Demand Estimation and Market Measurement

Introduction

Managerial decisions depend on an informed estimate of how much customers will buy, at what price, under which conditions, and over what time period. Demand estimation converts market information into quantitative evidence for pricing, capacity planning, product development, sales forecasting, advertising allocation, and investment appraisal. Market measurement is broader: it identifies the size, structure, growth, profitability, and accessibility of a market.

Demand analysis is not the same as sales reporting. Observed sales reflect not only customer demand but also stock availability, distribution coverage, competitor actions, sales-force effort, product quality, regulation, and price. A decline in sales may therefore indicate weaker demand, reduced availability, or an ineffective marketing programme. The managerial task is to separate these influences as far as possible.

This chapter explains market research, survey methods, experiments, econometric estimation, customer analytics, willingness-to-pay measurement, market potential, and the limitations of each approach.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Distinguish demand, sales, market size, market potential, and market share.
2. Identify the determinants of demand and formulate an estimable demand model.
3. Design and evaluate surveys, experiments, and observational studies.
4. Interpret price, income, cross-price, and advertising elasticities.
5. Estimate willingness to pay and assess market potential.
6. Use customer analytics and econometric evidence for managerial decisions.
7. Recognise sampling, measurement, causal-inference, privacy, and ethical limitations.

1. The Meaning and Scope of Demand

Demand is the quantity of a product or service that consumers are willing and able to purchase at alternative prices during a specified period, holding other relevant conditions constant. The phrase “willing and able” is important: a consumer may desire a product but lack income, access, or authorization to buy it.

A demand function expresses the relationship between quantity demanded and its determinants:

Here, Q_d is quantity demanded per period; P is the product’s price; Y is consumer income or budget; P_s and P_c are prices of substitute and complementary products; A is advertising or promotion; T represents tastes and preferences; E represents expectations; and N represents the number and composition of potential buyers. The function is applicable only when the variables are clearly defined and measured over comparable periods.

A useful linear specification is:

where β₀ is the intercept, β_i are parameters to be estimated, and u captures omitted influences and random variation. Economic theory commonly predicts β₁ < 0, β₂ > 0 for a normal good, and β₃ > 0 when the other product is a substitute. These signs are predictions, not guarantees; unusual goods and poorly measured variables may produce different results.

Demand, sales, and market measures

If firm sales are 40,000 units and total market sales are 200,000 units, market share is:

The denominator must be defined carefully. A firm may have a 20% national share but a 35% share in its target segment or distribution channel.

2. Market Definition and Measurement Architecture

Before collecting data, managers should specify five boundaries:

1. Product boundary: What products satisfy the same customer need?
2. Customer boundary: Which users, buyers, or decision makers are included?
3. Geographic boundary: City, region, country, or digital territory?
4. Time boundary: Month, quarter, year, or product life cycle?
5. Transaction boundary: Retail, wholesale, subscription, replacement, or resale?

A narrow definition can overstate competitive advantage; an excessively broad definition can make estimates irrelevant. For example, a premium plant-based beverage competes directly with similar beverages but may also compete for consumers’ breakfast and refreshment budgets.

A practical market-sizing framework is TAM–SAM–SOM:

• Total Addressable Market (TAM): demand if the firm could serve all relevant customers.
• Serviceable Available Market (SAM): the portion compatible with the firm’s geography, capabilities, channel, and product.
• Serviceable Obtainable Market (SOM): the realistic share that can be captured given competition and resources.

A top-down estimate may use industry revenue. A bottom-up estimate may use customers, purchase frequency, and average transaction value:

The units must be consistent. If 50,000 buyers purchase four units per year at $12 per unit, estimated annual value is:

This is a potential-value estimate, not a forecast. It assumes that all identified buyers are reachable, willing, and able to purchase at the assumed frequency and price.

3. Research Designs for Demand Estimation

No single research method is universally best. Strong managerial studies often triangulate several sources.

Secondary research

Secondary data include government statistics, industry reports, transaction records, syndicated databases, regulatory filings, web analytics, and competitor information. They are inexpensive and useful for market structure, trends, and benchmarking. Their weaknesses include unsuitable definitions, unknown data quality, publication lag, and inconsistent measurement.

Qualitative research

Interviews, focus groups, observation, ethnography, and diary studies reveal needs, vocabulary, usage occasions, barriers, and decision processes. They are especially valuable in early product development. Qualitative findings should not automatically be interpreted as population percentages because purposive samples are not designed for statistical generalisation.

