Estimating TAM, SAM, and SOM with AI Guidance
Estimating market size is one of the most critical, yet error-prone, exercises for founders, product managers, and marketers. Without a dedicated research team, it is tempting to ask a Large Language Model (LLM) to simply “calculate the market size” for a new product.
However, AI is a synthesis engine, not an oracle. If you treat AI as a search engine for market sizing, you will likely encounter outdated data, hallucinated statistics, or fundamentally flawed logic. To use AI effectively, you must understand the mechanics of market sizing, guide the AI to build structured models, and rigorously verify its assumptions.
The TAM, SAM, SOM Funnel and the “1% Fallacy”
Before prompting an AI, you must define the exact tier of the market you are analyzing. Market sizing is conventionally broken down into three tiers:
- TAM (Total Addressable Market): The total global revenue potential if your product achieved 100% market share. TAM represents the absolute ceiling of the market.
- SAM (Serviceable Available Market): The portion of the TAM that you can actually target, constrained by geography, regulatory environments, language, or specific product fit.
- SOM (Serviceable Obtainable Market): The realistic, short-term market share you can capture given your current resources, sales channels, and competitive landscape.
A common pitfall that AI will happily validate if unchecked is the “1% Fallacy.” This is the assumption that capturing just 1% of a massive $10 billion TAM is an easy, low-risk path to a $100 million business. Investors and seasoned executives routinely reject this top-down assumption because it ignores the friction of customer acquisition and competitive defense. To avoid this, your AI-assisted models must eventually anchor in bottom-up realities.
Choosing a Sizing Methodology
There are three primary ways to model a market. AI plays a distinctly different role in each.
| Sizing Method | Best For | AI’s Role | Key Limitation | Human Verification Needed |
|---|---|---|---|---|
| Top-Down | Early-stage macro validation, investor pitch decks. | Synthesizing industry reports (e.g., Gartner, Statista) to extract broad TAM figures. | Highly prone to the “1% Fallacy” and ignores unit economics. | Cross-referencing AI summaries against original analyst reports. |
| Bottom-Up | Operational planning, setting sales targets, SOM calculation. | Structuring the mathematical model and calculating scenarios based on user-provided unit economics. | Cannot reliably guess niche conversion rates or specialized pricing tiers. | Validating average revenue per user (ARPU) and realistic conversion rates. |
| Value-Theory | Disruptive products creating new markets with no historical data. | Drafting ROI scenarios by analyzing customer pain points and estimating time/money saved. | Relies heavily on assumptions about customer willingness to pay. | Primary research (interviews/surveys) to confirm the estimated value translates to actual spend. |
1. Top-Down Sizing: Proceed with Caution
Top-down sizing starts with broad macro data and applies filters to narrow it down. AI models like Claude or ChatGPT are highly efficient at pulling top-down data during the initial research phase. You can upload industry reports and ask the AI to extract specific market segments or CAGR (Compound Annual Growth Rate) projections.
The $30 Billion AI Error:
Relying solely on AI for top-down sizing without checking its logic can be disastrous. In a real-world case study by the consulting firm Strategex, an AI model was asked to calculate the market size for e-commerce software. The AI found the average software fee and multiplied it by the total number of global merchants.
The AI treated all merchants equally, ignoring the massive pricing disparity between enterprise giants like Amazon and small “mom-and-pop” Shopify stores. This fundamental misunderstanding of market stratification led to a TAM estimate that was off by over $30 billion.
2. Bottom-Up Sizing: The Rigorous Approach
Bottom-up sizing builds the market from your unit economics: price per user multiplied by realistic target customer segments. AI struggles to generate bottom-up models from scratch because it lacks your proprietary business context. Instead, you must use AI as a calculation and structuring assistant.
Example: Prompting for a Bottom-Up Model
Do not ask: “What is the SOM for a dental scheduling app?”
Instead, use a structured prompt that provides the variables:
> “Act as a financial analyst. Build a bottom-up SOM model for a B2B SaaS scheduling tool targeting independent dental clinics in California.
>
> Use the following unit economics:
> * Subscription: $200/month per clinic.
> * Total addressable clinics in CA: 30,000.
> * Year 1 reach: We can market to 10% of these clinics.
> * Conversion rate: Assume a 5% conversion rate on the reached audience.
>
> Output a step-by-step calculation table showing the Year 1 SOM in ARR (Annual Recurring Revenue), and outline the three biggest risk factors to these assumptions.”
By providing the unit economics, you force the AI to act as a rigorous modeling engine rather than a hallucination engine.
3. Value-Theory Sizing: Estimating the Unseen
For entirely new categories, historical data does not exist. Value-theory sizing estimates the financial value your product creates (e.g., hours saved, revenue increased) and calculates how much of that value you can capture through pricing.
AI is excellent at brainstorming value-theory variables. You can prompt it with: “List all the indirect costs associated with manual data entry in a mid-sized logistics firm, and estimate the hourly financial impact.” While the AI’s dollar estimates will be synthetic and require human validation, the framework it provides ensures you do not miss hidden cost-savings you can use to justify your market size.
Critical Caveats for AI Market Sizing
As you integrate AI into your market research workflow, you must guard against three systemic risks:
- Outdated Training Data: Foundational AI models often rely on training data that is 12 to 18 months old. In fast-moving tech sectors or post-disruption markets (like commercial real estate post-2020), this lag renders AI-generated TAM figures dangerously obsolete. Always use tools with live web-browsing capabilities to pull the latest figures, or upload current PDFs directly.
- The “Black Box” of Synthetic Data: Using AI to generate synthetic customer profiles or estimate purchase intent can lead to catastrophic product failures if the underlying data generation cannot be verified. If an AI claims “40% of users will upgrade to premium,” you must ask it to cite the proxy data or historical precedent it used to reach that number. If it cannot, discard the metric.
- Information vs. Judgment: AI can process thousands of documents instantly, but markets are ultimately shaped by human incentives, B2B relationships, and complex procurement negotiations that do not exist in public datasets. AI provides the information; human judgment must verify the reality. Relying solely on AI-generated top-down estimates without a bottom-up reality check remains a primary reason why startups fail due to misjudging market size.