# Market Research

> Produces sized, sourced, decision-ready market and competitive analysis rather than a collection of facts. Builds every market size twice, top-down and bottom-up, reconciles the two, separates the total market from the serviceable slice from what could realistically be won, maps competitors on evidence with pricing and dated strategic moves, and enforces the standard that every number carries a source and a year in the same line and that nothing is fabricated. Use this skill whenever someone asks to research a market, size a market, work out the total addressable market, analyse competitors, evaluate a new segment or geography, assess an acquisition or partnership target's market, prepare the market section of a board deck or fundraising material, or says things like "what does the market look like", "who else does this", "is this worth entering", "how big could this be", or "look into this space". Trigger for any request to gather external business information that will feed a decision, even when the ask is

- Skill: `ingridleiria/market-research` (Agent Skill)
- Install (CLI): `npx skillmds@latest add ingridleiria/market-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ingridleiria/market-research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: ingridleiria (https://skillmd.com/u/ingridleiria)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ingridleiria/market-research

---


# Market Research

Market research fails in two directions and both are expensive. The first is the fact collection: forty slides of context, five charts from published reports, a competitor grid, and no answer to the question the work was commissioned to settle. It reads as thorough and it changes nothing, because at no point did anyone write down what finding would have led to a different decision.

The second is worse because it survives longer. A single market size appears, sourced to a headline figure from a report nobody has read, and it becomes the number in the board pack, then the number in the fundraising deck, then the number in the plan. The first serious question about how it was built collapses the whole case, usually in front of the audience that mattered most. A market entry justified on a number that turns out to include hardware, adjacent categories, and a geography you cannot sell into costs a year of a team's time and the credibility of whoever presented it.

This skill exists to make research that changes a decision and survives being questioned. Before gathering anything, write down the decision the research feeds and the finding that would change it. Research that cannot change the decision is decoration.

## When to use this, and when not to

Use it for sizing a market or segment, deciding whether to enter a geography or a vertical, mapping who else serves a set of buyers and on what basis they win, assessing the market a partnership or acquisition target sits in, and building the market section of board or fundraising material.

Do not use it to brief someone on a single company or person before a meeting, which is `external-insights`. Do not use it to produce the seller-facing one-page comparison used in a live deal, which is `competitive-battlecard`. Do not use it to structure the argument the research feeds, which is `structured-problem-solving`, or to carry the resulting choice to a decider, which is `decision-memo`. Do not use it for evidence about your own customers, which comes from `customer-interview-synthesis`, or for your own pipeline and revenue, which is `pipeline-deep-dive` and `revenue-analysis-workbook`. Market research answers questions about the outside; those skills answer questions about the inside, and blending them without labelling which is which is how an internal conversion rate becomes a market assumption.

## What you need before starting

**The decision and the threshold that decides it.** Not just "should we enter", but the number above which the answer is yes. Missing: propose one in a sentence and get it confirmed before starting. Without a threshold, every result is interesting and none is conclusive.

**The deadline and what it is attached to.** A board meeting, a planning cycle, a term sheet. Missing: assume the next leadership meeting and say so, because the depth of the sizing is set by the time available and should be chosen deliberately rather than discovered late.

**The definition of the market, in units.** Who is the buyer, what is the unit of purchase, what counts as inside the boundary and what does not. Missing: write the definition yourself, state it in the first line of the output, and flag it as an assumption. Most disagreements about a market size are actually disagreements about the boundary.

**Your own commercial numbers, for the bottom-up leg.** Average contract value, win rate, sales cycle, the cost of serving a customer. Missing: use a range from the closest comparable segment you do serve and label it. Never build a bottom-up size from a target you have not yet achieved anywhere.

**The constraints already fixed.** A budget cap on entry, a commitment to a partner, a geography the board has ruled out. Missing: ask. Sizing a market you cannot serve is the most common way this work is wasted.

**Access to sources.** Company filings, regulator registers, national statistics, pricing pages, published reports, and a search tool. A connected enrichment or company-data service is useful for firmographics, funding, and headcount where one is available. Missing: use public primary sources, which are enough for most of this work, and record explicitly where a paid dataset would have improved confidence. Never make the method depend on a subscription.

## The method

### 1. Frame the decision and write the kill number

State the decision, the deadline, and the two or three questions that actually matter. A typical frame: "Should we open a dedicated team for segment X next year. Depends on: is the serviceable market above fifteen million, is it growing, and can we win against the incumbents on something durable."

Then write the kill number: the result that would end the question. Fifteen million, in that example. Writing it before the research starts is what stops the size being quietly negotiated upward once the effort has been spent.

