500 if a court finds it was deliberate, and every call counts separately. Five thousand messages without consent is $2.5 million on paper. Cases settle lower. “We bought the list” has never been a defence.

2.8 Synthetic voice is not a loophole. The FCC treats AI-generated voices as artificial voices, so a cloned voice is the same rule with a worse paper trail.

If you cannot tick 2.1 through 2.5 for a list, do not load that list. A broadcast sends five thousand mistakes before you hear the first one.

3. Procedure: send a campaign

3.1 Install the broadcast edition. Voice Broadcast is a separate download from the dialer. Both can sit on the same machine.

3.2 Point it at your SIP trunk. Enter the server, username, password and the caller ID you want to display.

Expected: the line registers, and a test call to your own mobile rings with the right number on screen.

3.3 Import the list. CSV or Excel. Format the phone column as text first, or every leading zero disappears. Remove duplicates before importing, not after.

3.4 Record or load the audio. Ecsow records from a microphone, or takes a WAV or MP3. There is no text to speech, so each language needs its own recording and its own speaker. Have the person who actually answers your phones do it. A headset in a quiet room beats a studio voice that sounds like an advert.

3.5 Write to length. Thirty seconds is about 75 words. That buys four things and no more: who you are, why you are calling, what to do next, and the number to call.

Example, about 25 seconds
Hello, this is Northside Dental calling about your appointment on Tuesday the fourteenth at three fifteen. If that still works, you do not need to do anything. To move it, call us on 555 0134. That is 555 0134. To stop these reminders, press nine.

Three things that script does right, and each one is a rule. It names the caller in the first three seconds. It says what to do if nothing is wrong. And it reads the number twice, because nobody writes down a number read once at speed.

3.6 Set the campaign parameters. Section 4 lists the ones that change results. Start with the recommended values there and change one at a time.

3.7 Turn on transfer or callback if you want replies. Campaign Settings, then Call Transfer. Enter the number that should ring when someone presses the key. Test it from a mobile before the campaign runs, because a transfer to a number your carrier cannot reach fails silently.

3.8 Send it to yourself first. Then to twenty numbers you control. Listen to the whole thing on a real phone line, not on the machine that recorded it. This is the step everybody skips and everybody regrets.

3.9 Start the campaign. Lines go off-hook in sequence. Watch the first fifty attempts before you walk away.

3.10 Export the results. Reporting is a file, not a live dashboard. Decide what you are measuring before you send: completed plays, keypresses, callbacks, and the numbers that failed.

4. Parameters that change results

Only six settings matter for most campaigns. Change one at a time, or you will not know which one moved the number.

Setting Effect Start at
Concurrent lines throughput, and nothing else 10, if the list is under 20,000
Retry attempts reach, at the cost of repeat calls 2
Calling window who is awake when you call 9am to 8pm local
Answering machine detection whether machines get the message too on, leave message on
VAD sensitivity how fast it decides a machine answered default, raise only if you drop real people
Campaign DNC list who is never dialled load it before the first send

5. Cost per campaign

Three lines of arithmetic, and the first one does most of the work.

Minutes = contacts x script seconds / 60.
Carrier cost = minutes x your per-minute rate.
Software = lines x monthly price, regardless of volume.

The carrier rate is the variable. At half a cent a minute, which is the cheap end of US domestic SIP, the message length alone does this to 10,000 contacts:

Script length Minutes Carrier cost
15 seconds 2,500 $12.50
30 seconds 5,000 $25.00
45 seconds 7,500 $37.50
60 seconds 10,000 $50.00

So a monthly programme of 20,000 messages at 30 seconds, on ten lines: $50 of carrier minutes, $50 of software, about $10 of numbers. $110 a month, roughly half a cent a message.

Compare that with reaching the same thousand people by voice, and the difference is not the dialer. It is the wages.

1,000 messages 1,000 conversations
Carrier minutes $2.50 $15
Software and numbers $3 $7
Agent wages $0 $400 to $1,800
Total $5.50 $422 to $1,822
Can answer a question no yes

Break-even. Cost of one avoided no-show divided by cost per message. A $40 no-show against half a cent a message breaks even at about 7,300 messages. If your numbers are better than that, a broadcast pays. If they are much worse, the list is the problem, not the tool.

6. Throughput

Hours = contacts x 40 seconds / 3600 / lines. Forty seconds covers ring, message and hang-up for a 30-second script.

That puts 10,000 messages at 111 hours on one line, 11 hours on ten, and three and a half hours on thirty-two. Lines are the only lever that matters here, and they are shared: twenty lines running a broadcast leaves twelve for your agents.

7. Operating limits

Maximum concurrent lines 32
Platform Windows desktop, one machine. The machine must stay awake.
Audio input Microphone recording, WAV or MP3 upload
Text to speech None. Each language is a separate recording.
Keypress handling One level: transfer to a number, or queue a callback. Not a multi-level menu tree.
Do Not Call Loads a list you supply. Does not query a national registry.
Reporting Exportable file. No live dashboard, no automatic A/B testing.
Support Email, usually answered within a day. Nobody is awake at 8pm for a bad list.

8. Fault finding

Symptom Probable cause Action
Every call fails immediately SIP registration rejected Re-enter credentials, check the port and the caller ID your provider allows
Calls connect but nobody hears anything Audio format the carrier will not carry Re-export the recording as a standard WAV and test one number
First words are cut off Script starts at full volume with no lead-in Add half a second of silence at the head of the file
Answer rate falls over a week Caller ID flagged by carriers Rotate numbers, keep each under about 100 calls a day, and clean the list
Press-1 never reaches anyone Transfer number unset, or unreachable from your carrier Test the transfer destination from a mobile before the next send
Campaign stops halfway Machine slept, or the network dropped Disable sleep, keep it plugged in, and restart from the failure list

9. Pilot before you commit

Send to fifty of your own customers first. It costs about a quarter and it answers the only question that matters: did anyone do what the message asked.

Play the recording to somebody who did not write it. If they ask who it was from, your first three seconds are broken. If they ask what they were supposed to do, the middle is.

