AMDY.IO
Start free
← All articles
Product UpdatesAug 22, 2026 10 min read

How to Write an AMD Vendor RFP for Your Call Center: 2026 Template and Evaluation Criteria

If you're evaluating answering machine detection vendors, you already know what's at stake. Legacy silence-based AMD misidentifies roughly 1 in 5 live humans as voicemail. For a 10-agent team, that false positive rate costs an estimated $24,000 per month — $288,000 per year — in calls that never reached an agent.

An RFP forces vendors to answer the right questions before you commit. This guide gives you a working 2026 template and the evaluation criteria that actually separate good AMD from bad.

Why AMD Vendor Selection Deserves a Formal RFP Process

Most call centers treat AMD as a configuration detail, not a revenue variable. That's a mistake.

The accuracy gap between legacy silence-based detection (around 80%) and a well-built ML engine (99%) is not a minor tuning difference. It's the difference between routing 4 out of 5 live answers correctly versus nearly all of them. At scale, that gap shows up directly in your contact rate, agent utilization, and revenue per hour dialed.

A formal RFP also protects you from vendors who obscure their accuracy methodology, require platform migrations to unlock AMD features, or won't share pricing until you've sat through a sales call. The criteria below are designed to surface those problems before you sign anything.

Section 1: Company and Product Background

Start by asking vendors to describe exactly what they're selling.

Questions to include:

  • Is your AMD solution a standalone engine or a feature bundled inside a broader contact center platform?
  • What is your core detection methodology: silence-based, machine learning, audio fingerprint analysis, or a hybrid?
  • How long has this specific AMD technology been running in production?
  • What dialer platforms does your solution natively support?

Why this matters: many vendors bundle AMD inside a platform you don't need and can't integrate with your existing ViciDial or Asterisk stack. A purpose-built AMD engine that installs on your current infrastructure is a fundamentally different product from a full contact center suite that happens to include AMD as a feature.

Section 2: Accuracy and Detection Performance

This is the most important section. Don't accept vague claims.

Questions to include:

  • What is your documented AMD accuracy rate, and how do you define accuracy?
  • What is your false positive rate — live humans classified as voicemail?
  • What is your false negative rate — voicemail classified as live?
  • How do you measure and verify accuracy in production environments?
  • How does accuracy hold up across different carrier networks, regional accents, and voicemail greeting lengths?
  • Do you publish accuracy benchmarks or provide access to test data?

What to watch for:

A vendor claiming 97% or 99% accuracy without naming the methodology isn't making a verifiable claim. Ask specifically whether detection uses silence duration thresholds, audio fingerprint analysis, or a trained ML model. Silence-based detection built into Asterisk and ViciDial typically runs around 80%. That 20-point gap is the entire problem you're trying to solve.

Also ask about False Answer Supervision (FAS) handling. FAS events — where a carrier signals answer before a human actually picks up — cause ghost calls and dead air. A vendor who can't explain how they handle FAS isn't ready for high-volume outbound.

Section 3: Integration and Deployment

Your call center already runs a stack. The vendor needs to work with it, not replace it.

Questions to include:

  • What is the integration path for ViciDial, Asterisk, GoAutoDial, FreePBX, and Issabel?
  • Does integration require a platform migration or changes to your existing dialer configuration?
  • What is the estimated deployment time from contract to live calls?
  • Do you support WebSocket API for SIP-based dialers not on your native integration list?
  • What technical overhead does your IT team need to carry?

What acceptable answers look like:

A one-line install with same-day deployment is achievable. If a vendor requires weeks of professional services, custom development, or a full platform migration, that's not a drop-in AMD fix — that's a new infrastructure project. Know the difference before you start.

Section 4: Ghost Call and Compliance Handling

Ghost calls aren't just an annoyance. They trigger spam flags, damage your caller ID reputation, and create TCPA and Ofcom silent call exposure.

Questions to include:

  • How does your solution eliminate ghost calls?
  • Does your AMD engine detect and suppress calls where no live human is confirmed before audio is delivered to an agent?
  • How does your solution support TCPA compliance for outbound predictive dialing?
  • Does your solution address Ofcom silent call regulations for UK-based operations?
  • How do you handle voicemail drop, and can it be configured per campaign?

Ghost call elimination should be a documented feature, not a side effect. If a vendor can't explain the mechanism, push harder.

Section 5: Latency and Real-Time Performance

Dead air kills conversions. The time between a human answering and an agent connecting is a direct revenue variable.

Questions to include:

  • What is your average detection latency from answer to agent connection?
  • How do you define and measure ultra-low latency in your system?
  • What is your uptime SLA, and what happens during an outage?
  • How does your system perform under high call volume — 500 or 1,000 concurrent calls?
  • Do you provide real-time call progress detection, or is there a batch processing delay?

Any perceptible pause after a human says "hello" increases hang-up rates. Ask for latency figures in milliseconds, not vague descriptions.

Section 6: Pricing and Total Cost of Ownership

Many AMD vendors require a sales call before they'll share pricing. For operations managers who need to build a business case quickly, that's a red flag.

