Before You Fire Your Lead Vendor, Read Your Dialer Logs
Bad contact rates usually get blamed on the list. Often the list was fine and the answers were thrown away after hello. A one-week log test settles it.
Before You Fire Your Lead Vendor, Read Your Dialer Logs
The firing follows a script. Contact rates drop for a month, the sales manager says the list is garbage, the vendor sends a defensive email with an accreditation badge in the signature, and the account moves to a competitor who sells the same data at five percent less. Six weeks later, contact rates are still soft, and nobody asks the question that would have saved the money: did anyone actually reach fewer humans, or did the dialer reach them and throw them away?
I have been on the operator side of that argument more than once. In the postmortems where we eventually pulled the logs, the vendor was usually innocent. The calls were answered at a normal rate. The humans were there. The system that was supposed to sort humans from machines was misfiring, live voices were being hung up inside two seconds, and the whole thing landed in the report as if the data had been bad. The vendor got fired for a crime committed inside our own stack.
If you are close to firing a lead vendor right now, hold it for one week and run the test below. Either you save the relationship, or you fire them with evidence instead of a hunch. Both outcomes beat the standard script.
How a Good List Gets Mangled After the Answer
A lead is not damaged goods when it arrives. It is a phone number with a person attached. Everything that happens after the ring is your infrastructure, and that is where good lists go to look like bad ones.
The misclassification hole
Default dialer detection, the kind that ships free inside the platform, drops an estimated 10 to 20 percent of live humans as voicemail. That range comes from our own measurements across customer floors, and it matches what operators see when they finally audit. One live person in five or ten answers, says hello, and gets hung up on by software before any human knows they existed.
From the report's point of view, that call never happened as a conversation. It shows as a disposition, a voicemail classification, and a cost. The lead looks worked. The number looks bad. The vendor looks guilty. We covered the direct revenue side of this in the real cost of AMD false positives, and the same dropped calls that cost you sales are busy costing your vendor their reputation with you.
The triple punishment
Every misclassified human costs you three times, and only one of those costs shows up anywhere:
| Punishment | What it costs | Where it shows up |
|---|---|---|
| The lead itself | The price per record you paid the vendor | Buried in the list spend |
| The call attempt | Carrier cost for the dial, plus the ring time | Buried in telecom spend |
| The lost conversation | A real prospect who never reached an agent | Nowhere. Invisible. |
That third line is the one that should bother you. A dropped live answer is the only kind of business failure that produces no entry in any report. If you want the mechanics of measuring what your dialer does to each disposition, the ViciDial AMD audit guide walks through it call by call.
The One-Week Test That Settles It
The test is simple in structure. You need two rates from the same week of dialing, on the vendor's list in question.
Answer rate: of the dials that completed, how many were answered by anything, human or machine. Human-contact rate: of the dials that completed, how many produced a verified conversation with a person.
The gap between those two numbers is where the mystery lives. If phones are being answered at a normal rate and humans are rare in the logs, the humans are being intercepted between the answer and the agent. That is not the list.
What to ask your dialer admin
Three requests, one afternoon:
- Export the week's dispositions with connect times, hangup times, and classification reasons.
- Sample 50 calls classified as voicemail in the first three seconds after connect. If recordings exist, re-listen to ten.
- Compute answered-but-dropped as a share of all answered calls. That is your misclassification rate on your own traffic.
Where the numbers live
Every dialer keeps the raw material for this somewhere. In ViciDial it is the carrier and disposition logs plus the per-call recording archive, joined on the unique ID. In other platforms it goes by names like call detail records, hangup dispositions, or classification events, but the shape is always the same: one row per answered call, when it connected, when it ended, and what the system decided it was. If your platform genuinely cannot export that, that is a finding about your platform, and it belongs in the same conversation as the vendor's fate, because a stack you cannot audit will hand you this same argument again with a different vendor next quarter.