The Voice Broadcast trial stops at 20 contacts per campaign, which is enough to hear your own voice on a real phone line. Licences start at $10 a month for one line, $50 for ten, $100 for thirty-two. No contract, 15-day refund.

If you have not chosen a carrier yet, the per-minute arithmetic is here. For the live-calling half of the same job, this explains why a conversation costs forty cents and a message costs half of one.

10. Questions from the field

How much does voice broadcast cost?

About half a cent per message. 20,000 thirty-second messages a month comes to roughly $110: $50 of carrier minutes, $50 for ten lines of software, and $10 for numbers. A longer recording costs more, because you pay per minute.

Is voice broadcast legal?

Yes, with rules. In the US, a recorded call to a mobile needs consent first, and marketing messages need written consent. Every message must say who is calling, and calls are limited to 8am to 9pm local time. The penalty is $500 per call, up to $1,500 if it was deliberate.

Can a broadcast transfer the call to a person?

Yes, with press-1 transfer. The recording asks the person to press a key and the call goes to a number you choose. Keypress callback works the same way and puts the number in a queue for an agent. Without one of those, a broadcast can only leave a message.

How long does it take to send 10,000 messages?

About three and a half hours on 32 lines, eleven hours on ten, and nearly five days on one line. Speed comes from the number of lines, not from the software.

Talk Time, Contact Rate and Cost Per Conversation

The same five-agent team costs 0.4 cents or 18 cents per conversation depending on the dialer. Where the hour goes, and why your list sets the ceiling.

Five agents. Five hours of dialing a day. Twenty-two days a month. That is 550 hours of human attention you are paying for.

Now ask how many of those hours turned into a conversation with somebody who could buy. Most owners answer with total calls, which is not the same number.

I sell dialer software, so I have this conversation a lot. The owner knows his licence price to the dollar and his carrier rate to four decimal places. What he does not know is what one conversation costs him.

That is the only number that connects the software invoice to the revenue it is supposed to produce. Here is the arithmetic.

The short version

  • An agent who is paid for 60 minutes gives you about 30 minutes of talking.
  • Five agents, 550 hours a month, is roughly 5,500 real conversations.
  • Software costs 0.4 to 18 cents per conversation, depending on the dialer.
  • The list matters more than the dialer. A bad list needs seven times the dials for the same result.
  • Five agents should not run predictive dialing at 4:1. The US 3% abandonment rule makes it hard to hold.
  • Measure four numbers weekly: talk minutes per hour, contact rate, conversations per agent per day, and cost per conversation.

Three terms, in plain English

Talk time is the minutes an agent spends speaking with a human. Not logged-in time. Not the time the phone rings. Eight hours at a desk, three hours talking, is three hours of talk time.

Contact rate, also called connect rate or answer rate, is how many dials reach a live person. Dial a hundred numbers, get six humans, that is 6%. Voicemail and ring-no-answer do not count.

Cost per conversation is your total phone spend — software, minutes, numbers — divided by real conversations. Not leads. Not dials. Conversations.

That third number is the only way to compare a $10 dialer with a $159 dialer honestly. Everything else is a unit somebody picked to make their own price look small.

Where the hour goes

One agent, one logged-in hour

A consumer campaign, measured across a normal week.

The hour60 min
Talking — 30 min
Notes — 6 min
Waiting — 14 min
Breaks, coaching — 10 min

Half the hour is the product. The other half is what the job requires to deliver it. Wrap-up and waiting are the two you can actually move.

This has one consequence. If an agent costs $4 an hour and talks half of it, every hour of conversation costs $8 in wages. At $18 an hour, it costs $36.

That is why “we save you $100 a seat” is a weak claim by itself. $100 against $36 an hour of agent time is less than three hours. The software price only matters in the context of how many conversations that agent time produces.

Dials to conversations

Contact rates in cold outbound are brutal. The gap between a good list and a bad one is wider than the gap between any two dialers.

Published benchmarks put cold B2B at 2% to 5.4% of dials. Verified mobiles reach 9% to 13%. Warm lists run 20% to 30%. Those are somebody’s samples, so treat them as ranges and measure your own.

Here is what they mean for our five-agent team, which needs 5,500 conversations a month.

Dials needed for 5,500 conversations

The same target at five different contact rates.

3% — stale list183,000
5.4% — average cold B2B102,000
8% — decent consumer list69,000
13.3% — verified mobiles41,000
22% — warm list25,000

The bad list needs seven times the dials for exactly the same result. That is the argument for spending money on data before software.

The 3% row is not just expensive. It is impossible. 183,000 dials across 550 agent-hours is 333 dials per hour, or one every eleven seconds. Nobody places a one-at-a-time call that fast, because the ring alone takes twenty seconds. You would have to dial several numbers at once, and that runs into the 3% rule below. The list sets your ceiling.

Those extra 158,000 dials are also nearly free in carrier terms. Most providers bill from the moment the call is answered, so a number that rings out costs nothing in minutes. What a bad list costs you is agent time, number reputation, and the price of a better list.

What a conversation costs

Same team: 550 agent-hours, 16,500 talk minutes, 5,500 conversations at three minutes each. Carrier minutes come to $82.50 at half a cent, plus $3 for numbers. That part is identical on any software.

So here is the software side.

Software cost per conversation

Five agents, 5,500 conversations a month, October 2026 prices.

Ecsow — 3 lines0.4¢
Self-hosted VICIdial — $400.7¢
Hosted VICIdial — $60/seat5.5¢
Convoso — about $90/seat8.2¢
BatchDialer — $95/seat8.6¢
Mojo — $99/seat9.0¢
CallTools — $101.99/seat9.3¢
Five9 Core — $159/seat14.5¢
ReadyMode — $199/seat18.1¢

List prices times five agents. Ecsow uses three lines, because five agents need about three calls at once. Add the same $85.50 of carrier and numbers to every row and the gap shrinks from 45× to about 8× against Five9 and 10× against ReadyMode.

Two things worth noticing.

First, the $85.50 does a lot of work. On a US campaign at half a cent a minute, software is a fifth of the bill at the cheap end and nine tenths at the expensive end. On a Philippines campaign at thirty cents, the carrier bill for 16,500 minutes is $4,950 and every bar above becomes a rounding error. That is the point I made in the SIP pricing post.

