Why this is hard to get right
The Analyst Who Needed Numbers That Could Survive a Board Room
Priya is a senior product manager at a Series B startup. Her company is entering the SMB HR tech space, and the board wants a market sizing slide in three weeks. She's done this before — kind of. Last time, she pulled a Gartner headline number, slapped "TAM: $12B" on a slide, and moved on. The board's first question was, "What's your SAM given your price point and target segment?" She didn't have a good answer.
This time, she's determined to do it right. She opens a conversation with an AI assistant and types: "Estimate the market size for our HR tech product in the US."
The response is five paragraphs of hedged generalities. The AI cites a vague "global HR tech market valued at $30B" figure with no segment breakdown, no pricing logic, and no methodology she can defend. It doesn't know her price point ($5 per user per month), her target segment (50–500 employee companies), or whether she needs 2024 or 2025 data. The output is, frankly, useless for a board deck.
Priya's core problem isn't the AI. It's that she gave it nothing to work with. Market sizing requires a scaffolding of decisions before the math begins: What's the pricing model? Which buyer segment? What penetration rate is defensible? Top-down or bottom-up methodology? Which sources count as credible?
When she structures her prompt to answer those questions upfront — specifying a bottom-up model, defining company-size tiers, naming her price range, requiring explicit assumptions and formulas, and demanding source citations — the output transforms. She gets a segmented table with TAM, SAM, and SOM figures. She gets a sensitivity range showing what happens if adoption rates shift by 20%. She gets a list of assumptions she can walk through with the board and a 120-word executive summary ready for the deck.
The difference wasn't the AI model she used. The difference was that a well-structured prompt replaced guesswork with methodology. It told the AI exactly what kind of analyst to be, what inputs to use, what outputs to deliver, and what standard of evidence to meet.
That's the gap most professionals hit with market sizing prompts. The analysis isn't hard for the AI — the setup is hard for the human. Knowing what to specify, in what order, at what level of detail, is the real skill. Once you nail the prompt structure, the AI becomes a credible analytical partner rather than a generic research assistant.
Common mistakes to avoid
Using a Top-Down Number Without Methodology
Citing a headline figure like "global HR tech is $30B" without a bottom-up check produces outputs that collapse under the first follow-up question. Stakeholders will ask how that number applies to your specific segment and price point. Always specify methodology — bottom-up (units x price) gives you defensible logic; top-down gives you a ceiling to calibrate against.
Omitting Pricing Model from the Prompt
Bottom-up market sizing is literally price times addressable units. If you don't tell the AI your pricing model — per seat, per company, tiered, freemium conversion — it invents one or skips the calculation entirely. Include your price range in the prompt. Even a rough range like $3–$8 per user per month anchors the entire model.
Leaving Segmentation Undefined
Asking for "the SMB market" gives the AI latitude to define SMB however it wants — sometimes 1–100 employees, sometimes 1–500. This breaks comparability across sources. Define your tiers explicitly (e.g., 10–99, 100–499, 500–999 employees) so the AI applies consistent cutoffs and you can cross-check figures against public datasets.
Not Requesting Explicit Assumptions
Market sizing always rests on penetration rates, adoption timelines, and addressable-share estimates. If the AI buries those assumptions or omits them, you can't audit, adjust, or defend the numbers. Explicitly require the AI to list every assumption with its rationale. This single step separates a slide filler from a defensible model.
Skipping the Sensitivity Range
A single-point estimate looks precise but is actually fragile. Boards and investors expect a range. Prompting for a base case plus a plus-or-minus scenario (e.g., 20% variance on penetration rate) forces the AI to show how sensitive the output is to key assumptions, which demonstrates analytical rigor and reduces credibility risk.
Forgetting to Set a Time Horizon
Market size figures shift year over year. If you don't anchor the analysis to a specific year, the AI mixes 2022 data with 2025 projections and the sourcing becomes internally inconsistent. Always specify the target year (e.g., 2025) so growth rates, population data, and pricing benchmarks align to the same period.
The transformation
Estimate the market size for our product in the US.
Act as a market analyst. Perform a TAM, SAM, SOM for B2B HR tech (employee engagement platforms) in the US for 2025. 1) Use a bottom-up model based on price tiers: $3–$7 user/month. 2) Segment by company size: 10–99, 100–999, 1,000+ employees. 3) State all assumptions (penetration, adoption rates) and show formulas. 4) Cite 3–5 reputable sources with links (e.g., BLS, Statista, Gartner). 5) Provide base case plus ±20% sensitivity. 6) Deliver: executive summary (120 words), table of TAM/SAM/SOM, assumptions list, risks/limitations. 7) Write in a concise, neutral tone.
