Demand Sizing and Opportunity Economics:
build decision-grade forecasts, ranges and economics

How large, probable and valuable is the opportunity?

A forecast is a conditional argument, not a fact about the future. Good opportunity economics makes assumptions visible, uses independent methods, expresses ranges and links uncertainty to staged commitment.

TAM/SAM/SOM labels are insufficient unless boundaries, units, timing, price, adoption, capacity and competition are explicit. The objective is not the largest defensible number; it is the smallest decision-useful model.

1. Define the quantity

Before calculating, specify:

• unit: users, accounts, transactions, volume, revenue, gross profit or cash;
• customer, use case and geography;
• time horizon and price basis;
• gross demand versus accessible demand;
• stock, annual flow or cumulative adoption;
• nominal or real values and currency;
• inclusions, exclusions and substitutes.

2. Triangulate size

Use at least two independent approaches:

• Top-down: official category or population totals narrowed by valid filters.
• Bottom-up: eligible accounts × incidence × frequency × units × realised price.
• Supply-side: capacity, shipments, channel throughput or competitor revenues.
• Analogue: adoption curves from comparable markets, adjusted for differences.

Reconciliation matters more than averaging. Explain why estimates differ; disagreement often reveals a boundary or adoption assumption.

3. Forecast drivers

Model demand as drivers rather than a trend line: eligible population, trigger rate, awareness, access, conversion, frequency, retention, price, substitution and constraints. Separate baseline, upside and downside scenarios. Identify leading indicators and thresholds.

Official sources are valuable anchors but have coverage and revision limits. WTO datasets supply comparable trade and market-access series across many economies,¹ while Hong Kong C&SD publishes population, expenditure, retail, trade and business indicators with definitions and reference periods.² Company decisions may require more granular primary evidence.

4. Opportunity economics

Translate demand into:

• realised revenue, discount and channel share;
• contribution margin and cost-to-serve;
• acquisition, implementation and retention cost;
• working capital and capacity investment;
• time to break-even and cash exposure;
• cannibalisation and opportunity cost;
• expected value under uncertainty.

Use probability-weighted cases only when probabilities are reasoned and auditable. For highly uncertain opportunities, calculate the value of learning and staged options instead of forcing a point forecast.

5. Calibration and back-testing

Archive each forecast version. Record assumptions, source dates and confidence. Compare forecasts with outcomes at fixed intervals and diagnose error into boundary, driver, timing, execution and external-shock components. Avoid quietly replacing an old forecast.

6. AI-native application

AI can extract comparable series, maintain model documentation, detect changed assumptions and run scenarios. It cannot repair incompatible definitions or establish causality from correlation. Require deterministic calculations outside the language model, source-level traceability and human approval of assumptions.

7. APAC and Hong Kong application

Use local currency, price, channel, policy and adoption drivers. Do not scale one APAC market by population alone. Hong Kong’s compact geography and cross-border flows can make resident population a poor denominator for categories affected by visitors, commuters, trade or regional delivery.

8. QA checklist

• boundaries and units explicit;
• no top-down and bottom-up double counting;
• price and volume separated;
• source reference periods aligned;
• uncertainty and sensitivity visible;
• capacity and competition included;
• totals reproduce from documented inputs;
• forecast can be back-tested.

9. Decision playbook

Build the model in layers. First establish an observed baseline. Next isolate structural drivers from temporary effects. Then model adoption and competitive capture. Finally connect volume to cash economics. Keep inputs, formulas and judgments separate so reviewers can change assumptions without rewriting the model.

For each critical input record a low, central and high case, source, date, rationale and elasticity of the decision. Focus new research on assumptions that are both uncertain and economically sensitive. This is more efficient than improving every estimate equally.

Use rolling forecasts for operating decisions and scenario ranges for strategic commitments. A rolling forecast updates expected results; a scenario tests a different causal world. Do not collapse them into one “best case / worst case” spreadsheet.

10. Failure modes

Typical errors include population multiplied by an arbitrary penetration rate, mixing annual flow with installed base, applying list price instead of realised price, treating demand as unconstrained supply, ignoring churn and cannibalisation, and using CAGR beyond its evidential horizon. Decision makers should also see when a number is model output rather than observed statistic.

Evaluation should be proportionate to uncertainty and consequence, as the Magenta Book recommends.³ A reversible pilot may need a range and trigger; an irreversible capital commitment needs deeper validation and independent model review.

Sources

1World Trade Organization, *WTO Stats Portal*. https://data.wto.org/dataset/wto_sts

2Hong Kong Census and Statistics Department, *Statistics and You* and linked datasets. https://www.censtatd.gov.hk/en/page_235.html

3HM Treasury and Evaluation Task Force, *Magenta Book* (2026), on proportionate evaluation, uncertainty and early learning. https://www.gov.uk/government/publications/the-magenta-book/magenta-book-central-government-guidance-on-evaluation-html