How to Measure Answering Machine Detection Accuracy in Your ViciDial Call Center: 2026 Audit Guide

Your AMD is hanging up on live customers right now. You just don't have a number for how many.

Legacy silence-based AMD in ViciDial and Asterisk produces false positive rates between 15 and 20 percent. At a 50-agent center running 3,000 live answers per day, that's 600 decision-makers per day hitting dead air and hanging up. That's $120,000 in lost revenue every month. Most operations managers never measure it directly, so the loss runs quietly in the background for months — sometimes years.

This guide walks you through a practical 2026 audit of your AMD dialer accuracy. By the end, you'll know your actual false positive rate, what it's costing you, and what a realistic benchmark looks like.

Why AMD Accuracy Is Hard to Measure Without a Framework

Your dialer doesn't log "hung up on a human." It logs a disposition. When AMD fires incorrectly and drops a live call, that call gets coded as abandoned, no answer, or dropped. Nothing in default ViciDial reporting tells you the AMD made the wrong call.

You have to build the measurement yourself.

There are two error types worth tracking separately:

False positives: AMD identifies a live human as a voicemail. The dialer drops the call. The prospect hears dead air or a click. Your agent never knew the call was live.

False negatives: AMD identifies a voicemail as a live human. An agent gets connected to a voicemail greeting. Less costly, but it wastes agent time.

For outbound call centers in debt collection, insurance, solar, and home services, false positives are the expensive error. Every false positive is a warm lead you paid to dial, paid to connect, and then threw away.

Step 1: Pull Your Baseline Disposition Data

Start in your ViciDial admin panel. Pull a disposition report for the last 30 days, filtered to outbound predictive dialer campaigns.

Look for these patterns:

• High AMD or ANSWERING MACHINE disposition volume relative to SALE, CALLBACK, or CONTACT
• Short call durations under 3 seconds coded as live contacts
• Abandoned call rates above 3 percent, which signals FAS or AMD misfires

If your AMD-coded calls exceed 40 to 50 percent of total answered calls on a campaign where you'd expect a 30 percent voicemail rate, you have a false positive problem. The excess is almost certainly live humans being dropped.

Export this data to a spreadsheet. You'll need it for the ROI calculation in Step 4.

Step 2: Run a Manual Sampling Audit

Disposition data gives you volume. A manual sample gives you ground truth.

Pull 200 to 300 call recordings from the last week. Filter for calls where AMD fired and the call was dropped or dispositioned as answering machine. Listen to each one.

For each call, note:

Did a human say "Hello" or speak before AMD fired?

Was there a voicemail greeting or beep tone?

Was the call duration under 1.5 seconds before drop? That's a strong signal of a premature AMD trigger.

Tally your results. If 15 to 20 percent of your AMD-fired calls show a human voice before the drop, your false positive rate matches the known baseline for legacy silence detection. If it's higher, you may have a configuration problem stacked on top of the accuracy floor.

This sample becomes your audit benchmark. Repeat it quarterly.

Step 3: Calculate Your CPD and FAS Exposure

Call Progress Detection (CPD) accuracy and False Answer Supervision (FAS) are related but distinct problems.

FAS happens when a carrier returns an answer signal before a human actually picks up. Your AMD fires against silence or ringback tone — not a human voice. This inflates your apparent false positive rate beyond what AMD alone would produce.

To isolate FAS exposure:

• Look for calls with answer supervision but zero audio duration before AMD fires
• Check your SIP trunk provider's answer supervision settings
• Compare false positive rates across different carrier routes if you use multiple trunks

FAS compounds AMD accuracy problems. Fixing your AMD engine without addressing FAS will still leave you dropping live calls. A proper audit covers both.

Step 4: Calculate the Revenue Cost of Your Current False Positive Rate

Once you have your false positive percentage from the manual sample, apply this formula:

Daily live answers × false positive rate = daily dropped live calls

Daily dropped live calls × close rate × average deal value = daily revenue loss

Daily revenue loss × 22 working days = monthly revenue loss

Example using standard 50-agent center numbers:

• 3,000 live answers per day × 18% false positive rate = 540 dropped live calls per day
• 540 × 1% close rate × $1,000 deal value = $5,400 per day
• $5,400 × 22 days = $118,800 per month

That's the floor. Higher deal values or a close rate above 1 percent scales the number up fast. AMDY.IO's ROI calculator lets you plug in your own figures to get a center-specific result.

At $1,440,000 annually for a 50-agent center, this isn't a configuration nuisance. It's a revenue line item.

Step 5: Benchmark Your AMD Dialer Against 2026 Accuracy Standards

Legacy silence-based AMD was built around a simple heuristic: if the audio after answer supervision is silent long enough, it's probably a voicemail. That logic held reasonably well in 2005 when voicemail greetings were long and predictable. It doesn't hold in 2026.

Modern voicemail systems answer faster. Mobile voicemail greetings are shorter. Carrier-level intercepts fire before a single ring. Human speech patterns vary by region, age, and context. Silence detection can't distinguish between a human pausing after "Hello?" and a voicemail system queuing a greeting.

The 2026 accuracy benchmark for a production AMD dialer is 99 percent or better. That's the threshold where false positives drop below 1 percent and revenue leakage becomes negligible.