Why this works
Methodology First
The After Prompt opens with "Use a bottom-up model based on price tiers: $3–$7 user/month." This forces the AI into a specific analytical framework before it touches any data. Methodology-first prompting eliminates the AI's tendency to default to a vague top-down industry figure. The output is now a calculation, not a citation.
Segmentation Anchors the Math
The prompt specifies three company-size bands: 10–99, 100–999, and 1,000+ employees. This directly maps to the "Segment by company size" instruction in Step 2. Explicit segmentation means the AI applies consistent definitions, enabling you to later cross-reference figures against Bureau of Labor Statistics or Census business data.
Transparency Requirement Builds Defensibility
Step 3 requires the AI to "state all assumptions and show formulas." This single instruction transforms the output from a black-box number into an auditable model. Requiring visible reasoning means you can stress-test, adjust penetration rates, and answer every stakeholder follow-up question with a documented rationale.
Source Mandate Raises Evidence Quality
The prompt requires "3–5 reputable sources with links (e.g., BLS, Statista, Gartner)." By naming expected source types in Step 4, you calibrate the AI toward authoritative data rather than hallucinated statistics. Named source categories also signal the quality bar — the AI understands that a blog post does not meet the standard a Gartner report does.
Structured Deliverables Eliminate Reformatting
Step 6 specifies four exact outputs: a 120-word executive summary, a TAM/SAM/SOM table, an assumptions list, and a risks section. Pre-defined deliverable formats mean the AI organizes its response for direct use in a board deck or investor memo, removing the rewrite step that kills momentum.
The framework behind the prompt
Market Sizing Fundamentals: TAM, SAM, and SOM
Market sizing is a structured analytical practice, not a data retrieval task. Understanding its foundations helps you write better prompts and evaluate outputs more critically.
The TAM/SAM/SOM Framework
The three-tier model was popularized in venture capital and product strategy as a way to communicate both the total opportunity and the realistic near-term addressable share in one visual. TAM (Total Addressable Market) is the revenue opportunity if you captured every eligible buyer globally or in your defined geography. SAM (Serviceable Addressable Market) filters TAM by the constraints of your product, price point, and go-to-market reach. SOM (Serviceable Obtainable Market) is the realistic share of SAM you can capture given competitive dynamics, sales capacity, and time horizon — typically 2–3 years.
Two Core Methodologies
Top-down analysis starts with an industry-level figure from a research firm (Gartner, IDC, Statista) and applies a series of filtering percentages to arrive at SAM and SOM. It's fast and anchored to recognized authority, but it's vulnerable to definition mismatches and opaque methodology.
Bottom-up analysis builds from unit economics: count the addressable buyers in your segment, multiply by your average contract value or annual spend, and sum across tiers. It's slower but produces a model you can stress-test assumption by assumption. Most credible investor presentations triangulate both methods and acknowledge the variance between them.
Penetration Rate Logic
Penetration rates are the most frequently abused assumption in market sizing. A new SaaS entrant achieving 5% of a $10B SAM in year three is almost always unrealistic, yet these numbers appear in pitch decks constantly. Realistic penetration benchmarks vary by category maturity: established software categories might see 15–30% penetration among target buyers; emerging categories rarely exceed 3–8% in the first three years.
Why AI Prompts Fail Without Structure
AI models approach unstructured market sizing requests by pattern-matching to the most common outputs they've seen — usually a headline industry number with vague growth projections. Without explicit instructions on methodology, segmentation, pricing model, and source requirements, the AI has no basis for choosing a defensible approach. Structured prompts that specify these inputs replace pattern-matching with genuine analytical reasoning.
Prompt variations
Act as a venture capital analyst. Perform a TAM, SAM, SOM analysis for on-demand legal services targeting independent contractors and freelancers in the United States for 2025.
- Use a bottom-up model based on a subscription price of $29–$79 per month per user.
- Segment the addressable base by worker type: gig platform workers, self-employed consultants, and creative freelancers.
- Apply realistic penetration rates for each segment and cite the source for each base population figure.
- Show all formulas and intermediate calculations.
- Provide a base case, a bull case (+30%), and a bear case (−30%) across all three segments.
- Cite 4–6 sources including at least one government dataset (BLS or Census) and one industry report.
- Deliver: a 150-word investment thesis summary, a segmented TAM/SAM/SOM table, and a risks section highlighting the top 3 market uncertainties.
