How to do an industry analysis

What is an industry analysis, and what should it produce?

An industry analysis is a structured read of a whole industry: its size and growth, how value moves from suppliers to end customers, who holds pricing power, and what will shift any of those over the client's horizon. The client pays not for the read but for three or four findings that change a decision: enter or stay out, invest or pass, build or partner.

A market size says how big the pond is. An industry analysis says who gets to fish in it, at what margin, and whether that is about to change: in AI clinical documentation, whether the EHR vendor takes the catch.

An industry analysis contains the other two. It needs a size, and it needs a competitive landscape, but it treats both as evidence for a judgment about structure rather than as the deliverable.

How do you do an industry analysis, step by step?

Six steps, in order. Every later step depends on the boundary set in the first: drawn around generative AI in clinical documentation, the 2025 market was $0.79bn; drawn around clinical documentation improvement, $4.01bn.

1. 01Frame the question and fix the boundary: write the client's decision as a hypothesis, define the industry by product, customer, geography and value-chain stage (ambient note drafting, not all clinical documentation software), and note the matching NAICS code.
2. 02Size the industry from at least two independent sources, recording each figure's unit, period, geography and what it counts, then set a range, as $1.0bn–$1.5bn was for ambient documentation in 2026.
3. 03Map the value chain from inputs to end customer, and mark where the margin sits, who captures it and why they can keep it: in clinical documentation, the EHR vendor that owns the clinician's screen.
4. 04Test the structure: concentration (an HHI above 1,800 is high), buyer and supplier power, barriers to entry and the threat of substitutes, such as an EHR building the scribe in.
5. 05Name the three to five drivers that will move the numbers over the client's horizon, each with a dated source: Getir's UK exit, announced April 29, 2024, reset quick commerce.
6. 06Write the so-what: three or four findings, each a slide headline that states the implication for the client's decision, such as 'embed in the EHR or expand into coding'.

US federal statistical agencies classify every business establishment under NAICS, the North American Industry Classification System, which groups establishments by how they produce rather than what they sell. A client's idea of its industry rarely matches one code, so state the gap: what the client means, what the code covers, and which way the difference cuts.

Two sources is the minimum, because one source cannot reveal its own scope. A gap is either definitional, because the sources count different things, or genuine, because they count the same thing and disagree. The first is settled by choosing the definition that fits the client's question, the second by testing the evidence behind each figure. Never average them. How to reconcile market size estimates is a discipline of its own.

For concentration, use the measure US antitrust agencies use. The Herfindahl-Hirschman Index squares each firm's market share and sums the results: four firms at 30%, 30%, 20% and 20% score 2,600. Under the 2023 Merger Guidelines, a market above 1,800 is highly concentrated, and one between 1,000 and 1,800 moderately so. The Census Bureau's Economic Census, taken every five years for years ending in 2 and 7, publishes the concentration of the largest firms by industry.

Porter's five forces remains the right checklist for the structure step: rivalry, buyer power, supplier power, new entrants and substitutes. His sharper point, in the 2008 restatement in Harvard Business Review, is that managers define competition too narrowly, as if it occurred only among today's direct competitors. In clinical documentation the decisive force sat outside the rival set: not another scribe, but the EHR that standalone scribes embed in, once Epic built the feature into it.

Which framework should you use for an industry analysis?

A slide titled 'Five forces' with five boxes of description tells the partner nothing; a slide headed 'The EHR vendor owns the clinician's screen, and with it distribution' tells the partner what to say. Choose each framework for the question it answers.

Two usually suffice: the clinical documentation study used a profit pool for structure and dated drivers for change, from Epic in February 2026 to Oracle Health in August. SWOT belongs in the company section of the deck. Applied to an industry, it produces a list of adjectives.

How do you get up to speed on a new industry in two days?

The client presentation is in 48 hours, the associate who did pharma last month is on logistics this month, and nobody on the team has covered the sector. Two working days is enough if the work runs in the order the analysis needs, not the order in which sources turn up.

The fastest primary source on a US industry is the annual report of its leaders. The SEC's investor guide describes Item 1 of Form 10-K as covering products, markets, competition and regulation, and notes that companies generally list Item 1A risk factors in order of importance. Three 10-Ks read side by side give the industry's own account of what threatens it. Item 7, management's discussion and analysis, adds the trends management expects to affect results.

For a UK industry, swap the Census Bureau for the national statistics office and the 10-K for annual results, where Tesco's Whoosh and Sainsbury's on-demand figures come from.

The time goes on finding numbers that can be defended and reconciling them when they disagree, not on reading. That part can be handed to a tool, and the test for AI market research is whether each figure arrives with its source and basis. Caspr returns a structured, cited report from a single-line prompt in under 15 minutes, which leaves the two days for judgment.

What goes wrong in an industry analysis?

One number with no basis. In AI clinical documentation, three publishers' 2025 market sizes run from $0.79bn to $4.01bn, roughly five times apart, and each is right for its own definition. A deck that shows one of them, unlabeled, invites the first question it cannot answer. The basis is what lets the team defend a market size estimate when the partner, and then the client, asks where it came from.

Figures older than the last structural break. Caspr's analysis of UK quick commerce found that Getir's exit from the UK, announced on April 29, 2024, which also ended Gorillas there, handed the channel to supermarkets and delivery apps: Tesco's Whoosh passed £400m of sales in FY2025/26, up 51%, and Sainsbury's on-demand sales passed £700m, up 69%. A sizing written before 2024 describes an industry of venture-funded dark stores that no longer exists. Ask of any figure not how old it is, but whether it predates the reset.

Description in place of conclusions. The test is whether every slide headline would still carry the argument if the client read nothing else on the page.

Sources with a position. Vendor-funded studies, trade association figures and investor presentations are useful and partial; Epic's 16 million monthly chart summaries is Epic's own count. Say who published each one and why; a reader who learns it later discounts everything near it.

What should an industry analysis report contain?

The report follows the steps, and each section ends in a finding, not a summary. The right-hand column shows the headline each section should produce, taken from the AI clinical documentation example.

Keep the sources on the slide, not in an appendix: every number with its publisher, its date and what it counts. A line under $0.98bn reading 'Mordor Intelligence, 2025, AI in clinical documentation' answers the first question before anyone asks it.

Frequently asked questions