Second, 0.4¢ and 18.1¢ are both small numbers. That is not an argument for the cheap one. It is an argument that if your offer converts, software price hardly matters. If it does not convert, no dialer saves you. Price matters most in the first six months, when you are still finding out whether the offer works. That is exactly where most Ecsow customers are.

Why five agents should not run 4:1

This part costs me sales, so pay attention.

In the US, the FCC rule 47 CFR 64.1200(a)(7) caps abandoned calls at 3% of calls answered by a person, measured over 30 days per campaign. A call is abandoned if it is not connected to a live agent within two seconds of the greeting. The same section requires a recorded identification message and an opt-out on any abandoned call, and forbids hanging up before 15 seconds or four rings.

Two seconds is the whole window a predictive dialer has to find a free agent. To do that at 4:1 you need enough agents that somebody is almost always free. The usual guidance is eight or more, with careful pacing and real supervision. With five agents, one bad afternoon pushes you over 3%, and the 3% is measured across the whole month.

So at five seats, run one call at a time per agent. It is slower per agent, and it is the only setting that is comfortably legal and comfortable for the person answering. If you have thirty agents, buy a real predictive engine and hire someone to tune it. If you have five, buying prediction you cannot legally use is how you pay $159 a seat for a dialer set to 1:1.

I am not a lawyer, and this is the US rule. India, the Philippines, Nigeria, the UAE and the EU each have their own, and several are stricter about consent and calling hours. Check yours before you scale.

What actually moves the needle

  1. The list. A verified mobile list connects at roughly three times the rate of a switchboard list. That beats every dialer feature combined, and costs less than a month of expensive software.
  2. Caller ID health. First Orion’s published research puts answer rates near 52% on a clean number and near 11% on one labelled “Potential Spam”. A 40-point swing is bigger than the gap between the best and worst list above.
  3. Time of day. Late afternoon beats late morning. Midweek beats Friday. Everyone agrees on the direction and disagrees on the exact hours, so test it on your own logs.
  4. Number rotation. Carriers start flagging at about a hundred calls per number per day. Five agents on one number burn it in a week. Five agents on four numbers do not.
  5. The dialer. Last, and mostly for labour. It can remove the dialing, the waiting and the note-taking from the hour. That is worth 10 to 20 minutes of talk time. Not a new business. If you want to see where those minutes leak away, how a dialer tells a human from a voicemail is the most common place.

Where Ecsow is worse

  • No workforce management. No forecasting, scheduling or adherence. You run shifts in a spreadsheet.
  • Reporting is an export, not a wallboard. Call records and reports in Excel, plus a live campaign view. No thirty-metric supervisor dashboard.
  • Pacing is manual. No adaptive engine learning your answer rates. Somebody has to look and change a setting.
  • No list cleaning or DNC scrubbing. You bring clean data, or you get the connect rate your data deserves.
  • Windows only, outbound only, 32 lines. No browser, no inbound, no blended campaigns.
  • Email support, about a day. If a campaign breaks on Saturday, you read the manual.

And the one that matters most: a cheap dialer lets you lose money more slowly, not faster. If you will not open the report once a week, the software price is the least of your problems.

Four numbers for the wall

Talk minutes per agent-hour. Below 25 means a pacing or list problem. Above 40 on cold data means somebody is dialing too hard.

Contact rate. Dials against conversations with a human. Split it by list source, not by campaign. The source is where the answer lives.

Conversations per agent per day. Your agents feel this one, and it moves first when a list goes stale.

Total phone spend divided by conversations. Software, carrier and numbers, once a month. When it drifts, you will know which part moved.

For the software half of that last number, Ecsow starts at $10 a month for one line and $22 for three, with no per-minute markup and a 15-day refund. Download it and run a campaign of twenty contacts on your own list. That is enough to see your real contact rate.

If you are still comparing, the published price list and the cheapest-options ladder cover the software side, and the VICIdial piece covers the free option without flattering it.

Questions

What is a good talk time for outbound agents?

On a dialer-assisted consumer campaign, 25 to 40 minutes of real conversation per logged-in hour is a workable band. Below 25 usually means a list or pacing problem. Sustained above 40 on cold data usually means pacing aggressive enough to create abandoned calls.

What is a good contact rate for cold calling?

Cold B2B lists land at 2% to 6% of dials. Verified mobiles reach 9% to 13%. Warm lists run 20% to 30%. Measure by list source, because two sources in one campaign often differ more than two industries.

How much does one outbound conversation cost?

At five agents producing 5,500 conversations a month, software runs from about 0.4 cents per conversation on a per-line dialer to about 18 cents at $199 a seat. Add carrier minutes, identical on both, and the all-in figures are roughly 2 cents and 20 cents.

Can a predictive dialer work with five agents?

It can run, but it cannot legally dial aggressively. The US rule caps abandoned calls at 3% over 30 days, with a two-second window to connect an agent, and that is hard to hold below about eight agents. At five seats, one call at a time is the sane setting.

What SIP Providers Actually Charge in 2026

The dialer is the cheap half of the bill. What 17 SIP providers charge per minute, what 20,000 minutes costs to Asia, and what to ask.

Most people shopping for a dialer compare the licence price, pick the cheapest one, and stop. That works right up until the first carrier invoice arrives.

Here is what nobody puts on a pricing page. The dialer is the cheap half of your phone bill. The expensive half is the company that carries the calls, and it is usually five to fifty times bigger.

Every rate below is published by the provider itself. Checked October 2026.

The short version

  • Your dialer is the cheap half of the bill. Your SIP provider is the expensive half.
  • US domestic minutes run $0.005 to $0.014. That is the cheapest traffic in the world.
  • 20,000 minutes to the Philippines costs about $6,000. The dialer that made those calls costs $30.
  • Ecsow charges $10 to $100 a month by line, and never per minute.
  • The per-minute rate you negotiate matters more than the software you choose.
  • Ask for the rate to your three main destinations, in writing, before you sign.

What a SIP provider is

A SIP provider, also called a SIP trunk provider or just your VoIP carrier, is the company that carries your calls over the internet.