Act as a startup market analyst preparing slides for a seed-round pitch. Size the US market for AI-powered customer onboarding software targeting mid-market SaaS companies (100–1,000 employees) in 2025.
- Use both a top-down approach (cite an industry report) and a bottom-up approach (company count x average contract value of $18,000–$40,000 per year).
- Define TAM as all mid-market SaaS companies in the US, SAM as those with active onboarding teams, and SOM as realistic 3-year capture at 2–5% penetration.
- List every assumption with a one-sentence rationale.
- Include a sensitivity table showing SOM at 1%, 3%, and 5% penetration rates.
- Cite 3 reputable sources.
- Format output as: executive summary (100 words), methodology note, TAM/SAM/SOM table, sensitivity table, and top 3 risks.
Act as a market analyst specializing in international expansion. Perform a TAM, SAM, SOM analysis for B2B fleet management software in Germany, France, and the United Kingdom for 2025.
- Build a bottom-up model using an average selling price of 120–200 EUR per vehicle per year.
- Segment by fleet size: 5–25 vehicles, 26–100 vehicles, and 100+ vehicles.
- Apply country-specific penetration rates based on current software adoption in each market. Cite your source for each country's fleet and commercial vehicle population.
- Convert all figures to both EUR and USD at current exchange rates.
- Compare the three countries side by side in a single table.
- State all assumptions and provide a 20% sensitivity range on penetration rates.
- Deliver: a 150-word regional executive summary, a country-comparison table, an assumptions list, and a 3-point risk section covering regulatory and FX factors.
Act as a corporate strategy analyst. Quickly size an adjacent market opportunity for a company currently selling expense management software to enterprises (1,000+ employees) that is considering expanding into travel management for the same buyer.
- Define TAM as total US enterprise spend on managed travel software in 2025.
- Use a top-down approach first (industry analyst data), then validate with a bottom-up check using estimated enterprise count x average contract value of $80,000–$200,000 per year.
- Estimate SAM by filtering to companies that already use integrated finance and HR platforms (indicate your assumption for this overlap percentage).
- Estimate SOM based on a realistic 18-month sales-cycle constraint and 3% market capture.
- Flag the top 3 competitive risks from incumbent vendors.
- Cite at least 3 sources.
- Deliver: a one-paragraph executive summary, a TAM/SAM/SOM table, a methodology note, and a competitive risk section.
When to use this prompt
Marketing Managers
Size the total addressable audience before planning campaigns and budget allocation for a new category.
Product Managers
Validate business cases by quantifying TAM/SAM/SOM for feature bets or new market entries.
Sales Leaders
Estimate territory potential and quota capacity with segmented market size by company tiers.
Founders and Strategy Leads
Build investor-ready TAM slides with transparent assumptions and sensitivity analysis.
Researchers and Analysts
Benchmark multiple markets using a consistent bottom-up framework and cited sources.
Pro tips
- 1
Specify your pricing model because it drives bottom-up accuracy and adoption assumptions.
- 2
Define segmentation that mirrors your go-to-market (industry, size, region) to align outputs with decisions.
- 3
Set a time horizon so growth rates and source data match the analysis year.
- 4
Request both top-down and bottom-up if you need triangulation and variance commentary.
Triangulation is the practice of sizing a market using two or more independent methods and then comparing the results. The gap between methods is as informative as the numbers themselves.
For a bottom-up model, you build from units: total addressable companies in the segment x average contract value = SAM revenue potential. For a top-down model, you start with a total industry figure from an analyst report and apply a realistic share percentage. When both methods land within 15–25% of each other, you have a credible anchor range. When they diverge by 50% or more, that signals either a definition mismatch or a data quality problem — both of which you need to surface before a board presentation.
Scenario modeling adds a second dimension: what if key assumptions are wrong? The After Prompt requests a ±20% sensitivity range, but advanced analysts build three full scenarios:
- Bear case: Lower penetration rate, longer sales cycle, narrower segment definition
- Base case: Most likely adoption curve, current pricing, mid-range segment estimates
- Bull case: Accelerated adoption, expanded segment, higher price realization
When prompting for scenarios, give the AI the specific variable to stress-test (e.g., penetration rate, not "general conditions") and the range to apply. Vague scenario requests produce vague outputs.
For boards and investor memos, present all three scenarios in a single table with the key assumption driver labeled for each column. This approach signals analytical maturity and preempts the most common skeptic questions.
