The Weekly Number That Predicts Next Month's Revenue

Your revenue report arrived, it is 9 percent below last month, and four people are now arguing about why. Wrong list. Bad script. Summer. The carrier. By the time anyone finds the real cause, another month is gone and the report about that month is being formatted.

I ran outbound floors for years before I sold detection for a living, and the most expensive habit I see is this one: steering the business by reports that describe the past. Revenue is a lagging indicator. It tells you what the floor did six weeks ago. The question you actually need answered every week is simpler: are we putting agents in front of live humans, or are we burning their hours on machines?

There is one number that answers it, you can compute it on Friday from data you already have, and when it moves, revenue follows. Not with certainty. With enough reliability that you stop finding out about problems from the invoice side.

The number

Live conversations delivered per agent-hour.

Count the calls where a person answered and an agent actually spoke with them. Divide by the agent hours your floor logged that week. That is the number.

It is one division, it is staff-normalized, and it captures the entire production capacity of an outbound floor in a single figure. Every path to revenue runs through it. No conversation, no pitch. No pitch, no sale. If your conversion rate holds anywhere near steady, weekly movement in this number shows up in monthly revenue with a lag you can verify on your own history in about ten minutes.

Here is the uncomfortable part. The default answering machine detection in most dialers drops an estimated 10 to 20 percent of live humans, measured by AMDY across real traffic. Those people answered. They said hello. The machine classifying the call decided they were voicemail and hung up. They never reach your revenue report at all, because your revenue report only counts what agents touched. The leak is invisible from the invoice side and obvious from the agent side.

Why revenue reports hide the cause

A revenue report is an autopsy. Something died weeks ago and you are reading the findings.

Worse, the cause is usually not in the report at all. When a detection problem silently removes 15 percent of your live answers, revenue dips, and nothing in the report points at detection. The list gets blamed. The script gets rewritten. A marketing VP gets a hard conversation. The dialer keeps deleting humans the entire time, and the cost per number you pay for data keeps buying answers you never let anyone speak with.

Lagging indicators share of fault and hide mechanism. Leading indicators isolate mechanism and buy you time. A week where conversations per agent-hour drops 12 percent is a week you can fix, because whatever broke is still warm. A revenue dip discovered a month later is archaeology.

There is a compliance dimension hiding here too. The FTC caps predictive dialer abandonment at 3 percent of live answers per campaign over 30 days under the Telemarketing Sales Rule, which you can read at ecfr.gov. Calls your detection drops on humans count against you. So the same number that leaks revenue also quietly builds regulatory exposure, and the weekly metric is how you notice both. I go deeper on the record-keeping side in the AMD logs compliance defense.

How to compute it, Friday, thirty minutes

You do not need a BI project. You need three numbers and a division.

Step 1: Count live conversations delivered

From your dialer's reporting, take the week's live-answer dispositions where an agent connected. Most floors already tag this. If your tagging is muddy, the stricter definition is better: a call counts only if an agent spoke and the other end was a human. Calls that rang to voicemail, dropped before connect, or auto-dispositioned do not count. You are measuring what reached your people, nothing else.

Step 2: Count agent-hours

From the same system, total logged agent hours for the week. Paid hours, not scheduled hours. If someone worked overtime taking calls, those hours are in the denominator because they are in your cost base.

Step 3: Divide

Conversations divided by hours. Write it down next to the week. Two decimal places is plenty. Do not build a dashboard yet. A column in a spreadsheet with one row per week is how this habit actually survives contact with a Monday.

One warning from experience: protect the definition. The moment someone redefines "live conversation" to include voicemail drops, the trend line is dead and you are comparing apples to a salad bar. Same definition, every week, forever.

The weekly scorecard

Here is the template. Five metrics, all from data you already collect, one owner each. The first row is the leading indicator. The rest explain it when it moves.