Ecsow is the software that dials the numbers and puts agents on the line. Your provider connects the call and bills you for the minutes. You buy them from two different companies, and both send a bill.

Ecsow charges a flat fee per line: $10 for one, $30 for five, $50 for ten, $100 for thirty-two. It never charges per minute.

Your carrier charges for every minute of every call, priced by country. That second number decides whether your operation makes money.

The dialer stays flat. Minutes grow.

One team, four call volumes

Ecsow licence and carrier minutes side by side, at $0.005 a minute.

5,000 min$55
15,000 min$105
50,000 min$300
100,000 min$550
Ecsow licence
Carrier minutes

The blue bar barely moves. You pay for lines, not minutes. Some carriers also charge per channel, which is one of the questions below.

So shop accordingly. If you dial a few thousand minutes a month, software price matters. Past about 50,000 minutes it is noise, and the only number that moves your business is your per-minute rate.

What providers charge

These are the advertised starting rates of the seventeen SIP providers Ecsow works with. Almost all are US domestic rates, and I will explain why that caveat matters in a moment.

Cost of 20,000 outbound minutes

Using each provider’s advertised starting rate.

Amazon Chime — $0.0049/min$98
Telnyx — $0.005/min$100
VoIP.ms — $0.005/min$100
QuestBlue — $0.0055/min$110
DirectCall — $0.006/min$120
Flowroute — $0.0083/min$166
Telxi — $0.01/min$200
Twilio — $0.0140/min$280
LocalPhone — $0.02/min$400
Anveo — $0.02/min$400
illyvoip — $0.093/min$1,860

Same 20,000 minutes. The cheapest carrier here charges a nineteenth of the most expensive one. That gap is not quality. It is packaging, minimums, and who is reselling whose network.

Five more do not publish a per-minute number at all. CallCentric sells monthly plans from $5.95. VoIPLine Global advertises unlimited outbound for $14 a month. RingCentral charges per seat. OnSIP charges per user. Voxter prices by destination. None of those models is wrong, but you cannot compare them without a quote.

The full list, with links and a setup guide for each, is on the SIP providers page. There is a calculator there that puts your own numbers into this.

The part that actually bites

Every rate above is US domestic. Almost nobody reading this calls only the US.

International calls are priced by destination, and the spread is enormous. Here is one carrier’s published table. Yours will differ, sometimes by three times, so check the countries you actually dial.

What 20,000 minutes costs to each destination

One carrier’s published mobile rates. In Asia, landlines often cost more than mobiles.

Bangladesh — $0.06/min$1,200
Brazil — $0.06/min$1,200
Pakistan — $0.08/min$1,600
Mexico — $0.08/min$1,600
India — $0.12/min$2,400
Egypt — $0.20/min$4,000
Philippines — $0.30/min$6,000
Nigeria — $0.30/min$6,000

Twenty thousand minutes into the Philippines costs about $6,000. The dialer that made those calls costs $30. If you remember one thing from this page, remember that ratio.

This is where small outbound teams discover the shape of their business. The software choice saves tens of dollars. The carrier choice saves thousands. I sell the software, and I am telling you the carrier is the bigger decision.

One worked month

Ten agents. Five lines. 15,000 outbound minutes at half a cent.

Ecsow licence, 5 lines $30
Carrier minutes, 15,000 × $0.005 $75
Phone numbers, 5 at about $1 $5
Total per month $110

Now put the same team on a per-seat dialer at $110 an agent. The licence goes from $30 to $1,100. The minutes do not change at all, because you buy those from a carrier either way.

That is the whole argument for per-line pricing. It does not make your calls cheaper. It stops your software bill from growing with your headcount.

Five questions for any SIP provider

  1. What is the exact rate to my three main destinations, in writing? Not the US headline rate. The countries you actually dial.
  2. Is billing in six-second or sixty-second increments? On short calls this changes the invoice more than the rate does.
  3. Do you charge per channel, or only for minutes used? A per-channel fee on top of per-minute billing is how a cheap rate becomes an expensive month.
  4. Which countries can you reach, and which codecs do you support? Coverage varies more than the price list suggests.
  5. Is there a minimum deposit, and does the credit expire? Prepaid balances that expire are a quiet cost.

One more if you dial at volume: ask about caller ID reputation. Carriers start flagging numbers at about a hundred calls per number per day, and a “Scam Likely” label can take two to three months to clear. That is a rotation problem, not a dialer problem, but it lands on your answer rate.

Where Ecsow is worse

  • We do not carry your calls. No network, no minutes, no rates. You bring a provider, which means one more account and one more invoice.
  • We add no per-minute markup, and no per-minute safety net. A carrier with bundled minutes is easier to budget. With us, a bad rate is entirely your rate.
  • No inbound. If you receive calls, this structure is the wrong shape for you.
  • Windows only, outbound only, 32 lines maximum.
  • Email support, answered within about a day. If your carrier goes down at 8pm, neither of us is picking up.

How to use this

Get one written quote from two or three carriers for your exact destinations. Multiply it by twelve. Put it next to the software price. That comparison takes an afternoon and usually settles the question.

If you have not chosen a carrier, the SIP providers page lists all seventeen with rates, and the calculator there does the totals for you.

The software side is easier. Download Ecsow, run a campaign of up to 20 contacts, and see if it dials the way your team needs. Licences start at $10 a month for one line, with no contract and a 15-day refund.

Questions

How much does a SIP provider cost per month?

US domestic rates run about $0.005 to $0.014 a minute, plus roughly $0.50 to $1.15 a month per phone number. International destinations cost far more, usually $0.05 to $0.60 a minute.

Is a SIP provider the same as a dialer?

No, and this is the most common confusion. The provider carries the call and bills per minute. The dialer decides which number to call and connects an agent. You need both, from two companies.

What is the cheapest way to call internationally?

Get the destination rate right first. Twenty thousand minutes to the Philippines costs about $6,000 at $0.30 a minute, against $30 for the dialer. The carrier rate is the cost that matters.

Do I need a phone number for outbound calling?

Usually yes, because the number is what the person sees. Numbers cost about $0.50 to $1.15 a month each. Most teams want several so they can rotate them and protect their reputation.

