Porters Five Forces Analysis: A Guide for AI Startups
Summary
A Porters five forces analysis rates a market on rivalry, buyer power, supplier power, new entrants and substitutes to show how much revenue can stay as profit. For an AI startup, define one narrow segment, rate each force low, medium or high with a dated fact, then name the force that sets the margin ceiling. Supplier power and substitutes usually dominate. Re-score every quarter and track the delta.
A Porters five forces analysis scores an industry on five pressures: rivalry, buyer power, supplier power, new entrants and substitutes. The score tells you how much of the revenue in a market can survive as profit. For an AI startup the useful version is narrow: pick one segment, rate each force low, medium or high, and attach one piece of evidence to every rating. Everything below follows that method, with a worked example at the end.
What does a Porter's five forces analysis actually measure?
It measures industry structure, not company quality. Michael Porter's argument in the Harvard Business Review original is that profit is contested by established rivals and by four other sources: customers, suppliers, potential entrants and substitute products. A strong team in a bad structure still earns thin margins.
That distinction is why founders and investors keep using it. A pitch deck argues about the team and the product. The five forces argue about the arena. If all five pressures are high, no amount of execution quality fixes the arithmetic of the segment.
The framework has one failure mode that matters here: defining competition too narrowly. Looking only at direct rivals is the shortcut, and it hides the buyer who can build in-house, the supplier who can move into your layer, and the free tool that makes the whole category optional.
How do you run it in five steps?
The process is short. The discipline is in step one and step three.
Define the segment. "AI" is not a segment. "Voice agents for dental clinics in the US" is. Product, customer type and geography must all be fixed before anything is rated.
List the participants. Rivals by strategic group, buyers by segment and channel, suppliers by tier, plausible entrants, and substitutes including doing nothing.
Rate each force low, medium or high, with evidence. A rating without a fact behind it is an opinion. Pricing pages, hiring boards, funding rounds and switching costs count as facts.
Look for the force that sets the ceiling. Usually one or two forces explain most of the margin pressure. Name them.
Decide what the structure implies. Enter, avoid, or pick a position that blunts the dominant force.
Step four is where most write-ups stop short. A table of five ratings is not a conclusion. The conclusion is a sentence such as "supplier power sets the ceiling, so the strategy is to reduce dependence on a single model provider".
Which of the five forces matters most for an AI startup?
For AI application companies, supplier power is the first one to examine, and it is rarely low. A startup whose product is a thin layer over a third-party model inherits that provider's pricing, rate limits and roadmap. The provider can cut prices, which helps margin, or ship the same feature natively, which removes the reason to buy.
Buyer power comes second. Enterprise buyers run pilots with three vendors at once, and procurement teams now ask for exit clauses and data portability by default. In SMB, buyer power looks low because each account is small, but churn does the same job that negotiation does in enterprise.
The threat of new entrants is structurally high. A working prototype now takes a weekend, and capital is available for the teams that show traction. The barrier is no longer building. It is distribution, proprietary data and workflow depth.
Substitutes deserve a separate look. For many AI products the substitute is a general assistant plus a prompt, or a spreadsheet and an intern. If a customer can get 70% of the outcome from a tool they already pay for, your pricing power is capped near the cost of the missing 30%.
Rivalry closes the loop. When entrants are many and products converge, rivalry shows up as price cuts and feature parity within a quarter.
What does each force look like when you rate it with evidence?
The list below converts each force into a question you can answer from public data. The delta matters more than the absolute reading, so record the date next to each rating.
Rivalry: how many funded competitors exist, and how fast do features converge? Check funding databases and changelogs.
Buyer power: can the customer switch in under a month or build it in-house? Check contract terms and integration depth.
Supplier power: which single vendor can reprice or compete with you? Check API pricing history and vendor roadmaps.
New entrants: what does parity cost, and who is hiring for it? Check job postings and new round announcements.
Substitutes: what does the buyer do today instead of paying you? Check customer interviews and tool stacks.
