55% of US adults are regular supplement users. It is a real number from a respected annual survey. In one supplement market model, it was doing a different job: the share of last year's buyers who buy again this year.

The survey cannot tell you that. And because a re-order rate is applied in every forecast year, the gap does not stay small. If the true rate were 40%, the model's customer base would end up a third too large, and its revenue with it.

Nothing was wrong with the formulas. The number was real and properly sourced. It just measured something other than what the model needed.

Below: what the 55% really measures, why the damage multiplies, what to do when the real figure is not available, and the one question that catches this before it reaches a model.

If you review target forecasts, market sizings or management decks before investing, this is one of the mistakes most worth catching early.

WHAT THE 55% ACTUALLY MEASURES

The figure comes from the CRN 2023 Consumer Survey on Dietary Supplements. The model read it as "55% of last year's buyers come back."

The survey cannot say that. It was one online poll of 3,192 US adults, run over five days in August 2023 (Ipsos). It asked people how they use supplements now. It did not follow the same people from one year to the next, so it cannot tell us who came back, who stopped or who is new.

The input differs from what the model needs in three ways:

• People: the model needs last year's buyers. The 55% covers all US adults.

• Time: the model needs the same people tracked across two periods. The survey covers one week in August 2023.

• Meaning: the model needs people who bought again. The survey counts people who call themselves "regular" users.

This is the mistake I call a snapshot used as a rate. A share of people who use a product today is a snapshot. Retention, churn and re-order rates are rates: they need the same people tracked across two periods.

WHY A SMALL GAP BECOMES A BIG ONE

A retention rate is applied every year, to a base that already includes every earlier year's survivors. If a business gains N new users a year and keeps a share r of its users each year, the user base settles at:

Customer base = new customers per year ÷ (1 − retention rate)

So the retention rate decides how many times larger than one year's new users the business becomes. Small moves in r produce large moves in the answer:

At 40% retention, the base settles at 1.7 times one year's new users. At 55%, it settles at 2.2 times. At 70%, 3.3 times. At 90%, 10 times.

Illustration:

if the true re-order rate were 40%, a model using 55% would overstate the settled user base by a third, and revenue with it. Nothing in the survey tells us whether the true figure is 40%, 55% or 70%. That is the problem.

WHEN THE REAL NUMBER ISN'T DISCLOSED

Often there is no true retention figure, and the model still needs one. A proxy can stand in, as long as it is handled in the open:

1. Label it as an assumption. Say what the number actually measures and what it is standing in for.

2. State which way it is biased, and what that does to revenue.

3. Test it against the company's reported totals. A rate that cannot reproduce them is wrong.

4. Show a range, not one value, so the reader sees how much rides on it.

5. Ask management, and replace the proxy when the real data arrives.

Illustration:

a subscription company does not disclose retention, but its deck quotes a 90% "repeat rate". It reports 200,000 active customers in Q1 and 150,000 in Q2. If 90% of Q1's customers had come back, Q2 would have started at 180,000 before a single new customer joined. The reported 150,000 means no more than 75% returned, and fewer than that, because the company was still adding new customers. The proxy should sit below 75%.

The snapshot itself is not wasted. The CRN 55% is useful context on how widespread supplement use is. It just cannot carry the forecast.

In diligence, this is where a target's forecast is often weakest. Companies rarely share real retention data. Instead, their decks quote numbers like "active users" or "repeat rate", and the forecast treats them as retention. On a target model, I ask for the real data and check the rate against the company's reported customer numbers.

THE ONE QUESTION TO ASK

Before any number becomes a retention, churn or re-order rate, ask: did the source track the same people across two periods? If not, it is a snapshot. Keep it as context, and find a rate built from repeat-purchase, subscription or cohort data for the actual product.

This is the first article in a series. Each one takes a single check from the list I run on research inputs before they go into a market sizing, forecast or valuation, and every check comes from a mistake I have actually found.

Next: a ratio measured in dollars applied to a count of people, and why a 10% share of spend can mean anything from 2% to 100% of buyers.

Vaishali Chawla is an independent consultant in financial modelling, valuation and investment research for SPAC sponsors, investors and deal teams. vaishalichawla.com