Market Research skill for Codex

A market estimate is credible when a skeptical reader can see every input, where it came from, and how much the answer moves if the input is wrong. The bottom-up number (who buys, how many, at what price) is the estimate; top-down figures are the sanity check.

This produces research for planning, not investment advice. Figures used in fundraising materials or filings should be reviewed by someone accountable for them.

Definitions (state them in the output)

Term	Definition	Built from
TAM	Annual revenue if every potential customer for this problem bought your type of solution	All target units x annual price
SAM	The part of TAM your product and go-to-market can actually serve now (geography, segment, channel, language, compliance)	TAM filtered by current constraints
SOM	The share of SAM you can realistically win in a stated period (usually 3-5 years)	SAM x achievable share, grounded in comparables or sales capacity

TAM is about a problem and a buyer, not about an industry label. “The $600B software market” is never a TAM.

Workflow

- [ ] 1. Define the buyer, the unit and the price metric
- [ ] 2. Bottom-up: count units and price per unit
- [ ] 3. Apply SAM filters, each with a source
- [ ] 4. Derive SOM from capacity or comparables
- [ ] 5. Top-down cross-check; reconcile any gap > 2x
- [ ] 6. Sensitivity: low / base / high on the 3 biggest drivers
- [ ] 7. Segment the SAM and rank segments
- [ ] 8. Write the assumptions log and the summary

1. Unit and price

Pick the unit that matches how the product is bought: companies, seats, developers, locations, vehicles, transactions. Pick the price metric customers actually pay (per seat per year, per location per month, % of GMV). Mixing units (seats in one step, companies in the next) is the most common error in market models.

2. Bottom-up count

Target unit:  US dental practices with 3+ chairs
Count:        ~X establishments (Census County Business Patterns, NAICS 621210,
              filter by employee size class as a proxy for chairs)
Price:        $300/location/month from 4 competitor pricing pages (median)
TAM (US):     X x $3,600/year

Good public sources (most need network access):

Data	Source
US establishments by industry and size	Census County Business Patterns, Statistics of US Businesses (by NAICS)
Occupations and headcounts	BLS Occupational Employment and Wage Statistics
EU businesses by sector and size	Eurostat structural business statistics
Developer populations	Stack Overflow Developer Survey, GitHub Octoverse, SlashData (note each one’s method)
Public competitors’ revenue, customers, segments	10-K / 20-F filings on SEC EDGAR, annual reports, investor decks
Pricing	Competitor pricing pages (archive them with the date), G2 listings
Private competitor signals	Press releases, job postings, funding announcements (treat as weak)

Analyst reports (Gartner, IDC, Statista) are often paywalled and use opaque methods; cite them only as a top-down cross-check, and only with the figure, year and definition.

3. SAM filters

Apply one filter at a time, each with a source or a labelled assumption:

TAM (US + EU)                              100%
x served regions (US only at launch)        58%   (count from step 2)
x segment fit (practices on cloud PMS)      40%   ASSUMPTION: vendor survey, n=300
x compliance (HIPAA-ready, yes)            100%
= SAM

4. SOM

Ground share in something observable, not a round “1% of the market”:

• Capacity: reps x deals per rep per year x ramp x win rate x years.
• Comparables: what share did similar companies reach in a similar number of years? (Public filings give customer counts.)
• Funnel: traffic or community size x conversion rates from your own data.

Use the lowest of the defensible methods as the base case.

5. Top-down cross-check

Find a total spend figure for the category (analyst report, sum of public competitors’ revenue divided by their estimated share). If bottom-up and top-down differ by more than ~2x, find the reason before publishing: wrong unit count, price too high, or the top-down figure includes adjacent products.

6. Sensitivity

Vary the three inputs with the most uncertainty and show the range:

Driver	Low	Base	High	Source of range
Price / unit / year	$2,400	$3,600	$4,800	competitor quartiles
Segment fit %	25%	40%	55%	survey CI
Win share by year 5	1%	3%	5%	comparables

Build it as a spreadsheet with inputs in one block and formulas referencing them, never hard-coded numbers in formulas. Hand formula work to $excel-formulas or $xlsx-spreadsheets when building the file.

7. Segmentation

Segment the SAM on variables that change buying behaviour: size, vertical, job-to-be-done, current solution, regulatory need, budget owner. For each segment estimate size, pain intensity (evidence from interviews, reviews, support tickets), willingness to pay, reachability (channels) and competition. Rank segments by attractiveness x ability to win.

Gotchas

• Revenue TAM vs user TAM: “50M developers” is not a market until multiplied by what each would pay.
• Double counting: seats and companies, or counting both a platform and its add-ons.
• Using list price when most deals are discounted; use realised or median plan price.
• Survey figures from a vendor’s own users skew toward early adopters.
• Currency and year: state both; convert with a dated rate.
• CAGR quotes from press releases (“growing 24% to 2032”) usually have no method. Use as context only.
• A tiny SAM is a useful finding. Do not inflate filters to reach a fundable number.

Verification

• Recompute TAM x filters = SAM and SAM x share = SOM in the sheet; units match at each step.
• Each input has a source with date, or is labelled ASSUMPTION with reasoning.
• Bottom-up and top-down are within ~2x or the gap is explained.
• The sensitivity table shows which single input moves the answer most.

Output template

# Market sizing: <product> for <buyer>
**Unit:** <unit> | **Price metric:** <metric> | **Currency/year:** USD 2026

| | Base | Range (low-high) |
|---|---|---|
| TAM | $X | $A-$B |
| SAM | $Y | ... |
| SOM (year 5) | $Z | ... |

**Method:** bottom-up count x price; filters below; top-down check from <source>.
**Biggest uncertainty:** <driver> moves SOM from $a to $b.

## Assumptions log
| # | Input | Value | Source / reasoning | Confidence |

## Segments (ranked)
| Segment | Size | Pain evidence | WTP | Reach | Competition | Rank |