The quality of your market size estimate is only as good as your source data. Here is a tiered framework for evaluating sources:
Tier 1 — Government and Official Data (Highest Credibility)
- US Bureau of Labor Statistics (BLS): employer counts, workforce data, industry employment
- US Census Bureau: business counts by size, industry, geography
- Eurostat, ONS (UK), Destatis (Germany): equivalents for international work
Tier 2 — Established Research Firms (High Credibility, Paywalled)
- Gartner, IDC, Forrester: IT and enterprise software markets
- IBISWorld, Statista: cross-industry sizing and trend data
- PitchBook, CB Insights: venture-backed market intelligence
Tier 3 — Industry Associations and Trade Reports (Moderate Credibility)
- Useful for niche verticals, but methodology is often opaque
- Always check whether figures are self-reported by member companies
Tier 4 — News Coverage and Press Releases (Low Credibility for Sizing)
- Useful for competitive intelligence, not for base market figures
- AI models frequently cite this tier when better sources are unavailable
When prompting the AI, specify at least one Tier 1 and one Tier 2 source by name. This raises the floor on output quality. Always verify any specific statistic the AI attributes to a named source — hallucinated citations are common, especially for precise figures.
The core prompt structure (methodology, segmentation, pricing, assumptions, sensitivity) applies across industries, but the inputs change significantly by sector. Here are the key adaptations by industry type:
SaaS and Software Markets
- Use per-seat or per-company ARR as the pricing unit
- Segment by company size (employees or revenue bands)
- Source company counts from LinkedIn data or Census SUSB tables
- Penetration rates typically range from 2–15% for established categories
Consumer Markets
- Use household income bands or demographic segments
- Annual spend per user is the pricing unit, not contract value
- Source population data from Census ACS (American Community Survey)
- Consider geographic density as a SAM filter (urban vs. rural adoption)
Healthcare and Life Sciences
- Segment by provider type (hospital systems, independent practices, ASCs)
- Use procedure volume or patient population as the unit base
- Reference CMS data, AHRQ, and state health department reports
- Regulatory constraints on market entry often create a practical ceiling below SAM
Industrial and Manufacturing
- Unit count (machines, facilities, fleet vehicles) is a more reliable base than revenue
- Maintenance cycle and replacement frequency drive adoption timelines
- Census Annual Survey of Manufactures and industry association data are primary sources
In every case, tell the AI which segmentation variable to use as the primary base unit. That decision shapes the entire model.
When not to use this prompt
This prompt structure assumes a defined product, price point, and target segment. If you don't yet have those inputs, the analysis produces false precision. In that case, use an exploratory prompt first to define those parameters before sizing the market.
Avoid using AI-generated market sizing as a sole source for legal filings, SEC disclosures, or financial instruments. The citation quality and source verification required for those contexts goes beyond what any AI output should provide without expert review.
This approach is also less effective for:
- Hyper-local markets (a single city or zip code) where national datasets don't segment finely enough
- Pre-category markets where no industry reports exist and proxy market assumptions carry very high uncertainty
- Regulated industries (pharmaceuticals, financial services) where addressable market depends heavily on compliance pathways that AI cannot reliably model
In those cases, supplement the AI output with primary research, expert interviews, or a commissioned analyst report. Use the AI-generated sizing as a starting framework, not a final answer.
Troubleshooting
The AI returns a single large market number with no segmentation or methodology
Your prompt is missing structural instructions. Add these three lines explicitly:
- "Use a bottom-up approach: show company count x average contract value for each segment"
- "List each segment separately in a table"
- "Show all intermediate calculations"
Alternatively, break the task into two prompts: first ask for the methodology and segmentation framework, then ask it to populate the numbers.
The AI cites sources that don't exist or provides statistics I can't verify
AI models frequently hallucinate precise statistics and report titles. Mitigate this with two prompt changes:
- Name specific source categories you trust: "Cite only from BLS, Statista, Gartner, or Census Bureau"
- Add: "If you cannot attribute a figure to a named source, flag it as an estimate"
Always verify any specific number against the original source before including it in a client-facing document or board deck.
TAM, SAM, and SOM figures seem inconsistent or don't nest properly
The AI likely defined each layer independently instead of as nested subsets. Restate the definitions explicitly in your prompt:
- "TAM = total market if you captured every eligible buyer"
- "SAM = subset of TAM filtered by your segment, geography, and price point"
- "SOM = realistic 3-year capture of SAM at X% penetration"
Ask the AI to show SOM as a percentage of SAM, and SAM as a percentage of TAM, so the nesting is transparent.