### 2. Define the market in units before sizing it

Name the buying unit, the unit of purchase, and the boundary. A dental practice is not the same buying unit as a dental group with forty sites, and a market sized in practices will be wrong by the ratio between them.

The rule: size in units you could sell to, then multiply by what they would pay. If you cannot name the buying unit, you cannot size the market, and everything downstream is arithmetic on an undefined quantity.

### 3. Size top-down

Start from a published figure, name the source and the year in the same line, and cut it down to the reachable slice with explicit filters: geography, segment, price point, channel. Each filter is a numbered assumption with a stated basis.

The test for a top-down anchor: do you know what is inside it. A figure that bundles hardware, services, and adjacent categories will overstate by a factor that nobody can see once it has been cut three times. Where you cannot establish what the anchor contains, do not use it as the anchor; use it as a sanity check.

### 4. Size bottom-up

Number of buying units multiplied by realistic annual spend per unit. State where each input comes from. Prefer counts from a source that exists for another purpose: a regulator's register, a trade association membership list, a national business register, a licensing body. Those counts are collected for administrative reasons and are more reliable than any figure produced for a market report.

For spend per unit, use observed prices where they are published, your own realised contract values where the segment is close enough to yours, or a stated range where neither exists.

### 5. Reconcile the two, and say which you trust

Put them side by side. If they disagree by less than two times, take the lower and state the range. If they disagree by more than two times, the divergence is a finding: something in the boundary definition differs between the two methods, and locating it is usually the single most useful hour in the whole piece of work.

Then separate the three quantities and never let them blur. The total market is everyone who could buy. The serviceable market is the slice the product can serve today, with the constraints that produce it named. What could realistically be won is a two to three year figure built from a share assumption you can defend against your own history in a comparable segment. State the assumptions that separate each from the one above it.

### 6. Map the competition by evidence, not by logo

For each relevant competitor: what they sell, to whom, at what price where pricing is public, how they distribute, funding or revenue where known, and their most recent strategic move with a date. A source on every row.

Then answer the only competitive question that matters: on what basis would a buyer choose us instead, and is that basis durable for more than a year. A comparison that ends in a feature grid has not answered it.

Include the non-obvious competitors, which are usually where the deals are actually lost: the status quo, meaning spreadsheets and doing nothing; an internal build; a services firm doing it manually; and the adjacent player one product move away from the market.

### 7. Verify before anything enters the document

Every number carries a source and a year in the same sentence or the same table row. Prefer primary sources: company filings, pricing pages, regulator data, national statistics, earnings calls, and job postings, which show what a company is building three to six months before it announces.

Two independent sources for any number that carries the recommendation. Where only one exists, mark it as single-sourced in the document itself, not in a footnote.

Dates matter as much as figures. A market size from five years ago quoted in a current memo is a different claim, and it must either be updated or flagged where it appears. Anything about pricing, leadership, funding, or headcount older than twelve months is treated as possibly stale.

Never fabricate a figure, a company detail, or a source. A gap stated plainly is more useful than a guess: "no public revenue figures; the closest proxy is headcount growth from 60 to 84 in a year, from their own careers page and archived versions of it" tells the reader exactly how much weight to put on it.

### 8. Deliver as a decision input

Answer the framing question in the first line, against the kill number. Then the sizing with the method shown, the competitive read, the two or three risks that would invalidate the answer, and what to watch that would change it later. Tables for competitor and sizing data, prose for the argument.

End on the finding. "Further research is recommended" is what a document says when nobody was willing to answer.

## Sourcing, in order of preference

1. **Administrative and regulatory counts.** Registers, licensing bodies, national statistics, filings. Collected for a purpose other than marketing, and correspondingly reliable.
2. **Company primary material.** Filings, pricing pages, annual reports, earnings transcripts, job postings, product documentation.
3. **Connected data services**, where one is available: firmographics, funding, headcount, technology signals. Useful for breadth and for anything at company level. Where none is available, the same facts can usually be assembled from filings and archived web pages at the cost of time.
4. **Quality secondary press and analyst work**, read for its method rather than its headline.
5. **Market report press releases.** A last resort, always labelled as such, never the anchor for a size that carries the recommendation.

Where a figure exists only in tier five, say so in the line where it appears.

## Worked example

**Situation.** Alderby Systems, a field service scheduling software company with 18 million euros in revenue, was deciding whether to open a dedicated team for the dental practice segment in one European country. The head of sales had two inbound wins in that segment and believed it was a large market. The board would review the annual plan in five weeks, and a dedicated team meant four hires and roughly 620,000 euros of annual cost.

**Task.** Establish whether the serviceable market justified a dedicated team, against a threshold the chief financial officer set at the start: at least 15 million euros of serviceable annual spend and a credible route to 10 percent of it within three years. Anything below that and the segment would be served through the existing generalist team.