Questions to include:

  • What is your pricing model: per agent, per minute, per call, or flat rate?
  • Is pricing publicly available, or does it require a custom quote?
  • Are there setup fees, onboarding fees, or minimum contract lengths?
  • What is included in the base price versus charged as an add-on?
  • Do you offer a free trial, and what does it include?

For context on the market:

MightyCall claims 97% accuracy at $25 to $45 per agent per month but requires full platform adoption — which means abandoning your ViciDial setup. Genesys Cloud CX runs $75 to $240 per agent per month and is built for enterprise orchestration, not AMD-only upgrades. Amazon Connect requires developer-heavy AWS integration. Most vendors in this space won't share a number until you've talked to sales.

A vendor willing to publish pricing and offer a no-setup-fee trial is signaling confidence in their product. That transparency is worth weighting in your evaluation.

Section 7: Reporting and ROI Visibility

You can't manage what you can't measure.

Questions to include:

  • What AMD performance data do you report, and at what granularity?
  • Do you provide false positive rates, false negative rates, and detection latency per campaign?
  • Do you offer an ROI calculator or revenue recovery estimate based on your accuracy improvement?
  • How do you report ghost call events and caller ID health?
  • Can reporting be exported to your existing analytics stack?

An AMD vendor who can't show you the before-and-after on false positive rates is asking you to trust a black box. Demand reporting that ties directly to contact rate and agent talk time.

AMD Vendor Evaluation Scorecard: 2026

Use this framework when comparing responses. Weight each category based on your operation's priorities.

Evaluation Category Weight Key Criteria
Detection Accuracy 30% Documented accuracy rate, methodology, false positive rate
Integration Compatibility 25% Native support for your dialer stack, deployment time
Latency and Performance 15% Detection latency in ms, uptime SLA, concurrent call capacity
Ghost Call and Compliance 15% Ghost call elimination, TCPA/Ofcom handling
Pricing Transparency 10% Published pricing, free trial, no hidden fees
Reporting and ROI Visibility 5% False positive reporting, ROI calculator, exportable data

If compliance exposure is your primary trigger — a recent TCPA warning or a spike in spam flags — move ghost call and compliance handling up significantly in your weighting.

What to Do After You Score Vendors

Once you have written responses and scores, run a live pilot before committing.

A 14-day trial on real call volume tells you more than any RFP response. During the pilot, track:

  • False positive rate on live calls
  • Agent talk time per hour compared to your baseline
  • Ghost call events per day
  • Any caller ID reputation changes

Compare pilot results against your current AMD performance. If your baseline is 80% accuracy and a vendor delivers 99% in a live environment, the math on recovered revenue is straightforward.

AMDY.IO is built specifically for this evaluation. It installs on ViciDial, Asterisk, GoAutoDial, FreePBX, and Issabel with a one-line command — no platform migration, no IT project. The 14-day free trial runs on your actual call volume with zero setup fees, so your pilot data reflects your real operation, not a demo environment.

Frequently Asked Questions

What is the most important AMD vendor evaluation criterion for a ViciDial call center?

Detection accuracy with a named methodology. A vendor claiming high accuracy without specifying whether they use silence-based thresholds or ML audio fingerprint analysis isn't giving you a verifiable number. For ViciDial operations, integration compatibility is the second critical criterion. If the vendor requires a platform migration, it's not a drop-in fix.

How do AMD false positives affect call center revenue?

Every false positive is a live human who never reached an agent. At a 15 to 20 percent false positive rate, a 10-agent outbound team loses an estimated $24,000 per month — $288,000 annually in missed contact opportunities.

What should I ask an AMD vendor about ghost call handling?

Ask specifically how their system prevents calls from connecting to agents before a live human is confirmed. Ghost calls happen when AMD fires before the call is fully established, delivering silence or dead air to a real person. Ask for the mechanism, not just a yes/no answer. Also ask how ghost call events are logged and reported.

Can I add a new AMD engine to ViciDial without replacing my dialer?

Yes. Purpose-built AMD engines like AMDY.IO install directly into ViciDial with a one-line command and don't require a platform change. Avoid vendors who require full platform adoption as a condition of accessing their AMD features.

What latency should I expect from a modern AMD engine?

Detection latency should be low enough that a live human experiences no perceptible dead air after answering. Ask vendors for latency figures in milliseconds and test them under your actual call volume during a pilot. Batch processing or delayed detection is not acceptable in a predictive dialing environment.

How do I evaluate AMD vendor pricing fairly when most require a custom quote?

Request a written pricing breakdown covering per-agent or per-minute costs, setup fees, minimum contract terms, and what's included in the base price. If a vendor won't provide this before a sales call, factor that opacity into your score. Transparent pricing is a signal of product confidence.

What compliance requirements should an AMD vendor address?

For US operations, ask about TCPA compliance for outbound predictive dialing — specifically how the system handles abandoned call rates and silent call prevention. For UK operations, ask about Ofcom silent call regulations. Ghost call elimination is the core mechanism that supports both. Any vendor without a documented answer to this question is a compliance risk.

Running this evaluation properly takes a few hours of preparation. It saves months of dealing with the wrong vendor. Start with accuracy, integration, and ghost call handling. Score every response against the same criteria. Then run a live pilot before you sign.

Start your 14-day free AMD trial at AMDY.IO and use your own call volume as the benchmark.