Metric	Source	Owner	Why it is on here
Live conversations per agent-hour	Dialer live-answer dispositions divided by logged agent hours	You	The leading indicator. Predicts revenue with a 3 to 6 week lag.
Agent minutes on machines	Dialer dispositions and handle times	Floor lead	Time you pay for that cannot buy anything.
Calls dropped, no disposition	Dialer logs	Dialer admin	Dead air. Often misclassified humans.
Detections logged per classification type	Detection logs (per-call logging)	Dialer admin	Shows whether classification or the list moved the needle.
Revenue	Finance report	You	The lagging confirmation. Put it on the same sheet.

That last row is the point of the whole exercise. When the top row moves and the bottom row follows a month later, twice, you will never go back to steering by the invoice.

An example with no invented numbers

I will not show you a case study, because a case study would require me to invent one, and you should not trust invented numbers. Instead, here is the formula and a hypothetical you can check against your own floor.

Take a floor with 20 agents averaging 30 logged hours a week, so 600 agent-hours. Suppose the dialer logs 4,200 live conversations delivered. That is 7.0 conversations per agent-hour. Now suppose detection is dropping an estimated 15 percent of live humans before they ever reach an agent. Recovered conversations would be 4,200 divided by 0.85, minus 4,200, which is about 740 a week, or roughly 1.2 per agent-hour. The metric would read 8.2 instead of 7.0.

Whether that matters is your arithmetic: 740 extra conversations times your conversion rate times your average deal. Use your real rates. The point of the example is not the outcome, it is that the number makes the leak countable. A 15 percent leak that never appears in any report is a budget line you are paying every week and cannot see. Once conversations-per-agent-hour exists, it shows up as a weekly reading you can act on.

The same logic works in reverse, which is the part most floors miss. When the metric slips from 7.0 to 6.5 over three weeks, that is not noise to be averaged away. It is roughly 300 conversations a week that stopped happening, and it will surface in revenue a month later with nobody able to explain it. The weekly number is your early warning, and it costs one division.

What movement means

A number you do not interpret is decoration. Here is how I read movement after years of watching floors.

If conversations per agent-hour drops and agent minutes on machines rises, your detection is drifting. Lists age and answer behavior shifts with seasons, and a static classifier slowly loses the plot. This is not hypothetical; it is the subject of AMD model drift, and it is why I distrust set-and-forget detection. Check the classification-type row on the scorecard before you touch the list.

If conversations per agent-hour drops and machines hold steady, the problem is upstream: list quality, dial rate, carrier issues, or something poisoning your answer rates. The scorecard just saved you from rewriting a script that was never the problem.

If conversations per agent-hour rises and machines fall, capacity came back. Either detection improved or the list got hot. The classification row tells you which, and that distinction matters, because list luck reverts and detection gains persist.

If everything is flat for a month, that is a finding too. Flat means the constraint is conversion, not delivery, and your money should go to scripting and training rather than infrastructure. Most floors never get that answer cleanly because they never isolate the number.

Connect it to money once a quarter

Once per quarter, run the regression on yourself. Take your weekly conversations-per-agent-hour figures and your revenue by week for two quarters. Compute the lag that maximizes the correlation. Most floors land somewhere between three and six weeks. That lag is now your forecast horizon, built from your own data, not from my experience or anyone else's.

Then the arithmetic writes itself. Conversations times your historical conversion rate times average revenue per deal equals expected revenue, one lag period out. Your weekly number just became a forecast, and you built it in a spreadsheet for free.

This is also the honest way to evaluate anything you change on the floor, including detection. We tell every prospect the same thing: run the free trial, 50,000 detections a month, no card, and watch one number for a week. Install is a single command by the dialer admin, about five minutes, running alongside your current setup. If conversations per agent-hour does not move, do not buy. The number is the referee, not the salesperson. The full method is in our features overview, and the ROI page turns the delta into dollars with your own volumes.

What about the FTC ceiling?

