The Voicemail Tax on Your Payroll

Watch your agents for one hour. Count how many times somebody gets bridged onto a call, says hello twice, waits, realizes it is a machine, and disconnects. Fifteen to thirty seconds gone, every time. Now multiply by the number of times it happens per agent per day. Then by your headcount. Then by your loaded hourly rate.

Most owners have never done that multiplication. I know because I ran a floor for years and never did it either. Voicemail time shows up nowhere on a report. It does not register as a cost because nobody cuts a check for it. But you pay salaries for those minutes, and those minutes produce nothing. It is a tax on your payroll, levied in dead air, hundreds of times a day.

This article is the calculation, and then what to do about it. No jargon, I promise. Just arithmetic.

The math nobody runs

Here is the whole formula. You need four numbers off your own floor.

Daily voicemail tax = misroutes per agent per day × average seconds per misroute × number of agents × loaded hourly rate ÷ 3,600

The 3,600 converts seconds to hours so the hourly rate applies cleanly. Loaded rate means base wage plus taxes, benefits, and your share of overhead. If you do not know it, take base wage × 1.3. That is close enough to make the point.

A misroute, to be precise, is any call where the dialer's answering machine detection hands a machine recording to a live agent instead of dropping it or parking it. The agent listens, realizes, hangs up, and re-sets. On most floors that burns 15 to 30 seconds each time once you include the wrap-up and the mental reset. I have watched good agents take the full 30 seconds on a long greeting from a telco voicemail box.

Worked example

Below is a table with example inputs. Every number in it is one you should substitute with your own. Run it honestly, with numbers your floor supervisor would sign off on, not numbers that flatter the argument.

Input	Example value	Your value
Misroutes per agent per day	60	______
Average seconds lost per misroute	22	______
Agents on dialer	40	______
Loaded hourly rate	$19.50	______
Hours in a work day	8	______
Seconds lost per agent per day	60 × 22 = 1,320	______
Share of the paid day lost	1,320 ÷ 28,800 = 4.6%	______
Daily payroll lost	40 × 8 × $19.50 × 4.6% = $287	______
Annual payroll lost (260 days)	$74,600	______

Read that last row again. Seventy-four thousand dollars a year, on a 40-seat floor, at numbers that are middle-of-the-road. Not paid to anyone for anything. Just paid, by you, for silence and hang-ups.

Plug in 30 seconds instead of 22 and the annual figure crosses $100,000. A 120-seat BPO floor at the same rates is looking at roughly $225,000 a year. Your mileage will vary. But run it with your numbers before you decide it is small.

Where the seconds actually go

It is worth breaking the 15-30 seconds down, because owners hear "22 seconds" and assume it is exaggerated. It is not. Watch a single misroute end to end and count.

The bridge itself eats a beat or two. The agent greets, sometimes twice, because telco voicemail greetings often start with a pause. Then comes the recording, which on carrier voicemail systems can run 10 seconds before the beep if the box has a long personal greeting. Then the agent realizes, disconnects, and dispositions the call. That disposition is not free either. If your wrap-up code requires a note or a dropdown, add another few seconds of clicking. Then the agent re-sets mentally for the next call.

The long-greeting case is the expensive one. An insurance agent's voicemail that says "you have reached the office of, please leave your name and number after the tone" with background hold music is a 25-30 second event on a floor without decent detection. Short beep-only boxes are closer to 15. Average them across a day of mixed traffic and 20-ish seconds per misroute is a defensible planning number, which is what the table uses.

None of this appears on a timesheet. It appears only if you go looking, which is exactly why it persists.

Sanity-checking your misroute count

The input owners argue about most is misroutes per agent per day. Fair. Here is how to get a real number instead of a guess.

Pull a day of recordings and sample 50 calls per agent where the call connected but no conversation developed. Count the ones that were machine answers. Or sit behind two agents for an hour with a stopwatch and a tally sheet. Both methods take under an hour and give you a defensible figure. If your dialer logs every classification, and most do, the misroute count is sitting in a field you already export.