Five9 Alternatives for Teams Under 50 Seats

The case for buying Five9, argued properly first, then the one page where it falls apart. Which side of the 50-seat floor you are on decides it.

You should probably buy Five9.

I sell a competing product, so that sentence costs me something. I am going to make the case for Five9 as well as I can, including the parts most comparison pages leave out. Then I will show you the one page where it falls apart. You can work out which side of it you are on.

If I do this properly, most readers should end the first half thinking I have talked myself out of a sale.

The case for buying Five9, argued properly

One. They publish their prices. That sounds like a low bar until you try to price this market. Convoso will not tell you. ReadyMode will not tell you. RingCentral shows a number and then hands you to a rep. Five9 puts $159 on a page. A vendor that publishes its price is a vendor that expects to be compared.

Two. Compliance is not a feature, it is an entry ticket. HIPAA with a signed agreement, SOC 2 Type 2, GDPR. If you handle health records or card data, none of that is optional. Building it yourself is a year of somebody’s life and an audit you pay for twice.

Three. The channels share one queue. Voice, chat, email, SMS, social and WhatsApp, routed together, reported together. You can assemble that from four vendors. You will then spend your Fridays reconciling four reports that count things differently.

Four. Workforce management is the actual product. Forecasting, scheduling, adherence, quality scoring, speech analytics. At twelve agents, that is a spreadsheet. At eighty, a three-point improvement in schedule adherence is worth more than the licence fee.

Five. Uptime with money behind it. Geo-redundancy, several regions, a service credit if they miss. If your dialer is down at ten in the morning, you have lost a day of production. At a hundred agents, that is not a support ticket. It is a payroll event.

Six. It will pass your security review. The infosec questionnaire, the data processing agreement, the sub-processor list. Those documents exist. At a company with a procurement department, that alone can save two months.

Seven. Somebody implements it. Partner network, professional services, a project manager. You are not the integrator, and at this size you should not be.

That is the case. I do not think it is wrong. I think it is a case for a different company than the one most people reading this are running.

That case is right, and here is exactly who it is right for

Let me be specific, because vague concessions are worthless.

You are at fifty seats or more. Some real share of your volume is inbound, and your agents blend. Your buyers ask for a BAA before they sign. You have an operations manager whose job is the schedule. A four-hour outage costs you more than the annual licence difference.

If that is you, buying Ecsow would be professional malpractice. Not because the product is bad. It is because you would be swapping a system that manages people for a system that dials numbers, then trying to cover the gap with spreadsheets. Stay. Pay the $159. It is the cheapest part of your operation.

The one assumption the case never states

All seven arguments above assume something none of them mentions.

They assume you can buy it.

Five9’s published plans carry a 50 concurrent-seat minimum. Not fifty users. Fifty seats logged in at the same moment. The smallest purchase they will take is 50 seats of Core at $159. That is $7,950 a month, or $95,400 a year, before implementation. Digital is $119 a seat and has no voice at all; even that floors at $71,400 a year.

Here is the part I find interesting, and it is the reason I do not think the minimum is arrogance. The product’s value sits in workforce management, compliance and channel routing. None of those pay for themselves at twelve seats. A 50-seat floor is not a company being greedy. It is a company telling you, accurately, who the product is for. Selling it to a ten-agent outbound team would be the dishonest move, and they decline to make it.

Which leaves a strange situation. You are not being outbid. You are being told the product is not for sale to you, at any price, by a vendor who is right about that.

Team size Core, per month Can you buy it?
10 seats $1,590 No
25 seats $3,975 No
50 seats $7,950 Yes, this is the floor
50 seats, first year with implementation $120,400 Yes

Where the case is genuinely weak, and where it is not

Only two of the seven arguments deserve an attack. I will skip the other five, because they hold.

The unit fits inbound and misfits outbound. This is the whole thing, so read it slowly. For a blended inbound team, one seat really does mean roughly one conversation, because calls arrive when they arrive. Per-seat is correct there, and I will not pretend otherwise. For an outbound team it is backwards: the dialer exists to create more calls than agents, which is the entire point of predictive dialing. Billing per agent charges you for the thing you deliberately use less of.

That is not a pricing trick. It is two different business models sharing a word.

Implementation is quoted separately, and it is not small. One 2026 pricing guide puts it at $25,000 to $150,000, with professional services adding 15% to 30% a year. At a hundred agents, that disappears into payroll. At twelve agents it is a decade of software budget before the first call goes out.

The remaining four — compliance, channels, WFM, uptime — I concede completely. If you need them, you need them, and nothing at $10 a line will substitute.

Now the case against me

Fairness runs both ways, so here is my own product argued against.

No inbound. None. No IVR, no queue, no blended campaign. If anyone calls you back, Ecsow cannot answer.

No certificates. No BAA, no SOC 2. If procurement asks for either, I fail the review and you should not spend time on the trial.

Windows only. Every remote agent needs a Windows machine. A team on Macs is not a customer.

32 lines, then a hard stop. That covers about a hundred agents and then ends.

Email support, about a day. No SLA, no phone line, nobody awake at 8pm when your list turns out to be wrong.

No CRM, no SMS, no chat, no workforce management. Per-campaign and per-agent reports, exported as a file.

Manual billing. No card on file, so you renew by hand.

If two or more of those are requirements rather than preferences, the case against me wins. Close the tab. I mean that as operating advice, not as a rhetorical flourish.

The rule that settles it

Three questions, in this order. The first two are disqualifiers, so answer them before you look at any price.

Is more than about a fifth of your volume inbound? Then you need a blended platform, and Ecsow is out. So is every per-line dialer.

Will anyone demand a signed BAA or SOC 2 evidence? Then you need a compliance vendor. Out.

Are you under fifty seats and outbound only? Then Five9 is not a decision you get to make, and the real comparison is against the per-seat dialers — CallTools around $102 to $120, Convoso around $90, ReadyMode $199 to $249 — and against hosted VICIdial at $50 to $75 a seat.

That third group is where most readers actually are. Ten agents on CallTools is about $1,020 a month on a twelve-month term, $12,239 a year. Ten agents here need about four concurrent lines, which is $26 a month.

Two products, two units, two business models. The one you want depends on whether your agents are waiting for calls or calls are waiting for your agents.