Surveys

A survey measures stated attitudes, intentions, preferences, awareness, and self-reported behaviour. A sound survey requires:

• a defined target population and sampling frame;
• a sample-size rationale;
• neutral, specific questions;
• clear response scales;
• pilot testing;
• procedures for nonresponse and data cleaning; and
• documentation of weighting and estimation.

Leading questions, double-barrelled questions, vague time periods, and hypothetical overstatement create measurement error. “Would you buy an affordable premium service?” is less useful than “How likely are you to purchase one subscription within the next 30 days at $25 per month?”

For a simple proportion estimate, an approximate sample-size formula is:

Here, n₀ is the initial sample size for a large population, z is the standard-normal critical value for the selected confidence level, p is the anticipated proportion, and e is the desired margin of error. The formula assumes a probability sample, independent observations, and a proportion estimate. When p is unknown, 0.5 gives the largest variance and therefore a conservative sample size.

For a 95% confidence level, z ≈ 1.96, p = 0.5, and a margin of error of 5%:

This does not mean that any sample of 384 people is representative. A biased convenience sample may remain biased regardless of its size. For a finite population of size N, a finite-population correction may be used:

Experiments and quasi-experiments

Experiments manipulate a treatment, such as price, advertisement, interface, or product feature, while comparing outcomes with a control condition. Random assignment helps ensure that average differences are attributable to the treatment rather than pre-existing group differences.

A field experiment might randomly expose comparable customers to two prices. Its advantages are behavioural measurement and stronger causal inference. Its risks include spillovers, short observation periods, novelty effects, operational errors, and external-validity problems.

When randomisation is impossible, quasi-experimental approaches may use difference-in-differences, matched controls, regression discontinuity, instrumental variables, or interrupted time series. Difference-in-differences compares changes over time in a treated group with changes in an untreated group:

The key assumption is parallel trends: without treatment, the treated and control groups would have followed similar trends. This assumption should be examined rather than merely asserted.

4. Econometric Demand Estimation

Econometric estimation uses observed variation to quantify relationships while controlling for several influences. A multiple regression demand model can be written as:

The subscript t denotes a time period; S_t may represent seasonality or distribution coverage. Ordinary least squares estimates parameters by choosing coefficients that minimise the sum of squared residuals:

where Q̂_t is predicted quantity. Regression estimates are informative only under suitable conditions, including a correctly specified model, meaningful variation, reliable measurement, limited influential errors, and an appropriate treatment of dependence and heteroskedasticity.

Elasticity

Elasticity measures percentage responsiveness and is unit-free. The point price elasticity of demand is:

For a linear model, ∂Q ÷ ∂P = β₁. If E_P = −1.5, a 1% price increase is associated with an approximately 1.5% decrease in quantity, within the observed range and holding other variables constant. Demand is elastic when |E_P| > 1, inelastic when |E_P| < 1, and unit elastic when |E_P| = 1.

The income elasticity is:

Cross-price elasticity is:

A positive value generally indicates substitutes; a negative value generally indicates complements. Advertising elasticity is:

Elasticities are local or sample-dependent measures. They should not be extrapolated far beyond the data range.

Logarithmic models

A constant-elasticity model is:

In this specification, β₁, β₂, and β₃ are elasticities directly, subject to the usual interpretation of a ceteris paribus association. Log models require positive values for variables transformed by the natural logarithm and need careful treatment of zeros.

Identification and endogeneity

Price is often endogenous: firms may lower prices when demand is weak, while weak demand itself causes low sales. A simple regression may then confuse the effect of price with unobserved demand conditions. Other problems include omitted competitor prices, reverse causality between advertising and sales, multicollinearity, simultaneous supply and demand, and autocorrelated time-series errors.

Possible remedies include randomised prices, instrumental variables, controlled experiments, natural experiments, fixed effects, lag structures, and richer market models. An instrumental variable must be relevant to the endogenous regressor and satisfy an exclusion condition: it affects the outcome only through that regressor. This is a demanding substantive assumption, not a statistical formality.

5. Measuring Willingness to Pay

Willingness to pay (WTP) is the maximum amount a customer is prepared to sacrifice for a product, service, or attribute, under specified conditions. It is not necessarily the price eventually paid. WTP may be measured through:

• Observed purchases: behaviour at actual prices, with strong realism but limited variation.
• Auction or incentive-compatible mechanisms: participants face real consequences, but implementation can be costly.
• Choice experiments: respondents choose among alternatives with systematically varied attributes and prices.
• Contingent valuation: respondents state or select a payment for a defined scenario.
• Direct questions: easy to administer but vulnerable to hypothetical and strategic bias.

