Preparing Data for Demand Estimation
Demand estimation is not fragile because the math is difficult. It is fragile because it will happily operate on data that should never have been estimated in the first place.
This toolkit exists to prevent that mistake.
Before fitting models, comparing curves, or optimizing anything, you should pause and ask a simpler question:
Is this data actually fit for estimation?
This toolkit helps you answer that question deliberately.
It does not teach estimation. It teaches when estimation makes sense at all.
The Goal of This Toolkit
The goal here is not to perfect your data. That is impossible at this stage.
The goal is to ensure that:
- the data means what you think it means,
- it aligns with the entrepreneurial question you are trying to answer,
- and estimation will clarify uncertainty rather than disguise it.
If you cannot confidently answer the questions in this toolkit, the correct response is not to “try the model anyway.” It is to return to design, execution, or validation before proceeding.
Step 1 — Confirm the Decision Context (Again)
Before looking at the data, restate the decision it is meant to inform.
- What decision are you trying to make now?
- What would you do differently if demand were higher, lower, or more uncertain than expected?
If you cannot answer this cleanly, stop here.
Demand curves do not create decisions. They only inform decisions that already exist.
Step 2 — Verify Unit and Period Consistency
Estimation only makes sense if every observation refers to the same decision.
Ask yourself:
- Is the unit consistent across all responses?
- One subscription?
- One visit?
- One meal?
- One license?
- Is the decision period consistent?
- Per month?
- Per year?
- One-time?
- Repeating?
If different respondents were implicitly answering different versions of the question, the data cannot be meaningfully aggregated.
This is not a minor flaw. It is a category error.
If unit or period ambiguity exists, estimation will produce a curve — but the curve will not correspond to any real decision.
Step 3 — Confirm the Type of Demand Being Estimated
Before estimation, you must know which kind of demand your data represents.
Yes/No Demand
- Quantity is fixed at one unit per period.
- Variation comes from who buys at which prices.
- Estimation is about adoption probability across prices.
How-Many Demand
- Quantity varies across customers.
- Estimation must reflect both price sensitivity and usage intensity.
- Quantity responses must be explicitly tied to the defined period.
If your data mixes these unintentionally — for example, treating repeated consumption as yes/no demand — estimation will be misleading even if the model fits well.
This is a design problem, not a modeling one.
Step 4 — Inspect the Raw Price–Quantity Structure
Before fitting anything, look at the data in its most basic form.
Ask:
- Are prices within a plausible range?
- Are quantities non-negative and realistic?
- Are there obvious outliers that dominate the scale?
- Is there enough variation in price to identify a relationship at all?
Estimation cannot recover information that was never collected.
If all respondents faced essentially the same price, the problem is not statistical. There is nothing to estimate.
Step 5 — Check Effective Sample Size
Demand estimation depends on informative variation, not just respondent count.
A dataset with:
- 200 respondents
- but only 3 distinct price points
- or only a handful of non-zero quantities
… may have far less effective information than it appears.
Ask:
- How many distinct price–quantity relationships are actually present?
- How much of the data survives basic cleaning and filtering?
Small effective samples do not make estimation impossible — but they do limit what can be inferred responsibly.
Treat results accordingly.
Step 6 — Identify What Estimation Can and Cannot Do Here
Before running the app, be explicit about what you expect estimation to provide.
Estimation can help:
- summarize noisy responses into a coherent relationship,
- rule out implausible prices,
- compare broad demand shapes.
Estimation cannot:
- fix poor survey design,
- recover missing context,
- validate demand on its own,
- eliminate uncertainty.
If you are hoping estimation will “settle” the decision, pause. That is not its role.
Step 7 — Decide Whether to Proceed
At this point, make an explicit choice.
You should proceed to estimation only if:
If any of these fail, the right next step is not modeling.
It is to:
- refine the survey,
- gather additional data,
- validate earlier responses,
- or narrow the decision.
Estimation is powerful — but only when used at the right moment.
How This Toolkit Fits the Larger Process
This toolkit sits deliberately between executing demand experiments and estimating demand.
It prevents the most common failure mode in early analytics: producing confident outputs from incoherent inputs.
Once these checks are satisfied, estimation becomes meaningful — not because it is precise, but because it is anchored.
Only then does it make sense to fit demand curves, compare models, and bring cost back into the picture.