Reading the result
| Signal from the week of logs | What it points to | Action |
|---|---|---|
| Answer rate normal, human-contact rate far below it | Detection dropping live voices | Fix the detection layer before touching the vendor |
| Answer rate itself low across all vendors | Numbers not picking up at all | Caller ID reputation or carrier quality, not data |
| One vendor's list low on both rates, others normal | That list really is weaker | Vendor conversation, now with evidence |
| Contact rate normal, close rate low | Conversations happen but do not convert | Script, offer, or agent problem |
The top row is by a wide margin the most common outcome in my experience, which is exactly why the vendor-firing script is so durable. The symptom looks identical from the dashboard's altitude.
Rules for the test week
A comparison only counts if the week stays clean. Four rules, none of them optional:
- Change nothing mid-test. No new dispositions, no repaving campaigns, no script edits. Every change you make during the window puts a smudge on the result and buys the vendor an excuse.
- Run the vendor in question at its normal volume. Starving the list during the test produces a sample too thin to mean anything, and it tips them off that something is happening.
- Do not fire anyone yet. Not the vendor, not the media buyer, not the campaign manager. Premature firings during the measurement week contaminate the measurement, and you will never get that week back.
- Write down your prediction before the week starts. Say on paper what you believe the answer is, human-contact rate and all. The point of the exercise is to be able to disprove yourself cheaply, and a written prediction is the only kind you can actually lose.
The discipline matters because the alternative to this test is not neutral. Firing a vendor on a hunch and onboarding a replacement is two to six weeks of soft performance on the new source while your real problem sits untouched in the dialer, still eating one live answer in eight. The test week costs you seven days of patience. The hunch costs you a quarter.
Why the Vendor Gets Blamed Anyway
The blame lands on the vendor because the vendor is the only party whose product visibly entered the equation. Nobody sends you an invoice for dropped humans. The lead vendor sends an invoice every month with the lead count on it, so when conversations fall, the arithmetic in everyone's head is conversations divided by that invoice.
There is also a timing illusion. Detection quality drifts. A dialer software update changes audio paths, a codec renegotiation quietly degrades what the model hears, and misclassification creeps up a few points over a quarter. Meanwhile the list changed too, because lists always change. Two things moved, only one of them has a billing relationship, and the wrong one gets the termination letter. If you have never checked whether your detection drifted, the campaign configuration mistakes piece lists the settings that quietly do this.
The other suspect you should clear first
Before you blame the list or the detection, check one more thing: your caller ID reputation. If your numbers have been flagged as spam likely by the major carriers, your answer rate collapses across every vendor at once, which is the second row of the table above. This is worth clearing because it is cheap to check and it mimics both other failures. We went through the mechanics in how caller ID reputation scoring changes your dialing. The one-week log test already separates it: flagged numbers produce low answer rates on every list, while a detection problem hides under normal answer rates.
The report that ends the argument
Once you run the week of logs, you can walk into the vendor conversation with human-contact rate by list source instead of a vibe. Good vendors welcome that. A data seller who would rather you not measure contact rates on their lists is telling you something too.
The Vendor's Side of the Table
It is worth being fair to lead vendors for a moment, because the innocent-vendor scenario is more common than the guilty one, and firing an innocent vendor has real costs of its own.
A good vendor is running their own quality loop. They source data, they verify numbers, they track complaint and bounce rates by segment, and they replace bad records, usually within a window, if you bring them specifics. None of that machinery can see what happens after the phone rings on your side. From where they sit, they delivered records that dial and connect. From where you sit, nobody converted. Both statements are true, and only your logs can reconcile them.
Firing a vendor also resets things you paid to build:
| What you lose | Why it costs |
|---|---|
| Segment learning | The vendor learned which slices of their data work for your offer. A new vendor starts at zero. |
| Replacement terms | Established relationships usually include record replacement. New ones may not. |
| Baseline data | Your historical contact rates stop being comparable the day you switch sources. |
| Onboarding time | New feeds, new field mappings, new compliance review. Two to six weeks of soft performance. |
None of that means you keep a bad vendor forever. It means the bar for firing should be evidence, and the evidence is one week of logs away. The vendor who survives the audit keeps the account with a cleaner relationship on both sides, because now they know you measure.