In a discrete-choice model, the probability that customer i chooses alternative j may be represented by the multinomial logit model:

where V_ij is the systematic utility of alternative j, often expressed as a function of attributes and price. The model assumes independent and identically distributed extreme-value errors, producing the independence-of-irrelevant-alternatives property. If that property is implausible, nested logit, mixed logit, or other models may be more appropriate.

A simple utility specification may be:

Under appropriate coding and a negative price coefficient, an attribute’s marginal WTP can be approximated by:

The signs and units must be interpreted carefully. If price is measured in dollars and the feature coefficient is positive, the ratio gives dollars per feature unit. Confidence intervals should be reported because WTP is estimated, not observed with certainty.

6. Customer Analytics and Segmentation

Customer analytics integrates transaction, behavioural, demographic, service, digital, and survey data. Common analyses include:

• RFM analysis: recency, frequency, and monetary value;
• customer lifetime value;
• churn and retention modelling;
• recommendation systems;
• propensity scoring;
• cohort analysis;
• customer segmentation; and
• marketing-mix and attribution analysis.

A simplified customer lifetime value formula for a stable subscription business is:

Here, m is contribution margin per period, r is the retention probability per period, d is the discount rate per period, and CAC is customer acquisition cost. The formula assumes constant margin and retention, a one-period timing convention, and an indefinite horizon. It should not be applied mechanically when margins, retention, or acquisition costs vary substantially.

For example, if monthly contribution margin is $20, monthly retention is 0.95, monthly discount rate is 0.01, and acquisition cost is $150:

This is an expected contribution estimate, not accounting profit. Returns, service costs, taxes, overhead, and changing behaviour may materially alter it.

Segmentation should be actionable, measurable, substantial, accessible, and stable enough to support decisions. Machine-learning clusters can identify patterns, but statistical clusters are not automatically meaningful market segments. Managers must connect segments to value propositions, channels, economics, and ethical safeguards.

7. Forecasting and Scenario Analysis

Demand estimation explains relationships; forecasting predicts future outcomes. Time-series forecasting may use moving averages, exponential smoothing, regression with seasonality, or causal models. A simple moving average of the previous k observations is:

It is easy to use but gives equal weight to past observations and may lag during turning points. Forecast accuracy can be assessed with mean absolute error:

where n is the number of evaluated periods. Percentage measures should be avoided or adjusted when actual values are zero or close to zero.

Scenario analysis is often more credible than a single-point estimate. Managers can construct conservative, base, and optimistic cases by varying price, adoption, distribution, competitor response, and retention. Sensitivity analysis identifies which assumptions most affect the result. Forecast intervals should accompany forecasts where possible; uncertainty is part of the decision, not a defect to conceal.

8. Illustrative Case: Launching a Low-Sugar Functional Drink

Background

A fictional beverage company, VitalSip, is considering a low-sugar functional drink for urban professionals. Management believes that health concerns and convenience create an opportunity but lacks evidence about price, market size, and repeat purchase.

Problem

The company must decide whether to launch nationally, begin with a regional pilot, or abandon the concept. It needs estimates of potential buyers, WTP, price elasticity, trial probability, and repeat purchase.

Analysis and approach

VitalSip first defined the market as non-alcoholic, ready-to-drink functional beverages sold through supermarkets, convenience stores, gyms, and direct digital channels. Secondary data estimated the relevant regional category at $120 million annually, but category data did not isolate the low-sugar segment.

The company then conducted qualitative interviews with consumers and retailers. Interviews identified three important attributes: sugar content, functional benefit, and price. A pilot survey of 600 adults measured consumption frequency, health motivations, and purchase intent. The sampling design included quotas for age and gender, but the company correctly treated the results as weighted estimates rather than perfect population measurements.

A choice experiment presented respondents with alternatives varying in price, sugar level, benefit, brand, and “no purchase.” The estimated model suggested that the target attribute had positive utility and price had negative utility. However, management reduced stated WTP by applying a calibration factor based on a small test-market comparison; this did not eliminate bias, but it acknowledged the gap between stated and revealed behaviour.

Next, VitalSip ran a four-week geo-randomised pilot. Matched stores in similar areas were assigned either the regular price of $3.00 or a promotional price of $2.50. Distribution and shelf position were held as constant as operationally possible. Sales were measured per store per week, and stockouts were recorded. The price test suggested that the lower price increased unit sales but reduced contribution margin per unit. Repeat purchase after the initial trial was more important than first-week volume.

