Sanity-Check Your Estimated Demand Curve

A fast way to catch curves you shouldn’t trust

This toolkit is a practical companion to Chapter 8 Estimating Demand. It does not explain how to operate the profit analytics app. It explains how to interpret the results it produces

The profit analytics app can fit a demand curve from almost any dataset. That is a feature—and a risk. A smooth curve can look authoritative even when it is summarizing confused evidence, misaligned units, or a sample that cannot support market inference.

The goal here is not to prove your curve is “true.”
The goal is to catch the curves that are not safe to use for decisions.

You can do this in five minutes.

Start With the Decision (One Sentence)

Write the decision you are trying to support as one sentence:

  • “We are deciding whether to enter.”
  • “We are deciding whether this price range can support our costs.”
  • “We are deciding whether to invest in fixed capacity.”

If you cannot write that sentence, stop.
A demand curve without a decision frame becomes an art project.

Unit + Period: The Two Silent Failure Modes

Before you look at the curve, confirm these two definitions in plain language.

Unit (what counts as “one”):
What is the thing the customer imagines choosing?

Examples:

  • one subscription
  • one visit
  • one meal
  • one session
  • one item

Period (how often the decision repeats):
How often can the customer make that choice?

Examples:

  • per month
  • per semester
  • per year
  • per week

If unit or period is ambiguous, the curve may still fit—but it will not mean what you think it means.

Does the Shape Pass the “Human Behavior” Test?

Before any fit statistic, ask whether the curve behaves like real customers.

A demand curve should usually satisfy three sanity conditions:

  1. Quantity is never negative
    If your model implies negative quantity in the relevant range, treat it as a warning sign.

  2. Demand does not explode at low prices
    If the model implies absurdly high quantities at very low prices, it may be extrapolating beyond what your evidence can justify.

  3. Demand approaches zero at high prices
    At some sufficiently high price, almost no one should buy.
    If the model implies substantial demand at very high prices, something is off—or your unit/period is inconsistent.

You do not need economics training for this.
You just need to ask: would a real customer behave like this?

Compare the Curve to Your Raw Evidence (Visually)

A fitted curve should not surprise you when you overlay it on the evidence.

Look for these red flags:

  • The curve sits consistently above most points (systematic overstatement)
  • The curve sits consistently below most points (systematic understatement)
  • The curve fits the middle but fails badly at the ends
  • A handful of points appear to “drag” the curve dramatically

One of the most useful sanity checks is simple:

If you hide the curve and look at the points, would you expect a curve like this?

If the fitted curve feels like it has a different story than the data, treat that as information.

“Does This Predict Something I Already Know?”

Pick one price that you have some intuitive anchor for—based on the market, a comparable product, or your own purchasing experience.

Then ask:

  • At that price, does the predicted quantity feel plausible for your defined unit and period?

This is not a test of truth.
It is a test of interpretability.

If the curve produces predictions you cannot even reason about, you do not yet have a usable demand object.

A Simple Anchor Test (One-Point Validation)

If you can do one validation step, do this:

Choose one price, $X, and ask a fresh sample:

“If the price were $X, would you buy one unit in the next [period]?”

Then compare:

  • the predicted adoption/quantity at $X from your curve
  • the observed adoption/quantity from the fresh sample

You are not trying to “prove” the curve.
You are checking whether the evidence is coherent.

If the validation result is wildly different, the most likely culprit is not the model. It is the upstream evidence: sampling, framing, or unit/period confusion.

(Chapter 7 and Toolkit 6C cover how to interpret this responsibly.)

What to Do If It Fails a Sanity Check

If your curve fails one of these checks, do not “fix” it by forcing a different model.

Instead, go upstream:

  • Re-check unit and period
  • Re-check whether demand is yes/no or how-many
  • Inspect whether non-customers leaked into the sample
  • Look for framing that made the WTP question feel transactional
  • Consider whether you need a validation step before modeling again

A demand curve is only as credible as the evidence beneath it.

The Goal: A Curve You Can Reason With

A usable demand curve is not one that looks smooth.

A usable demand curve is one that:

  • aligns with your unit and period,
  • behaves like real customers,
  • is consistent with the evidence,
  • and produces predictions you can interpret in the context of a decision.

If you have that, you have something rare in entrepreneurship: a structured object you can argue with.

That is what makes disciplined decisions possible.