What to Demand From Your Outsourced Dialer Partner: An AMD & Tech Stack Checklist for BPO Contracts

The Question Most Outsourcing Contracts Skip

When a US brand outsources outbound calling to a BPO partner, the RFP and vendor evaluation process usually covers the things that are easy to specify: seat count, shift coverage, language capability, security certifications, pricing per hour or per lead. What it almost never covers in real technical depth is the dialer technology stack itself — specifically, how good the BPO's AMD engine actually is, and whether that's even something the client-side team knows to ask about.

That's a gap, because AMD accuracy sits directly between your outsourcing spend and your actual results. You can hire the best agents in the world and still lose 15-20% of your live connects to a dialer that misclassifies real humans as answering machines before an agent ever gets bridged onto the call. If you're the brand paying per-seat or per-hour for an outsourced calling team, that's a cost you're absorbing without ever seeing it broken out on an invoice.

Why This Matters More With Outsourced Operations

When you run your own in-house dialer, you have direct visibility into AMD performance — you can pull recordings, run QA, and tune settings yourself. When calling is outsourced, that visibility usually disappears unless you specifically build it into the contract. A few reasons this gap widens with outsourced partners specifically:

• Legacy tech stacks are common. Many established BPOs are running VICIdial or Asterisk deployments that have been in place for years, often on default or lightly-tuned AMD settings, because nobody on either side has flagged it as a priority to revisit.
• Reporting is aggregated, not granular. Most BPO reporting dashboards show you connect rate and talk time at a campaign level. They rarely break out AMD-specific metrics like false positive rate, because the client side usually hasn't asked for it.
• Incentives aren't always aligned on this specific metric. A BPO paid per-hour has less direct incentive to chase down AMD accuracy than one paid per-successful-outcome — worth understanding which model you're in before you assume this gets optimized automatically.

The Checklist

Here's what to actually put in front of a BPO partner — during vendor selection, and again during quarterly reviews if you're already under contract.

Technology and AMD Specifics

• What dialer platform are you running (VICIdial, Asterisk-based custom build, a commercial platform)?
• What is your current AMD false positive rate, and how recently was it measured?
• Is your AMD engine the platform default, manually tuned, or a third-party AI-based layer?
• Can you provide a sample of 50-100 recent "machine" classified calls for our own QA review?
• How do you handle greeting variety across the specific carriers and regions we're targeting?

Compliance and Calling Practices

• What's your process for TCPA/DNC compliance on the lists we provide, and who owns that liability contractually? (Our outbound calling compliance guide is a useful baseline to compare their answer against.)
• What are your caller ID and number reputation management practices? Are the numbers dialing our campaigns dedicated to us, or shared across their client base?
• What's your maximum attempts-per-lead policy, and is it configurable per campaign?

Quality and Reporting

• What does your QA process actually look like — percentage of calls reviewed, scoring rubric, and how often are agents coached against it? This is worth cross-referencing against a proper call center QA framework rather than taking "we do QA" at face value.
• Can we get campaign-level reporting broken out by connect type (human answered, machine detected, no answer, busy) rather than a single blended connect rate?
• What's your escalation process when a metric — including AMD accuracy — falls below an agreed threshold?

Infrastructure

• What's your SIP trunk redundancy and failover setup? A dropped trunk mid-campaign has the same effect on your pipeline as a bad AMD engine — dead time nobody's tracking.
• Are you running on dedicated infrastructure or shared tenancy with other clients, and how does that affect call quality during peak hours?

A Sample Scoring Rubric

If you want to formalize this into the RFP itself rather than a loose conversation, a simple weighted scorecard works well:

Category	Weight	What "Good" Looks Like
AMD accuracy / false positive rate	25%	Under 5%, third-party verifiable, willing to share sample recordings
Compliance practices	20%	Documented TCPA/DNC process, clear liability terms
QA and coaching process	20%	Documented rubric, regular scoring cadence, coaching evidence
Reporting granularity	15%	Per-connect-type breakdown, not just blended metrics
Infrastructure resilience	10%	Documented failover, dedicated trunk capacity
Number/caller ID reputation practices	10%	Active monitoring, dedicated number pools

Score each prospective partner against this before signing, and revisit it quarterly with existing partners. It turns "the dialer stack" from an assumed detail into a line item you're actually holding vendors accountable for.

Where This Fits Into the Broader Relationship

None of this replaces the fundamentals of a good outsourcing relationship — clear SOPs, aligned incentives, and regular business reviews still matter more than any single technical metric. But AMD accuracy is one of the few items on this list that's both easy to verify objectively (pull recordings, count the misses) and directly tied to a dollar figure you can hold a vendor accountable for. Most other line items in a BPO contract — agent quality, coaching cadence, cultural fit — are judgment calls. AMD false positive rate isn't. It's a number, and it's one either side can measure the same way.

That makes it a good starting point if you're trying to move a vendor relationship from "trust the reporting we're given" to "verify the numbers ourselves" — and it tends to open the door to asking sharper questions about the rest of the stack once it's established as a baseline expectation.

The Practical Fix, Either Direction

If you're the brand: put AMD accuracy in the contract as a measurable, reviewable metric, not an assumption. If you're the BPO: being able to answer these questions with real numbers — not "we use VICIdial, it works fine" — is a genuine competitive differentiator in an increasingly crowded outsourcing market, especially as more sophisticated clients start asking this exact question.

Either way, the underlying fix is the same. amdify.io plugs into VICIdial and Asterisk-based dialer stacks and brings AMD false positive rates down from the 15-25% range typical of default configurations to 1-3% — giving BPOs a concrete, verifiable answer to put in front of clients, and giving brands a specific technical requirement to write into their next outsourcing contract. Learn more at amdify.io.