If your count comes in at 15 a day, the tax is real but modest. If it comes in at 100, you have a payroll problem disguised as a technology problem. I have seen both, on floors that looked identical from the owner's office.

What it does to your cost per conversation

Payroll lost is the direct hit. The sneakier one is what misroutes do to your effective cost per live conversation.

You pay agents for a full shift. Some of that shift becomes conversations, some becomes waiting, some becomes voicemail handling. Every minute that slides into the voicemail bucket raises the cost of every conversation that remains, because the denominator of talk time shrinks while the numerator of payroll stays fixed.

Take the 40-agent floor from the table. Suppose each agent holds 90 genuine conversations in an 8-hour shift. That floor produces 3,600 conversations a day. Now suppose you recover the 4.6% of paid time currently lost to misroutes and it converts at the same connect rate into conversations. Roughly 165 more conversations a day, from the same payroll, same leads, same everything. Your cost per conversation drops by about 4.4%, which on thin outbound margins is a real move.

The reverse framing is uglier. Every misroute is a conversation you paid to have and did not get.

I wrote more about treating these numbers as a unit economics exercise in cost per number as a life metric. The short version: when you measure cost per live conversation instead of cost per dial, detection quality stops being an IT topic and becomes a finance topic.

The attrition angle you are not measuring

Here is the part no spreadsheet captures. Misrouted voicemails do not just cost seconds. They cost energy, and energy is what your agents are actually selling you.

An agent bridged onto a machine was braced for a person. The greeting was ready. Then nothing, or a beep. That is a small deflation, and it compounds. Sixty times a day, a hundred times a day, the job becomes a rhythm of false starts. Agents describe it to me as exhausting in a way that live conversations are not, even difficult ones.

Why this shows up in turnover

Attrition on outbound floors is expensive and everyone knows it. Recruiting, training, ramp time, the productivity dip while a seat is empty. What floors rarely track is which conditions drive quit decisions. Exit interviews usually name pay and schedule. But talk to supervisors off the record and they will tell you the agents who burned out were the ones on campaigns with heavy machine-answer rates and sloppy detection.

I am not going to invent a statistic here, because none exists that I would trust. What I will give you is the mechanism, and it holds regardless of magnitude: agents quit jobs that feel wasteful faster than jobs that feel productive. Every misroute you remove removes a small dose of the wasteful feeling.

There is a second-order effect too. Your best agents, the ones who close, are the ones most punished by dead time, because their talk minutes are worth the most. A floor that wastes 5% of their day is underpaying your closers relative to what their conversations generate. They notice. Some of them leave for floors that waste less of their time.

Where the misroutes come from

At this point you may be asking why your dialer does this at all. The plain-English answer: your dialer has a built-in piece of software that listens to the first seconds of each answered call and guesses whether it heard a person or a machine. It guesses wrong a lot.

On the drops side, the default detector in a stock Vicidial-class dialer throws away an estimated 10 to 20% of live humans, meaning real people who said hello and got hung up on (AMDY measured that estimate on real detection traffic). On the misroute side, the same weak detection hands machines to agents. The two errors trade against each other. Tune the detector loose and you drop humans. Tune it tight and agents eat voicemails. Most floors run whatever shipped in the config file and get a mediocre mix of both.

I cover the trade-off in detail in AMD latency versus accuracy and in which error direction your dialer should prefer. The operator summary: a verdict that starts at 125 milliseconds, from a classifier that actually looks at the audio, beats a fixed-threshold guess from 2006 on both error types at once.

The compliance footnote

One more cost of sloppy detection that lands on payroll-adjacent budget: regulatory exposure. Under the FTC's Telemarketing Sales Rule, abandonment is capped at 3% of live answers per campaign, measured over 30 days (the rule lives at ecfr.gov, 16 CFR 310.4). Calls you hang up on because you thought a human was a machine count against you. Slow or wrong detection also drags out the silence a live answerer hears before your agent or message kicks in. If you want the full mechanics, I wrote them up in AMD logs as a compliance defense. The payroll point is simpler: fines and legal review are the most expensive dead minutes of all.