The trial dials 20 contacts per campaign with no card. Run one campaign on your own list and the argument stops being theoretical. If you want the per-seat arithmetic laid out in full, this is the invoice, and the CallTools week covers the closest per-seat alternative.

Questions this argument always produces

What is the cheapest Five9 alternative?

Among published prices, Ecsow starts at $10 a month per line and does not charge per agent. Self-hosted VICIdial has no licence fee, but needs a server and someone who knows Asterisk.

Does Five9 have a minimum seat requirement?

Yes. Its published pricing lists a 50 concurrent-seat minimum, so smaller teams cannot buy it directly at any published rate.

How much does Five9 cost per month?

$159 per concurrent user for Core, and $119 for Digital, which has no voice. Checked September 2026. AI minutes and SMS are billed on top, and implementation is quoted separately.

Can Ecsow replace Five9?

Only for outbound-only teams. Ecsow has no inbound menu, no channel routing, no compliance certificates and no workforce management. It replaces the dialer and nothing else.

The Cheapest Auto Dialer Software for Small Call Centers

Six receipts for the same job, five agents for a year, with the line each one leaves off. The cheapest licence and the cheapest year are different rows.

Six quotes for the same job.

Five agents, outbound, one year, software only. No carrier minutes in any of them, because no vendor quotes minutes the same way.

Read them side by side and the argument makes itself. Then read what each receipt leaves off, because that is where the cheap ones stop being cheap.

ReadyMode Starter
5 seats · 12 months
5 seats at $19911,940
Setupquoted
Year one11,940
Not on this receipt: minutes, and the annual commitment that produced this rate.
BatchDialer
5 seats · 12 months, monthly billing
5 seats at $1247,440
Setupquoted
Year one7,440
Not on this receipt: the annual rate, which is lower, and the months you cannot cancel.
CallTools
5 seats · 12-month term
5 seats at $101.996,119
Setup250–1,500
Year one6,119+
Not on this receipt: the $18 a seat you saved by giving up the right to leave.
DialSheet + Twilio
5 reps · 150 dials a day
DialSheet, 5 at $291,740
Twilio usage3,900
Year one5,640
DialSheet’s own estimate, about $94 a rep all-in. Not on this receipt: the Twilio account you now administer.
Self-hosted VICIdial
5 seats · one server
Licence0
Server, 12 at $40480
Admin, 72 h at $201,440
Year one1,920
Not on this receipt: the 72 hours. That line is the whole argument.
Ecsow, 3 lines
5 agents · 3 concurrent calls
3 lines at $22264
Setup0
Year one264
Not on this receipt: inbound, SMS, a CRM, a browser version, or phone support. This is a Windows outbound dialer and nothing else.

A 45-fold gap deserves suspicion

The top receipt is 45 times the bottom one. That is not a pricing error on either side, and you should not believe it until it is explained.

Two reasons, and only two. The unit, and the labour.

The unit. Four of these six charge per person who can log in. One charges per call in progress. One charges nothing and bills you in hours. Five agents on an outbound campaign usually keep three calls going, not five. The per-line receipt is buying three units while everyone else sells five.

The labour. VICIdial’s licence is genuinely zero. Its receipt is not, because a server needs somebody. Price those 72 hours at $20 and the free option lands at $1,920. Price them at zero and it lands at $480.

The two receipts that look cheapest and are not

DialSheet. $29 a seat is the most honest number in this market. It is also the most misleading, because the calls are not in it. DialSheet bills your Twilio account directly. Their own published estimate is about $94 a rep once calls are counted, which turns $1,740 into $5,640 without anyone raising a price.

VICIdial. Free software, $1,920 a year. The gap between those two facts is a person, and that person is usually you.

Neither of these is a trick. Both vendors publish what they publish. The trick is in reading a licence fee and calling it a cost.

The one case where free actually wins

If the server already exists — a machine you rent anyway, spare capacity on something else, a box under a desk — the marginal cost of VICIdial really is zero.

Then the receipt changes shape. $0 against $264, and nothing I sell can argue with zero.

That case is narrower than the phrase “VICIdial is free” makes it sound. It is also common enough that I would not talk anyone out of it. But it does require the server, and it does require the person.

What the two cheapest receipts leave out of the product

Price is not the only line. Here is the other side, on my own product first.

Ecsow is Windows only and outbound only. No inbound menu, no SMS, no built-in CRM, no browser version, 32 lines maximum. Email support answered within about a day, and manual monthly renewal because we do not hold your card.

VICIdial has none of those limits. It has inbound, a queue, blending, source code you can edit, and no ceiling worth mentioning. What it does not have is anybody whose job is to keep it running.

The other four sit in between and charge accordingly. ReadyMode and BatchDialer sell you a managed floor. CallTools sells you the best support in the category. DialSheet sells you a cheap seat and hands you the telephony.

Run these six receipts for your own team

It takes an afternoon and it beats every comparison table on the internet, including this one.

Write down two numbers. How many people log in, and how many calls are in progress at your busiest hour. Most teams guess the second one wrong and guess it high.

Ask each vendor one question in writing. What is my total for this exact seat count and this exact concurrency, for twelve months, including setup? The ones who answer with a number are the ones worth talking to.

Add the missing line yourself. Carrier minutes, your admin hours, or the Twilio bill. Whatever the receipt left off, put it back before you compare totals.

Then pick the receipt with the smallest number at the bottom and the fewest things missing from the middle. Those are rarely the same receipt.

Where Ecsow lands, stated plainly

Cheapest published per-line price here. Not the cheapest product, and not the right one for everybody.

$10 a line, $30 for five, $50 for ten. It installs in about ten minutes and connects to the SIP trunk you already pay for. It dials automatically or predictively, and it does not care how many agents log in. No setup fee, no contract, 15-day refund, and the single-line price has not moved since 2017.

Download it and run a campaign — 20 contacts, no card. If a month of real dialing does not beat what you use now, that is what the refund is for.

If the arithmetic points you at VICIdial, that comparison will tell you which side of the line you are on. Two more breakdowns: CallTools at five seats, and why Five9 will not sell to a ten-agent team. The dialer is only half the bill — SIP rates cover the other half.

