How to do a competitor analysis

A competitor analysis is finished when it produces a dated decision, not when the grid is full. The method below is the one that gets there: pick the three to five rivals your customers actually compared you against, collect six things about each that are genuinely published, record every finding with its date and its source, grade how sure you are, and then run the arithmetic that turns a finding into a decision. The worked example runs end to end on an invented retailer — including the part where matching a competitor’s 10% price cut turns out to need 36% more orders just to stand still, and where holding your price only costs you money if you would otherwise lose more than 27% of your orders.

What a competitor analysis is actually for

Most competitor analyses die in the same place. Someone builds a grid — competitors down the side, features across the top, ticks in the boxes — circulates it, and nobody ever opens it again. The work was real. The output changed nothing, because a grid is a description and a description is not a decision.

The fix is to write the decision down before collecting anything. Not a topic, a sentence with a verb and a date in it: by the end of the month we will decide whether to change our delivery threshold. Then every hour spent collecting can be tested against one question: does this move that decision? Most of what a competitor publishes does not, and knowing that in advance is what stops the exercise swallowing a week.

In practice, competitor research can change exactly three things. Everything else it produces is interesting rather than useful.

• Your price, or the conditions attached to it. The headline number, the delivery threshold, the minimum order, the discount ladder, the length of the free trial. These are published by almost everyone, they move, and they are directly comparable to yours.
• What you say. Not what you build — what you claim, and where. If a competitor has just raised their prices and you have not, you are cheaper on identical specification and almost certainly not saying so anywhere. That is a change you can make this week at no cost.
• Where you spend to get a customer. A competitor appearing far more often on a search term you rely on raises the price of that term for you. That is a budget decision, not an opinion.

Notice what is missing from that list: your product roadmap. Competitor features are the most collected and least actionable thing in the whole exercise, because by the time you can see a feature it has already shipped, and copying it puts you a release behind on someone else’s judgement of what mattered. Collect features if you like, but do not let them set the agenda.

So the shape of a finished analysis is: a short list of competitors, a dated record of what each has actually published, a grade saying how sure you are of each item, and — the only part anyone will read twice — a handful of decisions with dates against them, each one pointing back at the specific finding that caused it.

Choosing the three to five competitors that matter

The list of competitors you would write from memory is usually wrong. It contains the companies you think about, which is not the same as the companies your customers compared you against before they bought.

Three places already hold the real answer, and all three are free:

• Your own lost deals. Whatever you record when an enquiry does not convert. If you record nothing, start: one line, who else were they looking at, is the single highest-value field in any sales record.
• The search results you already appear in. Take the five terms that bring you the most enquiries and look at who else is on the page, organically and in the ads. Those companies are competing with you for the same intent whether or not they sell the same thing.
• The names customers say unprompted. In enquiries, in reviews, in support conversations. A name that arrives without you asking is worth more than a name you produced by thinking hard.

Then sort what you find into three rings, because they need different treatment:

Cap the direct list at five, and prefer three. Each name on the list costs the same collection effort every time you repeat the exercise, and repetition is where the value is — a single snapshot tells you almost nothing, because you cannot see a change from one observation. Three competitors tracked every quarter beats eight described once, every time.

The worked example for the rest of this page: Halden Supply, an invented online retailer of workshop equipment, with two invented direct competitors, Ridgeway Trade and Colvin & Sons. All three are made up, and so is every figure attached to them. They exist here so the arithmetic has something to bite on, not as a claim about anybody real.

The six things worth collecting, and where each one is public

The rule for collection is narrow and it saves a great deal of time: take what the company publishes about itself, not what is said about it. A competitor’s own pages are dated, attributable and checkable by anyone you later show the work to. A third-party summary of the same thing is a claim about a claim, and it is where most competitor analyses quietly stop being evidence.

Two things worth collecting with care

Review counts. The number of reviews a competitor has accumulated is public, and the rate at which it grows is a rough proxy for the direction of their order volume. It is a proxy and nothing more: review rates respond to prompting, to a change of platform, and to one unhappy month. Use it to notice that something changed, never as a figure.

Search interest. Google Trends will tell you the relative direction of interest in a brand name or a category term over time. It gives relative interest, not volumes, so it answers is this rising or falling and never how many. Treated as the first it is useful; treated as the second it is a fabricated number.

