How to size a market with AI

Define the market before counting it

State the product or service, buyer, use case, geography, time period, unit, and inclusion rule. Distinguish total addressable market, serviceable market, and an attainable planning scenario only if each has a concrete boundary. Do not use labels as substitutes for definitions.

For example, “AI training market” could mean software, instructor-led services, internal programme spend, or all workforce learning. A useful definition might be: “Annual external spend by US employers with paid staff on applied AI-workflow training for knowledge workers, excluding general software subscriptions.” The definition is synthetic and still requires data.

The US Census Bureau’s County Business Patterns data provides annual establishment, employment, and payroll statistics by detailed NAICS industry and geography, with disclosure and comparability notes. The Bureau of Labor Statistics QCEW publishes establishment, employment, and wage data by industry and geography. These are official denominator sources; neither directly measures demand for the synthetic category.

Build a top-down estimate

Start with a defensible population, then apply filters and rates supported by evidence or labeled assumptions. Keep units visible:

| Step | Base value | Range | Source or status | |---|---:|---:|---| | In-scope employer establishments | 120,000 | Fixed for selected codes | Official dataset, chosen vintage | | Share with target knowledge-work roles | 35% | 25–45% | Assumption pending validation | | Annual buyer rate | 8% | 3–15% | Hypothesis | | Average annual spend | $18,000 | $8,000–$30,000 | Interview hypothesis |

The synthetic base calculation is 120,000 × 35% × 8% × $18,000, or about $60.5 million. The arithmetic is exact; the market is not. Report the range and sensitivity, not false precision.

Build a bottom-up estimate

Model plausible buyer segments, number of accounts, adoption rate, cohort volume, frequency, and price or contract value. Keep price separate from willingness to pay and observed transactions separate from list prices.

A bottom-up route might segment large professional-services firms, technology suppliers, and AI-training providers, then estimate eligible accounts and annual programme volume. If it produces $22 million to $95 million, compare that range with the top-down result. Do not average contradictory models until category and unit differences are resolved.

Ask AI to generate reconciliation questions: Are both models counting employers or locations? Does one include internal labor? Are annual and multi-year values mixed? Open and inspect all source tables and definitions.

Triangulate and expose category mismatch

Use a third route when possible, such as disclosed supplier revenue, procurement records, or a carefully bounded survey. Each source should measure a different part of the question rather than repeating the same estimate.

A failure example multiplies all business establishments by a software price while the defined buyer is only employers running formal AI academies. The numerator and denominator describe different categories. Repair it by narrowing the establishment codes and role filters, then validating the buyer rate separately.

Track dataset vintage, suppressed cells, methodology changes, and geographic coverage. A newer source is not automatically more comparable.

Verify the market-sizing model

Before circulation, confirm:

The artifact is the model plus definition sheet, source map, calculation table, reconciliation log, sensitivity analysis, and refresh schedule.

• category, buyer, use case, geography, period, and unit are explicit;
• top-down and bottom-up methods are independent;
• every input is sourced or labeled assumption;
• data vintages and classifications are compatible;
• suppressed or missing values are handled visibly;
• internal labor, prices, and spend are not mixed;
• sensitivity identifies switching assumptions;
• reconciliation explains model differences; and
• output is a range with limitations and refresh triggers.

Sources and claim map

The market definition, establishments, rates, prices, and estimates are synthetic. This resource does not establish demand, revenue opportunity, attainable share, or validated market size.

• US Census Bureau, County Business Patterns APIs: supports establishment, employment, payroll, industry, and geography denominator data with stated limitations.
• US Bureau of Labor Statistics, Quarterly Census of Employment and Wages: supports industry and geographic establishment, employment, and wage data in its published scope.

Sources: [1] [2]