Three objections I hear, answered

"My agents use the dead time anyway, they disposition while listening." Some do. Watch whether the disposition is right when they do it split-attention. More often the agent is half-listening in case it is a slow-talking human, which is the whole problem: the agent cannot know it is a machine, so they cannot safely ignore it. That uncertainty is what makes the seconds unproductive even when they look busy.

"We tuned our detection already." Tuned how, and measured against what? Most floors that say this adjusted a couple of timer values years ago, when the campaign mix was different, and have not re-measured since. Detection accuracy drifts as your traffic mix, carriers, and lead sources change. I wrote about that drift in why AMD models drift and what to do about it. If you cannot quote your current misroute rate from a log, you have not tuned it, you have set it and forgotten it.

"Agents need breaks anyway, this is their breather." Agents do need recovery time. But recovery from a hard conversation is not the same animal as a dead-end hang-up into a machine. One resets, the other grinds. Ask your floor which they would rather have back. I have never had an agent pick the voicemails.

What the fix looks like

Modern answering machine detection is a server-side AI classifier. One sentence, as promised: instead of fixed timer thresholds, software listens to the answered audio and classifies it, person or machine, starting within about 125 milliseconds. That is the entire technology story you need.

What the fix means operationally matters more than how it works:

• Machines get caught before bridging. Agents stop receiving recordings.
• Humans who answer stop getting dropped by mistake, which is the 10-20% recovery we measure on the traffic itself.
• Agent routing does not change. Your call flows, dispositions, and scripts stay exactly as they are.
• Install is one command run by whoever administers your dialer, about 5 minutes, no carrier change, no hardware.

I will not pretend the change is free. It is a subscription. Current plans, for the record:

Plan	Monthly	Included detections	Overage per detection
Sandbox	$0, no card	50,000 (hard cap)	none, capped
Starter	$79	500,000	$0.00025
Growth	$299	5,000,000	$0.00015
Scale	$999	25,000,000	$0.00010

Compare the Growth number to the $74,600 annual payroll tax from the worked example. Even if your own misroute math comes out a fifth of that, the arithmetic is not close. And you do not have to take my word for any of it: the Sandbox plan is 50,000 detections a month, free, no card, and you can watch your own misroute rate fall on your own traffic before anyone pays anything. Details at pricing.

A rollout that does not disturb the floor

Do it on one campaign first. Keep the same agents, same list, same hours for a week, and compare misroute counts and conversations per agent-hour against the prior week. Detection changes are unusually easy to A/B because nothing about the human workflow moves. If the numbers do not convince you, you have spent five minutes of admin time. The ROI writeup walks through the same comparison with a calculator.

The peak-hour multiplier

One refinement worth adding to your math. Misroutes are not spread evenly across the day. Contact rates peak mid-morning and mid-afternoon, and so does machine-answer volume, because that is when nobody is home to pick up and voicemail boxes catch the call. Your worst detection hours and your most expensive hours are the same hours.

That matters for two reasons. First, the payroll tax lands exactly when a live conversation is worth the most, since a reached prospect at peak contact time is a prospect your competitors are also trying to reach that hour. Second, queue pressure at peak means the seconds an agent spends unwinding a misroute are seconds the rest of the queue waits, which stretches abandon rates in the wrong direction. Anything that shaves dead time at peak compounds, it does not just average out.

If you run the formula from this article, consider running it twice: once on daily averages, once on your busiest two-hour block with the misroute count that block actually produces. The second number is the one that justifies acting this month instead of this quarter.

Run your own number before you dismiss this

The voicemail tax survives being ignored because it is paid in small installments, seconds at a time, inside salaries you were paying anyway. It has no invoice, no line item, no angry vendor. It just quietly converts talk minutes you already bought into dead minutes nobody can sell.

The formula again: misroutes per agent per day × seconds per misroute × agents × loaded hourly rate ÷ 3,600. Ten minutes with your own numbers. If your answer is under a few hundred dollars a month, ignore me. If it is in the tens of thousands a year, the free tier is sitting there.

Here is the line I would keep: you would never let an agent bill eight hours and work seven and a half. Why let the dialer arrange it?