One more reason the weekly rhythm matters, and it has nothing to do with forecasting. The 3 percent abandonment cap is measured per campaign over 30 days, which means by the time a monthly compliance report shows a problem, the window has closed and the evidence is whatever your logs happen to contain. A weekly read of your dead-air and machine buckets turns that 30-day surprise into a weekly check. Floors that catch detection drift in week one do not spend week four explaining a number to anyone.

I am not a lawyer and this is not legal advice. It is an operating observation: the same scorecard that protects revenue also happens to be the one that would protect you, because both failures start with live answers not reaching agents. One sheet, two kinds of risk covered. That is a rare bargain in compliance-adjacent work, where the usual price of peace of mind is a second consultant.

A note on who owns the number

The scorecard template assigns the top row to you, the owner or VP, and that is deliberate. Delegate the pulling of the data, never the reading of it. The week this number becomes something a manager forwards with a shrug is the week it stops predicting anything, because the people who can act on it have stopped looking.

Fifteen minutes on a Friday. You already have the meeting. The data is already in the dialer. The only new thing is the division, and the only question is whether you want to hear about problems in three weeks or six.

What the first month of readings looks like

The first four weeks of keeping this number are diagnostic, and you should resist drawing conclusions until they are done.

Week one, you will discover your dispositions are muddier than anyone admitted. Half the floor tags machines one way, the other half another way, and a couple of agents have a personal code for "weird call." Clean it once, write the definition down, and do not let the cleanup become a project. Two disposition codes, human and machine, plus the dead-air bucket for everything that never connected to a disposition.

Week two, the number stabilizes and you get your first real reading. Write it down even if it looks wrong. It is not wrong; it is your floor, measured honestly, possibly for the first time.

Week three, you will be tempted to act on the first delta you see. Hold. One week of movement is weather. Two consecutive weeks moving the same direction is a signal. This is the single most common way owners misuse a leading indicator: they treat the first wiggle as a trend and burn credibility with the floor by reacting to noise.

Week four, you have a baseline with error bars, because you have now seen what normal weekly variance looks like on your own traffic. Every reading after that is measured against something real.

A month of discipline for a permanent early-warning system. Cheaper than the consultant you would otherwise hire to explain the Q3 revenue miss, and it drinks nothing.

Three ways this number gets gamed

Any metric that runs a floor gets gamed, and this one is no exception. Know the failure modes before you laminate it.

The first is definition creep. Someone reclassifies "agent connected but hung up in two seconds" as a live conversation, because it technically was one, and the number jumps without anything improving. Fix: the disposition code is frozen, in writing, and any change to it gets dated in the spreadsheet. Undated definition changes are how trend lines die.

The second is denominator games. If agents log in and sit idle during slow hours, agent-hours inflate and the number falls for reasons having nothing to do with delivery. That is actually the metric working as intended, because idle hours are real cost, but you will hear arguments for "productive hours" instead. Resist them. Paid hours is the honest denominator. The number is dollars in and conversations out, not a feelings report.

The third is chasing the number itself. Run it up by hammering a fresh list and conversion collapses, and revenue falls even though the metric rose. This is why the scorecard has five rows and not one. Conversations per agent-hour is the leading indicator, not the only one, and any week it moves you check the other rows before acting. A number that outruns its context becomes a target, and a target gets hit whether or not it pays.

The discipline, if you want it in one line: the number starts conversations about the floor. It does not end them.

The habit is the hard part

You will compute this number twice and then skip a Friday. I have done it. The fix is to attach it to something that already happens weekly. Send it in the same email as the schedule. Review it in the same meeting as the list order. A metric that requires a new meeting dies within a quarter.

Give it one home, one owner, one definition. Five rows on a spreadsheet. The first week you catch a problem three weeks before revenue reports would have shown it, the habit pays for itself, permanently, and you will wonder what you were looking at before.

What did last Friday's number say?