Questions about these receipts

What is the cheapest auto dialer software?

The cheapest licence is self-hosted VICIdial, which is free but needs a server and someone who knows Asterisk. Among products with a published price, Ecsow starts at $10 a month for one line and does not charge per agent.

Is there a free predictive dialer?

VICIdial is free and does predictive dialing. Running it is not free. Every other free tier in this market is a trial or a single-user plan, not a usable dialer for a team.

Why is per-line cheaper than per-seat?

They charge for different things. Per-seat bills every agent who can log in. Per-line bills how many calls happen at once. A ten-agent team usually runs three to five calls, so the two numbers drift far apart.

What hidden costs should I ask about?

Setup fees, minimum seat counts, contract term, any per-minute charge on top of your carrier, extra numbers and channels, recording storage, and whether you can export your lists and recordings if you leave.

What a CallTools Alternative Costs at Five Seats

The week you decide: the quote, the comparison, the question about inbound, the fee nobody mentions, and the number that settles it on Friday.

You did not ask for a dialer. You asked for a price, and you got a calendar invite.

What follows is the week that usually happens next, written out in order, because the decision is rarely made on features. It gets made on a Tuesday or a Thursday, on one number that nobody put in the deck.

Monday: the quote arrives

The call takes twenty minutes. The number, when it lands, is $119.99 per agent per month. Or $101.99 if you sign for a year.

You have five agents. You do the multiplication twice.

Five seats, month to month, is $599.95 a month. Over a year, $7,199. On the annual term it is $6,119, and the difference is $18 a seat for the privilege of being able to leave.

CallTools does not publish that price on its site. You got it because you got on a call. That is worth noticing, and it is not a criticism of CallTools — most of this market works the same way.

Tuesday: you put our number beside it

Three concurrent lines, $22 a month, $264 a year.

You look at the two figures and assume one of them is wrong. It is a reasonable reaction. The gap is 23 times.

So here is the honest explanation, before you ask. They are not the same product, and the price difference is not 23 times of value.

CallTools gives you inbound calls with a routing menu, and SMS and email from the agent screen. There is a built-in CRM with contact history, free manager seats, and one-click connectors to Salesforce and HubSpot. It runs in a browser, so any machine will do.

Then there is the support, which is the best in this category by a distance. Capterra has it at 4.8 from about 155 reviews. G2 has 4.8 from about 72. Trustpilot is a clean 5.0 from more than 220. People keep saying their issues were fixed in minutes.

Ecsow is a Windows program that dials out. That is the product. $10 a line, up to 32 lines, no contract, no setup fee.

If those sentences read as different businesses rather than different prices, you are reading them correctly.

Wednesday: someone asks the question that decides it

Usually it comes from the person who handles the phones, and usually it is worded badly: “What happens when someone calls us back?”

That is the whole decision, and it has nothing to do with price.

If customers call you back — returns, confirmations, a number on a mailer — you need inbound. CallTools does it. Ecsow does not, at all. No menu, no queue, no blended campaign. Not a limited version. None.

Same answer for three other things. If an agent needs to text a lead straight after a call, CallTools does that and Ecsow cannot. If you want contact history inside the dialer instead of in a spreadsheet, CallTools has a CRM and Ecsow imports a file. If someone on the team uses a Mac, CallTools runs in a browser and Ecsow does not run at all.

Any one of those is enough to end the comparison. That is not a sales line, it is the reason this article has a Wednesday in it.

Thursday: you find the fee nobody mentioned

Setup and onboarding, $250 to $1,500 depending on who you ask and how much you want them to do. Complex CRM integration is reported at $2,000 to $5,000.

Against your first month of $600, that fee is anywhere from two-fifths of a month to two and a half months. It is not hidden exactly. It just is not on the page, and it arrives after the demo, when the decision already feels made.

Ask for it as its own line on the quote, before the demo. It is the cheapest question in this whole week, and the answer changes the arithmetic more than the per-seat price does in year one.

Friday: you decide with three questions

Not a feature matrix. Three questions, in this order.

Do customers call us back? If yes, stay on CallTools and stop here. Everything after this is irrelevant.

Is the annual saving worth the lock-in? $18 a seat a month sounds small. It is the price of being able to walk away in month four. If your volume is seasonal or the business is young, that is the most expensive discount on the table.

Is the difference more or less than one agent? Five seats, CallTools annual, is $6,119 a year. Three lines is $264. The gap is $5,855. If that is less than what you pay one agent, the support and the inbound are probably worth it and you should sign. If it is more, the comparison has already decided for you.

Then the part that is not arithmetic. A $10 product from a company you have not heard of is a real risk to take on. It is paid by PayPal, support is email, and you renew by hand. The way to shrink that risk is not to read more comparison pages. It is to run one campaign.

The trial dials 20 contacts, asks for no card, and installs in about ten minutes. Twenty contacts on your own list will tell you more than this article can.

What the week does not settle

Two things stay open, and you should know them before Friday.

The first is number reputation. Dial hard from one number and carriers start labelling it. The rough threshold is about a hundred calls a number a day. Once a number is tagged, the label can take two to three months to fall off. That happens on CallTools and on Ecsow. Rotating numbers and cleaning the list is the fix, not changing vendor.

The second is that both products still leave you buying minutes. CallTools bundles domestic minutes into the seat price. Ecsow never touches minutes — you rent a SIP trunk and pay a carrier. On US domestic work that is cheap. On a campaign into the Philippines the carrier bill can be twenty times the software, which is a different article.

If you want the wider market on one page, this is the invoice. If the free route is tempting you, this is what it costs in Saturdays.

Questions people ask on the Friday

How much does CallTools cost per agent?

CallTools does not publish prices. Buyers and trackers report about $119.99 per agent per month month to month, or $101.99 on an annual contract, with setup fees of $250 to $1,500.

What is the cheapest CallTools alternative for a small team?

It depends on the unit. Per-seat alternatives start around $29 to $99 an agent. Per-line products like Ecsow start at $10 a month for one line, no matter how many agents log in. That is usually cheaper below ten seats.

Does Ecsow support inbound calls like CallTools?