Record each item in the same six fields every time: what exactly was found; the page it was found on; the date it was found; the previous observation and its date; how sure you are; and what would change your mind. That last field is the one that separates research from an opinion with dates attached, and it is also what makes the record worth keeping — in three months it tells you what to go and look at again.

Dating and grading every finding

Here is the rule that does the most work in the whole method, and the one most often broken:

One observation is a state. Two dated observations are a change. Three in the same direction are a trend — and only then.

A price you saw once is not a price rise. A page that exists today is not a new page. Almost every confident, wrong conclusion in competitor research comes from treating a first observation as a change, because the observer had no record of what it looked like before. The remedy is unglamorous: sample the same things on a fixed cadence and keep the dates.

Then grade each finding honestly. Three grades are enough, and the definitions have to be tight or everything becomes medium:

A low-confidence finding is worth recording and worth not acting on. Write the re-check date next to it — if the login page has not become a public trade-price page by 24 October, drop this — and it stops being a rumour that gets repeated until it becomes a fact by attrition.

Absence is a finding

If you sampled thirty of a competitor’s prices on two dates and every one was unchanged, the honest output is no material signal found. It is tempting to write holding firm on price instead, because it sounds like a finding and fills a line. It is not one: it is a flat month, described in a way that implies intent nobody observed. A colleague reading no material signal found knows exactly how much to rely on it. A colleague reading holding firm does not.

The same applies to the things you went looking for and could not get. A short list of what the pass deliberately failed to find is more useful than the equivalent space filled with inference, because it tells the next person where not to waste their time.

Turning a finding into a decision: the arithmetic

This is the part that gets skipped, and it is the part that pays. A finding is not a decision until somebody has worked out what acting on it costs.

Take the highest-confidence finding from the example. Ridgeway Trade cut the price of the equivalent product from £47.00 to £42.30, a 10% reduction, observed on two dated samples. Halden Supply sells its version at £48.00 and it costs them £30.00 a unit delivered. Every figure here is invented for the example; put your own in.

Now the question the grid never answers: how much more would you have to sell, at the lower price, to end up where you started?

Extra orders needed = old gross profit per order ÷ new gross profit per order

£18.00 ÷ £13.20 = 1.364, so you need 36.4% more orders. At Halden’s illustrative 1,200 orders a month that is 1,637 orders — an extra 437 a month, every month, to stand exactly still. Check it: 1,200 × £18.00 = £21,600, and 1,637 × £13.20 = £21,608. A 10% price cut has to be paid for with a 36% volume increase, and it has to be paid for permanently, because the price stays cut.

The sum that settles it

Run it the other way round and the decision becomes obvious. If Halden holds at £48.00 and loses some orders to the cheaper competitor, how many can it lose before holding is worse than matching?

Matching at unchanged volume gives 1,200 × £13.20 = £15,840 of gross profit. Holding at £48.00 with N orders lost gives (1,200 − N) × £18.00. Setting them equal:

(1,200 − N) × £18.00 = £15,840 → 1,200 − N = 880 → N = 320 orders

So matching the cut only makes Halden better off if holding would otherwise cost them more than 320 orders a month — 26.7% of everything they sell. That is an enormous amount of customer behaviour to assume from a competitor changing a number on a page, and nothing in the research shows a single customer switching. It shows a competitor changing a number on a page.

Which turns the finding into a decision with a shape: do not match; test. Cut one category by 5% for three weeks, measure conversion against the same weeks last year, and decide on 23 October on the evidence rather than on the announcement. That costs a fraction of the permanent margin and it answers the question the research genuinely cannot.

The finding that costs nothing to act on

The second high-confidence finding usually gets ignored because it is not dramatic. Suppose Colvin & Sons raised prices on fourteen lines by 6–9% while Halden did nothing. Halden was about 2% cheaper on those lines before; it is now 8–11% cheaper on identical specification, it did nothing to earn that, and it says so nowhere. Publishing a dated, like-for-like comparison on those fourteen product pages costs no margin, needs no discount, and is available this week. Most competitor analyses produce one finding like this and leave it in the grid.

If you want the margin side of these sums done for you, the profit margin calculator converts between cost, price, margin and markup in both directions, and how to price a product works through the full cost stack a price has to carry before any of this is worth arguing about.

What a competitor analysis cannot tell you

Being explicit about the boundary is what makes the rest trustworthy. Five things sit permanently outside what public research can reach, and a pass that pretends otherwise is worth less than no pass at all.