**Action.** The top-down leg was attempted first and produced the wrong turn. A published figure of 480 million euros for practice management software across the region was cut by country share, then by the software share of it, giving something near 62 million euros. It looked decisive and it was abandoned in the second week, for a specific reason: reading the report's own methodology note showed the figure bundled imaging hardware, chairside equipment software, and payment processing. Nobody could say what fraction was scheduling and workforce software, and three successive cuts to an unknown base is arithmetic on a number, not a size.

The bottom-up leg replaced it and was built from a primary count. The national dental council register listed 12,430 registered practices. That count was then corrected for the buying unit, which was the correction that changed the answer: 31 percent of practices belonged to groups operating five or more sites, and those groups bought centrally, so the actual number of buying units was about 9,100, not 12,430.

Spend per unit came from two observed sources: the published list pricing of two incumbent products, which ranged from 240 to 410 euros per site per month, and Alderby's own two realised contracts in the segment, at 3,900 and 4,600 euros a year. Taking a conservative 3,800 euros per buying unit per year gave a serviceable market of about 34.6 million euros, and applying it only to the practices with three or more staff, which was the smallest size at which the product made sense, cut it to about 21 million euros.

The two methods disagreed by roughly three times, so the divergence was investigated rather than averaged. The gap was almost entirely the bundled hardware and payments inside the top-down anchor. That finding was reported in one line and the top-down figure was retained only as an upper bound.

The competitive map covered five named competitors plus two non-obvious ones. The most useful row was the status quo: interviews with the two existing customers, plus fourteen practice websites and three trade forum threads, indicated that a large share of small practices ran scheduling on a general-purpose calendar and a paper day book. The most recent strategic move that mattered was dated: the largest incumbent had raised funding eleven months earlier and posted six roles in that country in the previous quarter, which was visible on their own careers page.

**Result.** The serviceable market came in at 21 million euros, above the 15 million euro threshold. The share question failed instead. Alderby's realised share in the closest comparable segment it already served was 3.1 percent after four years, and reaching 10 percent of dental in three years would have required roughly three times that rate against a better-funded incumbent that was actively hiring in the country.

The recommendation was not to open a dedicated team, and instead to sign one reseller already serving the segment, at a cost of about 90,000 euros in the first year against 620,000 euros. The board took that decision. Whether the reseller route reaches the same revenue is unknown; at the time of writing it had produced nine opportunities in two quarters, which is ahead of the plan but too early to read.

The part that mattered was the buying-unit correction. It moved the number by a quarter, and it came from a register that took twenty minutes to find and cost nothing.

### A second scenario, where it goes differently

A different question: sizing a market for a product category that does not yet exist as a category, in this case automated compliance evidence collection for mid-sized manufacturers. There is no published market figure, no analyst category, and no register of buyers.

The top-down leg is unavailable, so it is not faked. The method substitutes displacement sizing: the spend already going somewhere else that this product would take. Here that was audit preparation time, established from three sources, and the fees paid to the firms currently doing the work, established from two published fee schedules and one interview. The bottom-up leg is built from the national business register count of manufacturers above a staff threshold.

The deliverable changes shape. Instead of a single serviceable figure it presents a range, from 8 to 26 million euros, with the one assumption that dominates the range named explicitly: what share of the current external fee is displaceable rather than mandatory. The recommendation is correspondingly conditional, and the document says in its first line that the range is wide and why.

What changed is not the discipline. It is that with only one sizing method available, the honest output is a range with its dominant assumption exposed, and presenting a point estimate from that evidence base would have been the failure.

## Output

The default deliverable is a short memo. One page for a decision, with the tables carrying the data.

```
MARKET RESEARCH: [segment or market]
DECISION IT FEEDS: [the choice, and who makes it]
THRESHOLD:         [the kill number agreed at the start]
MARKET DEFINITION: [buying unit, unit of purchase, boundary]
DATE / VINTAGE:    [when the research was done, oldest source used]

ANSWER
[One or two sentences, against the threshold.]

SIZE
[The reconciled figure, the method, and the range.]

COMPETITIVE READ
[On what basis a buyer would choose us, and whether it is durable.]

RISKS TO THIS ANSWER
[Two or three, each with what would show it was happening.]

WHAT TO WATCH
[The two or three signals that would change the answer later, and where they appear.]
```

**Sizing table**, both methods shown:

| Method | Step | Figure | Source and year | Assumption or filter applied |
|---|---|---|---|---|
| Top-down | Anchor | | | |
| Top-down | After filters | | | |
| Bottom-up | Buying units | | | |
| Bottom-up | Spend per unit | | | |
| Reconciled | Serviceable | | | |

**Competitor table**, one row each, including the status quo and the adjacent player:

| Competitor | What they sell | To whom | Price (public) | Distribution | Funding or revenue | Latest move, dated | Source |
|---|---|---|---|---|---|---|---|

Where the research feeds a deck, the memo is written first and the slides are built from it. Charts stay monochrome, with series separated by marker and dash pattern rather than colour and the legend outside the plot area, so they survive being printed and projected.