A provisional bottom-up estimate used 80,000 reachable target consumers, an expected annual purchase frequency of 10 units, and an average realised price of $2.80:

The figure was then reduced in the base scenario to reflect distribution constraints and competitive response. Rather than treating $2.24 million as a forecast, management reported a range and specified the assumptions behind each scenario.

Outcome

VitalSip chose a regional launch with a controlled price test, retailer-level stock monitoring, and a retention objective. It did not use the survey’s purchase-intention percentage as a direct sales forecast. After launch, actual sales and repeat rates were fed into an updated demand model. The company also rejected a proposal to use sensitive health data for individual targeting without explicit consent.

Learning points

1. Market size depends on boundaries and assumptions.
2. Stated WTP is informative but not equivalent to observed payment.
3. Experiments can improve causal understanding of price and promotion.
4. Repeat purchase may matter more than trial volume.
5. A range of scenarios is more decision-useful than false precision.
6. Ethical data governance is part of professional demand analysis.

9. Limitations, Ethics, and Governance

Demand estimates are uncertain because customers may change behaviour, competitors may react, and measured variables may be incomplete. Sampling error is only one source of uncertainty; nonresponse, coverage error, social desirability, recall error, model misspecification, and selection bias may be more serious.

Managers should distinguish correlation from causation, report confidence intervals or credible ranges, document exclusions and transformations, and conduct robustness checks. Analysts should test whether results change under alternative definitions, samples, model specifications, and time periods.

Customer data create legal and ethical obligations. Organisations should collect only necessary data, obtain valid consent where required, explain purposes, restrict access, protect data, define retention periods, and provide mechanisms for correction or withdrawal. Sensitive attributes should not be used to discriminate in pricing, credit, access, or service. Automated segmentation and propensity scores can reproduce historical bias. Explainability, fairness testing, human review, and audit trails are therefore important.

AI can accelerate survey coding, forecasting, recommendation, and anomaly detection, but it does not remove the need for valid research design. Generative systems may fabricate sources, infer sensitive traits, or disguise uncertainty. A responsible workflow keeps a human accountable for variable definitions, sampling, model validation, interpretation, and decision approval.

10. Best-Practice Workflow

A defensible demand study normally follows these steps:

1. Define the decision and the market boundaries.
2. Formulate an economic theory of customer behaviour.
3. Identify the target population and relevant unit of analysis.
4. Audit available secondary and internal data.
5. Select complementary qualitative, survey, experimental, and econometric methods.
6. Pilot instruments and operational processes.
7. Pre-specify key outcomes, assumptions, and evaluation criteria.
8. Clean, document, and protect the data.
9. Estimate models with appropriate controls and diagnostics.
10. Validate against holdout data, experiments, or market outcomes.
11. Report uncertainty, limitations, and alternative scenarios.
12. Monitor results after implementation and update the model.

The most valuable output is not a single number. It is a transparent decision system linking assumptions, evidence, uncertainty, managerial action, and subsequent learning.

Chapter Summary

Demand estimation quantifies how much customers may buy and how quantity responds to price, income, related products, promotion, preferences, expectations, and market structure. Market measurement requires precise product, customer, geographic, time, and transaction boundaries. TAM, SAM, and SOM distinguish theoretical opportunity from reachable and realistically obtainable demand.

Secondary data are efficient but may not fit the decision. Qualitative research reveals motivations; surveys measure stated attitudes; experiments provide stronger evidence about causal effects; and econometric models quantify relationships while controlling for multiple variables. Price, income, cross-price, and advertising elasticities express percentage responsiveness, but their interpretation depends on model specification and data quality.

Willingness to pay can be studied through observed purchases, auctions, choice experiments, and contingent valuation. Customer analytics supports segmentation, retention, CLV, and forecasting, but automated methods require validation and ethical governance. Reliable managers triangulate methods, test assumptions, report uncertainty, protect customer data, and update estimates as market evidence accumulates.

Key Terms

Demand: Quantity customers are willing and able to purchase at specified conditions.

Market size: Total sales or consumption within a defined market boundary.

Market potential: Maximum realistic demand under specified assumptions.

Market share: A firm’s sales divided by total market sales.

TAM: Total Addressable Market.

SAM: Serviceable Available Market.

SOM: Serviceable Obtainable Market.

