AMDY.IO
Start free
← All articles
Product UpdatesAug 23, 2026 8 min read

Outbound Dialer Setup: How to Add AI AMD to Any SIP-Based Platform Without Replacing Your Stack

Your outbound dialer is running. Agents are seated, campaigns are loaded, calls are going out. The problem is that a real share of those calls never reach a live person — because your answering machine detection is misfiring, and now iOS 26 call screening is adding a third category your dialer can't even see.

You don't need to tear out your stack to fix this. Here's how to add AI-powered AMD to any SIP-based outbound dialer without touching your dialplan, retraining your agents, or switching platforms.

Why Legacy AMD Breaks Down in 2026

Stock Asterisk AMD works by measuring silence. It listens for a greeting, times the pause, and guesses whether a human or machine picked up. That method tops out at around 75 to 80 percent accuracy even with careful manual tuning — a ceiling that's been confirmed repeatedly in vicidial.org community threads.

That means roughly 15 to 20 percent of calls get misclassified. Live humans get dropped. Voicemails get routed to agents. Ghost calls burn channel capacity. Across a 20- or 50-seat operation running predictive dial, that math compounds fast.

There's also a structural blind spot baked into the silence-based approach: it was built when the only two outcomes were HUMAN or MACHINE. In 2026, there's a third. iOS 26 Call Screening, Google Call Screen, Samsung Bixby, T-Mobile Scam Shield, AT&T ActiveArmor, and Verizon Call Filter all intercept calls before a human voice ever appears. Legacy AMD can't classify these interceptions — it either drops the call or routes it wrong, and either way you lose the contact.

What AI-Powered AMD Actually Does Differently

Instead of timing silence, a machine learning AMD engine analyzes the audio fingerprint of the call in real time. It reads acoustic patterns — not pauses — to classify what it's hearing.

AMDY.IO uses this approach to produce three distinct classifications:

  • HUMAN — a live person answered
  • MACHINE — voicemail or an automated system answered
  • CALLGUARD — the call was intercepted by a screening system before a human could respond

That third classification is what changes the operational picture in 2026. When a call comes back as CALLGUARD, your system knows it wasn't ignored and it wasn't voicemail — it was screened. You can retry on a different number, reroute to another channel, or leave a targeted message instead of writing off the contact entirely.

CALLGUARD is live in production today and covers iOS 26 Call Screening, Google Call Screen, Samsung Bixby, T-Mobile Scam Shield, AT&T ActiveArmor, and Verizon Call Filter.

How to Add AI AMD Without Replacing Your Dialer

The integration path depends on your platform. The principle is the same across all of them: you're replacing the detection layer only — not the dialplan, not the agent interface, not the campaign logic.

ViciDial

ViciDial's built-in AMD carries the same silence-based limitations as stock Asterisk. Swapping it for an ML engine requires nothing beyond a one-line install. Your campaigns, agent screens, and disposition codes stay exactly as they are.

Asterisk and FreePBX

Asterisk's AMD() application is the most widely deployed silence detector in open-source telephony — and the most discussed on forums when it breaks. The AMDY.IO integration drops into your Asterisk or FreePBX installation with a single command, replacing the detection call without touching your dialplan logic.

Issabel and GoAutoDial

Both platforms are Asterisk-based, so the same install path applies. Any dialplan customizations you've made stay untouched.

3CX

3CX has its own architecture, but it supports external AMD integration through a compatible connection point. The one-line install applies here as well.

Any Other SIP-Based Dialer

If your platform isn't on the list above, the WebSocket API covers it. Any SIP-based dialer that can make an API call can send audio to the AMD engine and receive a HUMAN, MACHINE, or CALLGUARD classification in real time. This is also the right path for cloud contact center developers who want to embed ML-based detection into a custom dialer without building the model themselves.

What the Integration Does Not Touch

This is worth being specific about. The concern that stops most dialer managers from evaluating a new AMD layer is the fear of breaking something that's already working.

Adding AI AMD through a one-line install or WebSocket API does not change:

  • Your dialplan logic or call routing rules
  • Agent scripts, dispositions, or campaign settings
  • Your SIP trunk configuration or carrier relationships
  • Any compliance-related call recording or logging setup
  • Your existing voicemail drop recordings or retry schedules

The AMD engine sits between the call answer event and the agent connection decision. Everything upstream and downstream of that moment stays the same.

The False Positive Cost Is Not Hypothetical

If your current AMD is running at 75 to 80 percent accuracy, you already have a measurable cost. Every human call classified as MACHINE is a dropped contact. Every voicemail routed to an agent is wasted handle time.