## Failure modes

**The logo wall.** Recognise it when the competitive section is a grid of names with no prices, no dates, and no source. It looks like coverage and answers nothing. Fix by requiring a dated most-recent-move and a source on every row, which quietly removes the competitors nobody could find anything about.

**The press-release anchor.** A headline market figure whose composition is unknown, then cut three times. Recognise it by asking what is inside the number and getting no answer. Fix by demoting it to a sanity check and building the size bottom-up.

**The single confident number.** One figure, no method, no range. Recognise it because the document cannot answer "how did you get that". It will not survive one board question. Fix by showing both methods, always.

**Total market inflation.** Counting everyone who could conceivably buy, including segments the product cannot serve and geographies you cannot sell into. Recognise it when the serviceable figure is more than a quarter of the total and no filter explains why. Fix by making every filter a numbered assumption.

**Stale figures presented as current.** Recognise it by the absence of years. Fix by putting the year in the same line as every number, which makes staleness visible rather than requiring vigilance.

**The status quo left out of the competitive set.** Most losses in a new segment are to doing nothing. Recognise it when the competitor list contains only vendors. Fix by pricing the status quo: what it costs the buyer today in time and money.

## Edge cases

**No published market data exists.** Use displacement sizing, as in the second scenario: find the budget the spend would come from and size that. Report a range with the dominant assumption named.

**The main competitors are private with no disclosures.** Build from observable proxies with sources: headcount over time from their own careers page and archived versions, pricing pages, customer counts they publish, job postings by function, and filings where a jurisdiction requires them. State that revenue is unknown rather than estimating it to a precision the proxies cannot support.

**The answer becomes obvious in the first week.** Stop and deliver it. A three-line answer with two sourced numbers, delivered in week one, is worth more than the same answer in week five with thirty slides behind it. Say what you did not do and what would change the conclusion.

**The decision has already been made.** Recognise it when the request is for the market section of a document whose conclusion is fixed. Say so, and offer to make the numbers defensible rather than to make them supportive. Research commissioned to justify a decision is discoverable by anyone who asks how it was built.

## Quality bar

- The first line answers the framing question against the threshold agreed before the research started.
- Every market size is built twice, the two methods are shown side by side, and the divergence is explained rather than averaged.
- The buying unit is defined in units somebody could sell to, and the boundary of the market is stated.
- Every number carries a source and a year in the same line, and single-sourced numbers are marked as such in the body.
- Total, serviceable, and winnable are distinguished, and each filter between them is a numbered assumption.
- The competitive set includes the status quo and at least one adjacent player, and every row has a dated latest move.
- What is known, what is estimated, and what is assumed are visibly different, and no adjective does the work of a number.

## Adapting this to your context

The sourcing order, the two-times reconciliation rule and the register-based count come from software and services markets in Europe and North America, where registers are public and pricing is often published.

- **Registers as the bottom-up count.** Where no public register exists, substitute a trade association list, a procurement portal or an industry directory, and say what it excludes.
- **Published list pricing.** Common in software, rare in industrial, medical device and defence contracting. Where price is negotiated, build spend per unit from your own realised contracts or from published tender awards, and mark it as the dominant assumption.
- **The two-times reconciliation rule.** A tolerance for a mature category with a published anchor. For an emerging category, use the displacement method in the second scenario rather than forcing a top-down leg that does not exist.
- **The twelve-month staleness rule.** Right for pricing and funding in fast-moving software. A structural figure in a slow regulated market may hold for three years; under regulatory change, six months is old.
- **What not to change.** The kill number is written before the research starts, every figure carries its source and its year in the same line, and a gap is stated as a gap rather than filled.

## Related skills

`structured-problem-solving` frames the question this research answers and decides which branches deserve the effort. `decision-memo` carries the resulting recommendation to a named decider. `external-insights` covers the single company or person brief, including the pre-meeting version. `competitive-battlecard` turns the competitive read into the seller-facing page used in live deals. `customer-interview-synthesis` supplies the demand-side evidence that a bottom-up size cannot produce on its own. `board-deck` and `pitch-deck` are where this work is usually presented. `partnership-assessment` and `vendor-evaluation` use this method on a single counterparty. `revenue-forecast` takes the winnable share figure and turns it into a plan.