MightyCall, the only major competitor with a published accuracy figure, claims 97 percent. That sounds close — but at 3,000 daily live answers, 3 percent false positives still costs you $30,000 to $40,000 per month. The gap between 97 and 99 percent is not cosmetic.

AMDY.IO reaches 99 percent accuracy by analyzing audio fingerprints with machine learning rather than measuring silence intervals. The engine identifies the acoustic signature of a human voice versus a voicemail greeting in real time, with sub-millisecond latency. No dead air. No ghost calls.

Step 6: Evaluate Integration Cost and Switching Risk

A common objection when auditing AMD performance is that replacing the built-in AMD feels risky or complex. It isn't — if you pick the right tool.

The relevant questions for your audit:

Does the replacement require a platform migration? MightyCall does. AMDY.IO does not.

Does it require new hardware or a new SIP trunk? It shouldn't.

What's the install time? One line of code is the right answer for a ViciDial or Asterisk stack.

AMDY.IO installs with a single line added to your ViciDial or Asterisk configuration. GoAutoDial, FreePBX, Issabel, and 3CX are supported natively. For any other SIP-based dialer, the WebSocket API connects without platform migration.

Compare that to Genesys Cloud CX at $75 to $240 per agent per month, which requires full platform adoption and multi-year commitments. Or Amazon Connect, which demands AWS ecosystem dependency and developer-heavy setup. Neither makes sense when AMD accuracy is the specific problem you're solving.

Step 7: Set Up Ongoing AMD Accuracy Monitoring

A one-time audit is useful. Ongoing monitoring is what protects revenue long-term.

After establishing your baseline, set up a monthly review cadence:

• Pull disposition reports on the first of each month
• Compare AMD-coded call volume against expected voicemail rates for each campaign
• Sample 100 recordings from AMD-fired calls and log false positive counts
• Track abandoned call rate as a proxy for FAS and dead air events
• Compare month-over-month to catch degradation early

If your false positive rate climbs above 2 percent in any month, investigate before it compounds. Check for new carrier routes, changes in campaign target lists, or configuration drift in your AMD settings.

What a Passing Audit Looks Like in 2026

A well-configured AMD dialer in 2026 should show:

• False positive rate below 1 percent on manual sampling
• Abandoned call rate below 3 percent, per TCPA and Ofcom guidelines
• No ghost call complaints or caller ID spam flags from prospects
• AMD-coded dispositions aligned with the expected voicemail rate for your target list
• Agent talk time increasing as a percentage of total dial time

If your current setup doesn't hit these benchmarks, the audit has done its job. You now have a number, a cost, and a clear path to fix it.

Start with your own disposition data today. To see what 99 percent accuracy means against your current false positive rate, run the numbers with AMDY.IO's ROI calculator or start a 14-day free trial with zero setup fees.

Frequently Asked Questions

What is AMD accuracy and how is it measured in a ViciDial call center?

AMD accuracy is the percentage of calls where your answering machine detection correctly identifies whether the answered call is a live human or a voicemail. You measure it by sampling call recordings where AMD fired, then manually reviewing each to determine whether a human voice was present before the drop. A false positive rate above 1 percent means your AMD dialer is underperforming against 2026 standards.

What causes high false positive rates in ViciDial AMD?

Legacy ViciDial AMD uses silence detection to distinguish humans from voicemails — measuring the duration and pattern of silence after answer supervision fires. Modern voicemail systems, mobile carrier intercepts, and regional speech patterns all produce audio that silence detection misreads. The result is a 15 to 20 percent false positive rate on most legacy configurations.

How does False Answer Supervision (FAS) affect AMD accuracy?

FAS occurs when a carrier returns an answer signal before a human actually picks up. Your AMD engine fires against silence or ringback tone rather than a real voice, inflating your false positive count beyond what AMD accuracy alone would produce. Fixing AMD without auditing your SIP trunk for FAS exposure leaves part of the problem unresolved.

What is the revenue impact of a 15 to 20 percent AMD false positive rate?

At a 50-agent center running 3,000 live answers per day, an 18 percent false positive rate drops roughly 540 live calls per day. At a 1 percent close rate and $1,000 average deal value, that's approximately $120,000 in lost revenue per month — $1,440,000 annually. Higher deal values or close rates scale the loss proportionally.

How often should a call center audit its AMD dialer accuracy?

Monthly monitoring is the right cadence for most outbound call centers. Pull disposition reports, compare AMD-coded call volume against expected voicemail rates, and manually sample 100 recordings from AMD-fired calls. A quarterly deep audit with a larger sample catches configuration drift and carrier-level changes before they compound.

Can I replace ViciDial's built-in AMD without migrating to a new platform?

Yes. AMDY.IO installs with a single line added to your ViciDial or Asterisk configuration — no platform migration, no new hardware, no SIP trunk changes required. GoAutoDial, FreePBX, Issabel, and 3CX are also supported natively. Any other SIP-based dialer connects via the WebSocket API.

What AMD accuracy benchmark should a call center target in 2026?

The 2026 production benchmark is 99 percent accuracy or better. At that level, false positives drop below 1 percent and revenue leakage from dropped live calls becomes negligible. Legacy silence-based AMD tops out around 80 to 85 percent. MightyCall, the closest competitor with a published figure, claims 97 percent. AMDY.IO delivers 99 percent through ML audio fingerprint analysis.