The executive summary is too long and includes methodology details
Specify an exact word count and scope in the prompt. The After Prompt uses "executive summary (120 words)" — that constraint works. Strengthen it by adding scope guidance: "Write a 100–130 word executive summary covering only the key TAM, SAM, and SOM figures, the single most important assumption, and the primary market risk. Do not include methodology details in the summary."
The sensitivity analysis shows only one scenario instead of a range
The AI likely interpreted "sensitivity" as a note rather than a calculation. Be explicit about format:
- "Create a sensitivity table with three columns: bear case (−20% on penetration), base case, and bull case (+20% on penetration)"
- "Show TAM, SAM, and SOM figures in each column"
If you want sensitivity on multiple variables (price and penetration), state each variable separately and ask for a matrix.
How to measure success
How to Evaluate Your Market Sizing Output
A strong AI-generated market size analysis should meet these standards before you use it in a deck or memo:
Structural Completeness
- TAM, SAM, and SOM are defined as nested subsets, not independent numbers
- Each figure includes the formula or calculation path that produced it
- All assumptions are listed explicitly with a rationale
Methodological Soundness
- The bottom-up model includes: addressable unit count, price per unit, and penetration rate — all stated separately
- A sensitivity range is provided, not just a single-point estimate
- Top-down and bottom-up figures are within a defensible variance range (typically under 30%)
Evidence Quality
- Every base statistic is attributed to a named source
- At least one government dataset (BLS, Census) anchors the population figures
- No figure appears without a source label
Presentation Readiness
- Executive summary is under 150 words and free of methodology detail
- Output includes a formatted table suitable for a slide
- Risks and limitations section names at least two specific uncertainties
Now try it on something of your own
Reading about the framework is one thing. Watching it sharpen your own prompt is another — takes 90 seconds, no signup.
Build a defensible TAM/SAM/SOM model with cited sources, visible assumptions, and a sensitivity range — ready for your next board deck.
Try one of these
Frequently asked questions
Use both when possible, but prioritize bottom-up for credibility. Top-down (industry report x market share %) gives you a ceiling and a sanity check. Bottom-up (addressable units x price) gives you a model you can defend assumption by assumption. In your prompt, specify which method you want primary and ask for the other as a triangulation check. When they diverge by more than 30%, that gap is itself a useful finding worth flagging.
Name the source types you expect in your prompt. The After Prompt example specifies "BLS, Statista, Gartner" explicitly. This signals the quality bar. You should also instruct the AI to include URLs or report titles and flag any figure it cannot attribute to a named source. Always verify citations independently — AI models can hallucinate specific report titles or statistics even when the source category is real.
For thin-data markets, adjust your prompt to:
- Use proxy markets (e.g., "use the adjacent project management software market as a proxy")
- Build from first principles (company count x department headcount x problem frequency)
- Accept wider sensitivity ranges (±40% instead of ±20%)
- Explicitly acknowledge data gaps in the output
The key is building a logic chain the reader can follow, even if the inputs are estimates.
TAM is total market demand if you captured every buyer. SAM is the portion you can realistically serve given your product, geography, and price point. SOM is your realistic capture within 2–3 years. Most AI outputs conflate these unless you define them explicitly in your prompt. Use the After Prompt structure: define each layer's filter criteria (segment, penetration rate, time horizon) so the AI calculates them as nested subsets, not independent estimates.
This usually means your prompt lacked structure. Add three explicit requirements:
- "Show all formulas and intermediate calculations"
- "List every assumption with a one-sentence rationale"
- "Provide a sensitivity range of ±20% on penetration rate"
These three lines force the AI out of summary mode and into analytical mode. If outputs are still thin, break the prompt into stages: first ask for the framework, then ask it to populate each step.
Yes, with two adjustments. First, specify each country individually and ask the AI to source country-specific population or business-count data (national statistics offices are better than US-centric databases). Second, address currency explicitly — ask the AI to work in local currency and provide a USD conversion. See the Geographic Expansion variation above for a complete example covering Germany, France, and the UK.
The After Prompt on this page is 7 numbered steps and runs about 130 words — that's a good target range. Shorter prompts produce shallower analysis; longer prompts risk confusing the AI with conflicting instructions. Focus on five critical elements: methodology, segmentation, pricing, required assumptions, and output format. Those five elements alone will produce a dramatically better result than a single-sentence request.
Specify your exact deliverables in the prompt, as Step 6 of the After Prompt does: "executive summary (120 words), table of TAM/SAM/SOM, assumptions list, risks/limitations." Pre-defining the output structure means the AI organizes its response in the order your deck needs. Ask for tables in markdown or specify that numbers should be formatted in millions or billions to match your slide standards.