No. Ecsow is outbound only. No IVR, no queue, no blended inbound. Teams that need blended calling should stay with CallTools or VICIdial.

Can I keep my lists and recordings if I switch?

Yes. Export contacts as CSV or Excel and import them. Format phone columns as text so leading zeros survive. Recordings are ordinary files and move independently.

The VICIdial Alternative for Small Call Centers

Eleven months running a free dialer for a four-seat team: the year in numbers, the Friday it dropped calls, and who should never leave it.

The calls started dropping at 19:40 on a Friday.

Not failing. Dropping — agents mid-sentence, the line going dead, the customer hearing nothing. By the time I opened my laptop, four of six campaigns had stopped and the fifth was dialing into a wall.

I had built this system myself, for a client with four seats. It had run for eleven months without complaint. That evening is the reason I no longer recommend it to teams like that one, and the reason has nothing to do with the software.

Month one: the install felt like a win

VICIdial installs from an ISO. You boot it, answer a few questions, and ninety minutes later you have a working phone system. No licence, no per-seat fee, no sales call.

I remember thinking it was the best deal in the industry. Four seats on a commercial platform would have been $240 a month. This was forty dollars of server and an afternoon.

That arithmetic is real, and it is why the first month feels so good.

Month three: the campaign that took a weekend

A campaign on VICIdial is not one form. It is a list, a set of call results, a user group, a dial level, an inbound group, a script, and a routing plan. Each one is a page of settings, and each one has a default that is wrong for your business.

I got it working in a weekend. The list imported cleanly. The dial level was set to something aggressive because aggressive sounds efficient.

It was not efficient. It burned through the list at a low answer rate and pushed the caller ID hard enough that people started not picking up. I turned the dial level down on Monday and the answer rate recovered over the following week. That was the first lesson, and the software did not teach it to me — the phone did.

Month five: nothing broke, and that was the trap

Four months of quiet. Calls went out, agents worked, recordings piled up on disk.

Piled up is the important phrase. VICIdial records every call by default. Nobody had set a retention rule, because nobody owned the server. That is not a software defect. It is what happens when a system has no owner.

Month eleven: what the Friday actually was

The disk was full. Recordings had filled it, MySQL could not write, and the dialer kept sending calls it could no longer track.

Forty minutes to fix: delete old recordings, add a cron job, restart. Two hours to diagnose, because the error messages point at the database and the cause was three layers down.

We lost most of an evening’s dialing. The money was not the problem. The problem came on Monday, when the client asked whether it would happen again.

I had to answer honestly. It could, because nobody was watching the disk.

That is the sentence I would put on every VICIdial review page. Free software does not remove the operations. It hands them to you.

Month eleven, part two: the year in numbers

I finally added it up. Four seats, one year.

Cost of ownership, four seats
First twelve months, self-hosted
Server, 12 months at $40480
Licence0
Admin, 72 hours at $201,440
Total1,920
Six hours a month at $20. It was closer to nine in the first quarter.

Three lines on a per-line dialer for the same year: $264.

Now the part that keeps me honest, and it is not the part people expect. Even with the admin time counted as free, the box costs $480 a year against $264 for three lines. The server on its own is more than the alternative.

So VICIdial only really wins in one situation: when the server already exists. A machine you rent anyway, spare capacity on something else, a box under a desk. Then the marginal cost is genuinely zero, and nothing I sell can compete with zero.

That is a narrower case than “VICIdial is free” makes it sound, and it is also a common one. Most people who say free mean they already had the server.

What VICIdial is better at, without hedging

Inbound. A queue, an IVR, a blended campaign on the same screen. VICIdial does all of it, and the alternative I sell does none of it. If a real share of your volume comes in, stay.

Depth. Adaptive dialing, a hopper you can actually tune, timezone rules, per-list reporting, an API that will do whatever you can describe. Sixteen years of development shows.

Ownership. The code is open, the database is yours, and no vendor can raise your renewal or disappear. For a fifty-seat operation with an Asterisk person on staff, that combination is close to unbeatable.

Runs anywhere. It is a browser application. The alternative I sell is a Windows desktop program and nothing else.

What moving actually costs

Less than people fear, and in one place more.

The list exports as CSV and imports cleanly. One trap: phone numbers with a leading zero lose it unless the column is formatted as text. Recordings are ordinary files and travel fine. Dispositions and callbacks export too, though old reports will not match exactly, because the two systems count some things differently. Your SIP trunk does not move at all — that is the point of both products.

Scripts get rewritten rather than moved, which is usually an improvement, because the second version is shorter.

The expensive part is trust, and the only cure is time. Run both systems for two weeks on your most boring campaign — not your biggest one. Let the same agents use both. Compare connect rate and talk time per hour, not feature lists. Then decide.

What I would tell someone starting today

If you have an Asterisk person and you need inbound, install VICIdial this week. Then stop reading comparison pages.

If you are three to twenty seats and outbound only, and nobody wants to own a Linux server, here is the honest answer. VICIdial will work, and you will resent it. Not on day one. Somewhere around month five, when a disk fills and the error message points at the wrong thing.

That was my month eleven. I moved the client.

I also set VICIdial up for another client two years later, for free. I still think it is the best free dialer ever written. I still did not want to run it.

Both of those things are true at once, which is why this piece has no tidy ending.

If you want to test the alternative, the trial dials 20 contacts per campaign with no card and installs in about ten minutes. Licences start at $10 a month for one line, with a 15-day refund. It will either make my case or VICIdial’s. For the per-seat arithmetic first, this is the invoice.

Things people ask me afterwards

Is VICIdial really free?

The software is free and open source. Operating it is not. You need a server, someone who knows Asterisk, and time to configure each campaign. Free software and cheap to run are different claims.

How long does a migration take?

One campaign takes about half a day to configure and two weeks to trust. Lists, recordings and dispositions move cleanly. Historical reports do not, because the systems count some metrics differently.

Will I lose my recordings?

No. They are ordinary files on your own server. Download them before you switch the old system off and match them by phone number or lead ID.

Can I run both at once?

Yes, and it is what I would do. Keep VICIdial running while you test one campaign on the alternative. Two weeks side by side costs nothing and removes the guesswork.