AMD for solar companies: settings, drop rates, and compliance

Solar call centers are running some of the highest call volumes in the outbound industry — and doing it under some of the strictest regulatory scrutiny. The FTC has the solar sector in its sights, state attorneys general are active on TCPA enforcement, and the class-action plaintiff’s bar has found solar companies to be reliable targets. In this environment, the answering machine detection settings you chose three years ago are an active liability.

Why solar call centers have unique AMD challenges

Solar’s call lists skew heavily toward homeowners — people who own their home and have sufficient roof space and sun exposure. That demographic profile has three characteristics that make AMD harder than a generic consumer list.

Long, personal voicemail greetings. Homeowners, particularly in the 45+ age bracket that owns most solar-addressable homes, often have voicemails that run long. A heuristic AMD that measures silence gaps and greeting length to detect machines will work fine on these — but it will also be tuned conservatively enough that it starts dropping humans who answer at a normal pace. The AMD can’t tell the difference between the start of a long voicemail greeting and a human who says “Hello, this is [name], how can I help you?”

Near-total mobile penetration. Solar call lists are almost entirely mobile numbers. Mobiles answer differently than landlines — there is more ambient noise, more codec compression, more variation in how quickly someone speaks after the pickup click. Heuristic AMD was tuned on landline audio patterns and has never caught up. The false positive rate on mobile-heavy lists (calling a live human a machine) runs significantly higher than on landline lists.

Volume amplifies every error. A mid-size solar call center running 50 agents on a predictive dialer might generate 30,000–50,000 AMD decisions a day. At 85% accuracy, that’s 4,500–7,500 misclassifications per day. Even if only a quarter of those are live humans dropped as machines, you’re losing 1,000+ prospects daily to AMD error alone — before any other efficiency problem is counted.

TCPA exposure specific to solar

The FTC’s 3% abandoned call rule is the most operationally significant compliance requirement for solar dialers. An abandoned call is defined as a call that is answered but not connected to a live agent within two seconds of the consumer’s greeting. The cap is 3% of all answered calls, per campaign, per 30-day period.

Here is where AMD errors create direct compliance risk. If your AMD mis-classifies a live human as a machine and routes the call to voicemail deposit (or just drops it), that call was answered by a live person and was not connected to an agent. Depending on how you count it, it may count as an abandoned call against your 3% cap. How the 3% rule works in practice — and where the hidden risk lives →

Solar companies have settled TCPA class actions for eight-figure amounts. The FTC has specifically targeted solar in DNC enforcement actions. The margin for error is low, and AMD accuracy sits directly at the intersection of where the compliance risk is.

Recommended AMD settings for solar call lists

If you’re using ViciDial’s built-in heuristic AMD, the two most important settings to review are initialSilence and greeting in your amd.conf.

initialSilence controls how long the AMD waits before deciding there is no audio. If a mobile user answers in a noisy environment and the first 500ms is quiet (waiting for the click to register), the AMD may time out and misclassify. Raising this value reduces that error but also slows classification, which increases the chance of a human being left waiting too long and hanging up.

greeting is the maximum allowed greeting length for a human. If a homeowner says “Hello, this is Mike speaking, who is this?” the greeting may exceed the default threshold and trip a machine classification. Raising the greeting threshold catches more humans with longer answers but makes the AMD slower to drop actual voicemails.

The fundamental problem is that tuning one direction makes things worse in another. This isn’t a configuration issue you can tune your way out of — it’s the structural ceiling of timing-based detection on mobile audio. The practical range for heuristic AMD on a solar list is 70–85% accuracy, no matter how carefully you tune it.

Caller-ID reputation for solar

Solar companies end up “Spam Likely” faster than most industries, for two reasons. First, call volume per DID is high — carrier analytics platforms flag numbers that generate many calls in a short window. Second, bad AMD creates the ghost abandonment pattern (answer → no agent speaks → prospect hears silence and hangs up) that carriers treat as a strong spam signal.

Once a number is labeled, answer rates drop 30–60%. You rotate to a new DID and start the clock again. Solar companies in heavy dialing phases can burn through DIDs at a rate that makes number hygiene a significant operational cost. How bad AMD wrecks your caller-ID reputation →

The honeypot angle matters especially for solar. Solar lists circulate widely — they’re bought, resold, scraped from permit databases, and exchanged between operations. Spam trap numbers get seeded into those lists intentionally. Calling one flags your DID with every carrier analytics platform it reports to, and a single bad call can accelerate a clean number into “Spam Likely” territory. AMD that can identify honeypot audio signatures protects your DID pool in a way that list cleaning alone cannot, because you can’t always tell which numbers are traps before you dial.

The case for AI AMD over heuristic AMD in solar

The argument for AI AMD is simple in most industries. In solar, the math is especially clear.

Higher mobile penetration means the heuristic accuracy floor is lower than average — closer to 70% than 85% on a typical solar list. Stricter compliance requirements mean the cost of each false-positive error is higher than average — every dropped live human is a potential abandoned call against your 3% cap. Higher call volume means the absolute number of errors per day is larger. All three of these factors compound in the same direction.

AI AMD classifies by acoustic signature rather than timing rules. It hears the difference between a human voice pattern and a voicemail recording even when the human speaks quickly, quietly, or in a noisy environment. On AMDY’s production network, accuracy runs at 99% — compared to the 70–85% ceiling of heuristic AMD — and detection starts in 1/8 of a second, before an agent is connected.

What the accuracy gap means in dollars for a solar operation

Take a dialer generating 40,000 AMD decisions per day. At 80% accuracy, you have 8,000 errors. If half are live humans incorrectly dropped (a conservative estimate for mobile-heavy solar lists), that’s 4,000 lost prospects per day — people who answered the phone, heard silence, and hung up. At 99% accuracy, that drops to 400. The difference is 3,600 prospect conversations per day that either happen or don’t, depending on your AMD.

Solar appointment-setting conversion rates from live connects vary by list quality, but even a conservative 5% contact-to-appointment rate means 180 additional appointments per day from accurate AMD. At an average solar deal value well above $20,000 and a reasonable appointment-to-close rate, the revenue difference is not a rounding error.

The compliance value is harder to put a dollar figure on, but settlements in the $10M–$50M range have become common enough in solar TCPA litigation that reducing abandoned-call exposure is worth pricing into your operations stack.