• Their margins, unit costs and supplier terms. Not published, not inferable from a price. A price tells you what a customer pays; it tells you nothing about what the seller keeps. Any competitor margin you have seen quoted was reverse-engineered from assumptions, and the assumptions were the answer.
• Why they did it. You can often establish that a competitor raised prices. The cause is almost never visible, and a supplier cost increase, a range refresh and a deliberate margin decision look identical from outside. Reporting the move is research; reporting the motive is fiction with a date on it.
• Whether it is working for them. A competitor who cuts prices and a competitor who cuts prices successfully produce exactly the same evidence for months.
• Intent. A job advert is a budget decision that has already been made, not a plan that will be carried out. Roles are posted and never filled; pages exist for years without launching.
• Private customer behaviour. Their churn, their repeat rate, their contract terms, the size of their largest account. All decisive, none public.

There is also a trap specific to doing this well, which is that a good competitor analysis is quietly seductive. The more evidence you assemble about what a rival is doing, the more of your own agenda they end up setting. The counterweight is the decision list: if a quarter’s research produces no decision you would not otherwise have made, the honest conclusion is that nothing material changed, and the correct response is to spend the next quarter’s hours on your own customers instead.

Arithmetic and general information only — not financial, tax, legal or investment advice. Your own figures, and your own accountant, decide what any of this means for you.

Frequently asked questions

How do you do a competitor analysis?

Write down the decision you are trying to make and its date first. Then pick the three to five competitors your customers actually compared you against, taken from your lost-deal notes, the search results you already appear in, and the names customers say unprompted. Collect six things each competitor publishes about itself: listed prices, the delivery or service promise, the positioning claim, hiring, channel presence, and filed accounts where they exist. Record every item with its date and source, grade how sure you are, and finish by working out what acting on each finding would cost. The output is a short list of dated decisions, not a grid.

How many competitors should you analyse?

Three to five direct competitors, and prefer three. Each name costs the same collection effort every time you repeat the exercise, and repetition is where the value sits, because a single snapshot cannot show you a change. Three competitors tracked every quarter is worth far more than eight described once. Keep a looser watch on substitutes, which take the whole category rather than a share of it when they move, and remember that the largest competitor by volume is usually the customer deciding to do nothing at all.

What information about a competitor is actually public?

What they publish about themselves: listed prices and the conditions attached to them, delivery thresholds and lead times, the claims on their homepage and category pages, their job adverts, which search terms and marketplaces they appear on, and for UK companies their filed accounts on the public register at Companies House. Review counts and search-interest trends are available too, but both are direction rather than measurement. What is never public is the thing people most want: their margins, their unit costs, their supplier terms and their churn.

How often should you redo a competitor analysis?

Often enough to have two dated observations of the same things, because one observation is a state and only two make a change. A fixed quarterly cadence on a fixed list of items works for most businesses; monthly is worth it only where prices and promotions move monthly. The cadence matters more than the depth, because the whole method depends on being able to compare what you found today against what you recorded last time.

Should you match a competitor who cuts their price?

Work out the two numbers before deciding. First, the extra volume a matched cut has to produce: old gross profit per order divided by new gross profit per order. On the worked example, an 18 pound gross profit falling to 13.20 pounds needs 36.4% more orders just to stand still, permanently. Second, the volume you would have to lose for holding to be the worse option, which on the same figures is 320 orders out of 1,200, or 26.7%. If the research shows a price change but no evidence of customers actually switching, a small dated test on one category answers the question for a fraction of the margin.

What is the difference between a competitor analysis and a SWOT analysis?

A SWOT is about you: strengths, weaknesses, opportunities and threats, assembled mostly from internal judgement. A competitor analysis is about evidence someone else published, collected on dated samples and graded for confidence. They answer different questions and the failure modes differ too. A SWOT fails by becoming a list of opinions nobody can check; a competitor analysis fails by becoming a grid that changes no decision.

Sources

• Companies House: Search the register (find and update company information) — the UK public register of company filings, including filed accounts and document images — checked 2026-09-25
• Google Trends — relative search interest over time for a brand or category term, which shows direction rather than volume — checked 2026-09-25
• Instilus: how to price a product, for the full cost stack a price has to carry — checked 2026-09-25

Other guides

• How much does Etsy take from a sale?
• How to build a cash flow forecast
• Agency utilisation rate: the formula, and the denominator that decides it