How to run the conversation when the logs clear them
If the test says your detection was the problem, tell the vendor that, in those words, before you do anything else. The message that works is short: we audited our own handling, we found we were dropping live answers at X percent, we are fixing it, and we will re-baseline your list's performance over the next two weeks. Any vendor worth keeping will respond well to that, partially because almost nobody sends it. You will get better service out of that one email than out of a year of quiet dissatisfaction, because you have told them the measurement you use, and vendors manage to the measurements their customers actually run.
What the Recovered List Is Worth
Work the example with stated assumptions, then rebuild it with yours.
| Input | Example value |
|---|---|
| Dials per month on the vendor's list | 100,000 |
| Calls answered by a live human | 8,000 |
| Humans dropped by misclassification at 15 percent | 1,200 |
| Humans recovered at 85 percent with proper detection | 1,020 |
| Close rate on a real conversation | 10 percent |
| Revenue per closed deal | $150 |
| Monthly revenue recovered | $15,300 |
Fifteen thousand a month was sitting in a list you were about to cancel. Even at half these inputs the conclusion survives, because the recovery costs a fraction of the list spend. We keep making versions of this point, as in how to recover lost revenue with AI AMD, because the money is real and repeatedly almost-fired over.
What recovery does to your cost per lead
Cost per lead is usually computed as list spend divided by closes, which means every recovered conversation lowers it without buying one new record. Run the same example with a $20,000 monthly list spend.
| Scenario | Closes | Effective CPL |
|---|---|---|
| With 15 percent of humans dropped | 680 | $29.41 |
| After recovery at 85 percent | 782 | $25.58 |
Same vendor, same invoice, same list. The list got 13 percent cheaper the day you stopped hanging up on its people. We made the broader case in why your CPL is wrong, and this is the cleanest version of it: the number you use to judge vendors is partly a measurement of your own answer handling, so measure the handling before you grade the data.
One more angle for the finance-minded: a lead vendor is innocent until your logs say otherwise, but you cannot know innocence without logs. The same call-by-call record that protects the vendor relationship protects you in the other direction, because deliberate abandonment and a detection failure look identical from the outside, and regulators do not grade on intent. The Federal Trade Commission's telemarketing rules cap how many answered calls you may abandon, and "our software misfired" is not a category they recognize. Knowing your true drop rate is cheaper than explaining it later.
How to Run the Test Without Touching Production
You do not rip out your current detection to run this. Modern detection runs as a layer beside the dialer: the same calls, the same trunks, your admin installs it with one command, and you compare its verdicts against your existing dispositions for a week. Nothing on the floor changes during the test. The free tier covers 50,000 detections a month, no card, which is enough for a decisive sample on most floors. Paid plans are monthly and start at $79 a month, so the decision after the test is a month-to-month choice, not a marriage.
Then fire or keep the vendor on purpose. The list might genuinely be bad; that happens, and the same logs will show it cleanly, in the second or third row of the table above. But make the call with the logs open, not with the dashboard's altitude view and a month of frustration. The vendor you almost fired may be the cheapest part of your whole pipeline. The hello you keep throwing away is the expensive part.
If the List Really Is Bad
Sometimes the audit convicts the vendor, and then you should act on it, with the same evidence discipline. Bring the logs to the termination conversation: answer rates by batch, human-contact rates by segment, the dates, the counts. A vendor caught with data that does not dial will usually negotiate replacement records rather than lose the account, and your documented numbers are what makes that conversation short.
Then hold the replacement vendor to the measurement you now run. Tell them on day one that you track human-contact rate by source and that you re-audit quarterly. The good ones will price and perform to it. The ones who go quiet were going to be a problem in month three anyway, and now you found out in week zero, which is the cheapest week to find out.
One warning from the operator files: do not fix the list and the detection in the same month and then argue forever about which fix produced the improvement. Sequence them. Fix detection first, because it is the faster change and it re-rates every list you own. Then judge the data on clean numbers. You will never untangle the two if you move them together, and the vendor relationships you rebuild next year deserve a verdict made on purpose.