Willingness to pay: Maximum amount a customer is prepared to pay for an offering.

Elasticity: Percentage responsiveness of one variable to a percentage change in another.

Price elasticity: Responsiveness of quantity demanded to price.

Cross-price elasticity: Responsiveness of demand for one product to another product’s price.

Regression: A statistical method for estimating relationships among variables.

Endogeneity: A condition in which a predictor is related to the model’s error term.

Experiment: A design that manipulates a treatment and compares resulting outcomes.

Sampling error: Random difference between a sample estimate and the population value.

Nonresponse bias: Bias caused when participation differs systematically across population groups.

Customer lifetime value: Expected future contribution from a customer net of relevant acquisition cost.

Segmentation: Division of a market into groups with meaningful similarities.

Scenario analysis: Evaluation of outcomes under alternative assumptions.

Causal inference: Estimation of the effect of an intervention or treatment.

Review Questions

Short Answer Questions

1. Distinguish demand from sales.
2. What are TAM, SAM, and SOM?
3. Define price elasticity of demand.
4. Why can a large convenience sample still be biased?
5. State two ethical risks in customer analytics.

Long Answer Questions

1. Explain how market boundaries influence a market-size estimate.
2. Compare surveys, experiments, and econometric studies as methods of demand estimation.
3. Explain the meaning and managerial use of price, income, and cross-price elasticities.
4. Discuss the main sources of endogeneity in demand estimation and possible remedies.
5. Evaluate the usefulness and limitations of willingness-to-pay research.

Analytical / Application Questions

1. A firm estimates demand using price, advertising, and competitor price but ignores stockouts. Explain how this omission may distort the results and propose a better design.
2. Design a regional price experiment for a subscription service, specifying treatment, control, outcome measures, and threats to validity.
3. A market has 100,000 reachable buyers, an expected purchase frequency of six units annually, and an average price of $15. Calculate potential annual revenue and explain why the result should not automatically be treated as a forecast.

Multiple Choice Questions with Answers

Question 1

Which measure represents the realistically obtainable portion of a market for a firm?

A. TAM

B. SAM

C. SOM

D. Category penetration

Correct Answer: C. SOM

Question 2

If price elasticity of demand is −1.5, a 1% price increase is associated, approximately, with:

A. A 1.5% increase in quantity

B. A 1.5% decrease in quantity

C. A 0.67% decrease in quantity

D. No change in quantity

Correct Answer: B. A 1.5% decrease in quantity

Question 3

In a simple market-sizing calculation, 50,000 buyers purchase four units annually at $12 per unit. Estimated annual market value is:

A. $600,000

B. $1,200,000

C. $2,400,000

D. $24,000,000

Correct Answer: C. $2,400,000

Question 4

Which method generally provides the strongest basis for attributing an outcome difference to a treatment?

A. A randomised field experiment

B. A convenience survey

C. A focus group

D. A descriptive sales report

Correct Answer: A. A randomised field experiment

Question 5

A positive cross-price elasticity most commonly suggests that two products are:

A. Complements

B. Substitutes

C. Inferior goods

D. Unrelated by definition

Correct Answer: B. Substitutes

Question 6

In the sample-size formula for a population proportion, using p = 0.5 when the true proportion is unknown is conservative because it:

A. Minimises the estimated variance

B. Produces the maximum Bernoulli variance

C. Eliminates nonresponse bias

D. Guarantees a representative sample

Correct Answer: B. Produces the maximum Bernoulli variance

Question 7

A major identification concern in estimating price effects is that:

A. Price may be correlated with unobserved demand conditions

B. Price is always measured in percentages

C. Quantity cannot be observed

D. Elasticity has no units

Correct Answer: A. Price may be correlated with unobserved demand conditions

Question 8

Using the simplified CLV formula, if monthly margin is $20, retention is 0.95, discount rate is 0.01, and CAC is $150, CLV is approximately:

A. $16.67

B. $150.00

C. $166.67

D. $316.67

Correct Answer: C. $166.67

Question 9

Which statement about stated willingness to pay is most accurate?

A. It always equals the price customers will pay

B. It is irrelevant when purchase data exist

C. It can differ from observed behaviour because of hypothetical bias

D. It directly measures market share

Correct Answer: C. It can differ from observed behaviour because of hypothetical bias

Question 10

In a difference-in-differences design, the central identifying assumption is commonly called:

A. Constant elasticity

B. Parallel trends

C. Perfect competition

D. Independent pricing

Correct Answer: B. Parallel trends