Funding and hiring data do most of the work for rivalry and entrants. That is where a startup database earns its fee: PitchBook covers private-market rounds and investors in depth, and it is priced for funds rather than for solo founders. Free or cheaper databases cover early-stage rounds with a lag of weeks. Pick the tool by latency on the segment you are rating, not by brand.
For the demand side, traffic and audience tools give you a read on how big the buyer pool is and where it already looks.
How do you rate substitutes and rivalry without guessing?
Start with rivalry, because it is the easiest to measure and the most over-weighted. Count funded direct competitors in your exact segment, then check whether their pricing pages have moved in the last two quarters. Three funded rivals with diverging prices is a healthier structure than three with identical tiers.
Competitive intelligence platforms automate part of that tracking. Tools such as Crayon, Klue and Contify watch competitor sites, releases and press and package the changes for sales and strategy teams. They suit a company that already has rivals to monitor. For a pre-seed team, a manual weekly review of five pricing pages gives the same signal at no cost.
Substitutes need a different method: talk to buyers. Ask what they used last quarter for this job and what it cost in hours. Audience research tools such as SparkToro help you find where those buyers gather and what they already read, which narrows the interview list.
One rule keeps this honest. If your only evidence for "low substitute threat" is that you could not think of one, rate it medium until a buyer says otherwise.
A worked example: AI meeting notetakers
Take one segment: AI notetakers that join sales calls for mid-market B2B teams. Ratings are illustrative and based on the public structure of the segment, not on a dataset.
Rivalry: high. Many funded tools, near-identical transcripts and summaries, free tiers as the default entry point.
Buyer power: medium to high. Switching is a calendar integration away. Annual contracts and CRM sync are the only real locks.
Supplier power: medium. Transcription and summarization run on third-party models. The dependency is real but several providers are interchangeable.
New entrants: high. Video platforms ship native notetakers, and any CRM vendor can bundle one.
Substitutes: high. The built-in recording feature of the video platform already covers part of the job at zero marginal price.
Verdict: the ceiling is set by substitutes and entrants together. Both come from platforms that already own the meeting. The position that survives is the one platforms will not copy cheaply, which means deep CRM workflow and coaching analytics rather than transcription. That sentence is the output of the analysis. The five ratings are only the scaffolding.
What does a bad analysis look like?
Four patterns show up repeatedly in decks and strategy memos.
Everything is "medium". Five medium ratings mean nobody gathered evidence. A real segment has at least one force that is clearly high or clearly low.
The segment is the whole industry. Rating "AI software" produces a table that could sit in any deck. Rating "contract review for mid-size US law firms" produces something a buyer can argue with.
Forces are scored as opinions of the author. Replace adjectives with observable facts: number of funded rivals, notice period in contracts, share of cost that goes to one vendor, months to reach feature parity.
The result is never used. If the output does not change a pricing decision, a build-or-buy decision or a go-to-market choice, the exercise was decoration. Attach one decision to each high rating before you close the document.
Where does the framework break for fast-moving AI markets?
Three limits are worth stating plainly.
It is a snapshot. Model prices fell and capabilities moved several times in a year. A rating from last quarter can be wrong on supplier power today. Re-score on a fixed cadence and keep the history.
It ignores complements. Platforms, standards and regulators shape profit but do not fit the five boxes cleanly. Add them as a sixth line when they matter. Rules such as the EU AI Act change buyer behavior and entry cost, and belong in the evidence column.
It says nothing about the quality of your execution. Two companies in the same structure earn different returns. The framework tells you the arena. It does not tell you who wins in it.
What we'd actually do with the result
If you are a founder, run the analysis before you write the deck and put the dominant force on slide three, with the mitigation next to it. Investors have seen the framework labelled and left blank. They have rarely seen it filled with dated evidence.
If you are an analyst, score the same segment twice, six months apart, and report the change. The delta compares segments better than any single rating does, and it is the part a reader cannot get from a generic template.
Skip the exercise if you cannot define the segment in one sentence. The ratings will be noise. Fix the definition first, then score.