The AMDY.IO ROI calculator lets you input your call volume, agent count, and current AMD accuracy to estimate what that gap is actually worth in recovered contacts. The number varies by operation, so rather than citing a fixed figure, the calculator is the right tool for your specific setup.

Comparing Your Options

The alternative to a drop-in AMD layer is switching to a platform that bundles better detection. That's a legitimate option — but it comes with a different set of trade-offs.

Genesys Cloud CX, Talkdesk, Amazon Connect, NobelBiz OMNI+, CallTools, and Squaretalk all include AMD as part of their proprietary dialer platforms. To access their detection, you adopt their seats, their pricing model, and their contract structure. Genesys Cloud CX runs $75 to $240 per user per month. For an operation with 20 to 50 agents already running ViciDial or Asterisk, that's not a line-item decision.

MightyCall claims 97 percent AMD accuracy at $25 to $45 per user per month based on competitor analysis, but it requires full platform adoption and doesn't support existing ViciDial or Asterisk infrastructure. For a 10-plus agent shop, per-user pricing at that range climbs above a flat monthly rate quickly.

AMDY.IO is $79 per month flat, with unlimited concurrent channels, no per-seat fees, and no setup costs. A 14-day free trial is available at app.amdy.io if you want to test detection accuracy against your actual call traffic before committing.

What to Expect After Installation

Once the AMD layer is active, the changes in your dialer are immediate:

  • Calls classified as HUMAN connect to agents without the silence delay that legacy AMD introduces
  • Calls classified as MACHINE trigger your voicemail drop or skip logic automatically
  • Calls classified as CALLGUARD surface as a separate disposition, giving your campaign manager real data to act on instead of a silent drop

The CALLGUARD classification is especially useful for diagnosing whether a campaign is getting screened at scale. If a significant share of your outbound calls to a specific area code or carrier are coming back as CALLGUARD, that's actionable intelligence about how your caller ID is being treated by carrier-level filters — not just noise in your drop rate.

FAQs

Does adding AI AMD require me to change my dialplan?
No. The integration installs into ViciDial, Asterisk, FreePBX, Issabel, GoAutoDial, and 3CX with a one-line command that replaces the detection layer only. Your dialplan logic, routing rules, and agent configuration are not modified.

What is CALLGUARD and why does it matter for outbound dialers in 2026?
CALLGUARD is a third classification output — separate from HUMAN and MACHINE — that identifies calls intercepted by screening systems before a live person responds. It covers iOS 26 Call Screening, Google Call Screen, Samsung Bixby, T-Mobile Scam Shield, AT&T ActiveArmor, and Verizon Call Filter. Without it, legacy AMD either drops these calls or misroutes them, and you lose the contact with no data to show for it.

How accurate is ML-based AMD compared to stock Asterisk AMD?
Stock Asterisk AMD tops out at 75 to 80 percent accuracy with manual tuning, based on community-confirmed figures from vicidial.org. AMDY.IO's ML audio fingerprint engine delivers 99 percent accuracy — a gap of 19 to 24 percentage points.

Can I use the WebSocket API if my dialer isn't on the supported platform list?
Yes. Any SIP-based dialer can connect to the AMD engine through the WebSocket API. You send audio, receive a HUMAN, MACHINE, or CALLGUARD classification in real time, and handle the result however your dialplan requires.

Is there a per-seat fee or channel limit?
No. Pricing is $79 per month flat with unlimited concurrent channels. No per-seat fees, no setup costs. A 14-day free trial is available.

Will this work if I've customized my ViciDial or Asterisk dialplan?
Yes. The AMD layer integrates at the detection point only. Custom dialplan logic, disposition handling, and campaign configurations are not affected.

How do I know if call screening is already affecting my campaign results?
If you're running legacy AMD and seeing unexplained drops or silence on answered calls, CALLGUARD classification will surface those events as a distinct category. The AMDY.IO ROI calculator can help you estimate the scale of the impact based on your current call volume and AMD accuracy.

Start With What You Have

The most practical path to better outbound performance in 2026 isn't a platform migration. It's replacing the one component that's actively degrading your results: the detection layer.

Your ViciDial installation, your Asterisk dialplan, your FreePBX setup, your agent workflows — none of that needs to change. You add one line, connect to an ML engine that classifies calls at 99 percent accuracy, and gain visibility into a class of screening events that legacy AMD can't see at all.

If you want to see what the accuracy gap looks like against your own call traffic, the 14-day free trial at AMDY.IO requires no setup fee and no changes to your production environment to get started.