RevOp  /  Post-mortems  /  14
Post-mortem 14

The month we missed by a fifth, with a conversion rate that never moved.

Module 19 · Forecasting Stage · repeatable, and scaling Instrument · The Forecast Rubric

I had always assumed that a badly inflated pipeline would show up as a falling conversion rate. It does not. It cannot, and the arithmetic of why took me far too long to work out.

What happened

A couple of months missed badly. Mid-six-figures each, which was material to profit and close to invisible on the topline. That is part of why it took a while to become a crisis rather than a bad month.

The conversion rate held steady the entire time. Month after month, roughly the same number. We looked at it repeatedly, because it was the first place anybody looks, and it kept telling us the machine was fine.

The pipeline was inflated. Reps were assigning 90% to deals that were not 90%, and nobody could say otherwise, because the facts that would have settled it were never asked for. A commit was a rep’s opinion with a dollar value attached to it.

What we said at the time

“Conversion’s holding. It’s a volume problem, we just need more at the top.”

“It was a tough month. The market got choppy.”

“A couple of deals slipped. They’ll land next month.”

“He’s been right before. If he says it’s 90, it’s 90.”

The mechanism

A measured conversion rate is a blend of two numbers, and you only ever see the product.

Every rate you measure is the real rate on real opportunities, multiplied by the share of your pipeline that was ever real at all. Two numbers, one readout.

So if both of them are stable (and they usually are, because a team’s habits are stable) the blend is stable too. A stable conversion rate is not evidence of health. It is evidence of consistency, which is a different thing entirely.

The rate cannot tell you which company you are.

And it gets worse in exactly the moment you need it most.

Two very different companies, reading the same

Here are two businesses. One is healthy with a modest amount of noise in the pipeline. The other is in serious trouble. Their reported conversion rates are indistinguishable.

 Real ratePhantom shareYou measure
Company A
healthy, a little noise
32%5%30.4%
Company B
excellent sellers, fictional pipeline
60%50%30.0%

A model, not measured actuals, but the arithmetic is exact: what you measure is the real rate multiplied by the share of pipeline that was genuine.

Company B has the better sales team. Their people convert nearly twice as well on anything real. They are also one bad quarter from a serious miss, and their conversion rate will never once warn them.

Why misses cluster in a tight market

This is the part that changed how I read a bad quarter.

A tight market does not lower your conversion rate. It unblends it. When budgets tighten, the deals that disappear first are the ones that were never qualified: the price-checks, the people being polite, the opportunities that existed so a box would not be empty. The real deals mostly survive.

So the phantom share collapses, the measured rate can even rise, and the number still misses badly, because the volume that vanished was volume you had been counting.

Which is why everyone concludes the market caused it. The market only revealed it. The inflation was already there, being paid for every month, and a tight quarter simply stopped hiding it.

What it cost

Mid six figuresMissed in each of a couple of months, material to profit, invisible on the topline
No warningFrom the one metric everybody checks first, for the entire period
7–10 pointsOf realised discount, in the years when the shortfall got paid off at month end instead

The discounting is the part worth dwelling on, because it looked like a separate problem and it was not. A forecast you cannot trust gets discovered late, and something discovered late can only be fixed with the most expensive lever you have left.

What replaced it

A commit now requires facts, not a percentage. A short, fixed set of things that must be known and answerable about a deal before it can sit in the number: who has seen it, what is holding them back, whether a figure has been attached, what has to be finished by when, and who else is in the conversation.

Then the forecast is produced twice: once as the reps called it, once on the evidence. The gap between the two is not a forecasting error. It is the coaching conversation, and it names the exact deals and the exact missing facts.

The three things that came out of it

1 · Large deals are excluded from the commit by default. Not discounted. Excluded. A miss driven by one whale is uncorrectable; a miss spread across many small deals can be worked. This converts the biggest source of downside into the only source of upside.

2 · The levers get more expensive as the date approaches, in a fixed order, and then they run out.

When you find outWhat is still available
A full cycle outGeneration. Costs effort, and a planned promotion lives here, modelled in advance, priced in
Ten days outActivity, and an incentive that does not erode margin. Cheap
Four or five days outRealistically nothing
Month endThe miss, or a discount, which is the miss in another currency

3 · Month-end discounting was never the problem. I spent years treating it as a pricing-discipline failure and putting approvals on it. Approvals decay under pressure. Once the forecast got honest the discounting mostly stopped on its own, because there was nothing left to rescue at the last minute.

There are two discounts and they only look alike. One is modelled before the period starts and is a demand instrument. One is a reach in the final week and is a detection failure being paid off at the register. The test is the calendar, not the amount, and the count of month-end discounts is a lagging indicator of forecast quality, not a pricing statistic.

Is your conversion rate hiding this right now?

You cannot tell from the rate. That is the whole finding. What you can do is look at the shape of your plan: how much of the number rests on how few deals, and what happens to it when the mix moves. The model is free, takes about four minutes, and the link it gives you holds your figures so you can come back to them.

Run the model, free No account. Your numbers stay in the link.