Most competitor research produces a document nobody reads, written by someone who spent two hours on the competitor’s website and came back having absorbed their positioning. You can tell it happened because the summary uses their words. “They’re going after the enterprise segment with an AI-native platform.” That is not a finding. That is their homepage, retyped.
The fix is a rule and a running order. The rule is this: read artifacts, not assertions. Rank every source by how expensive it would be for them to fake, and spend your time at the expensive end.
The source ladder
Cheapest to fake at the top. Most expensive — and therefore most informative — at the bottom.
- Homepage and marketing copy. Free to change on a Tuesday afternoon. Tells you positioning and nothing else. Useful only as a baseline to compare against everything below it.
- Blog, social, conference talks. Cheap. Tells you what they want attributed to them.
- Pricing page. Moderately expensive — they have to honour it. Tells you their theory of value.
- Support docs and API reference. Expensive. Documentation only exists for things that exist.
- Changelog and release notes. Expensive. Somebody built each line.
- Review sites and app store reviews. Customer-generated. Expensive to fake at volume, and the negative ones are nearly impossible to fake.
- Paid ad libraries. They are spending cash per impression. Nobody funds a message they do not believe converts.
- Job postings. Very expensive. A posting is a salary commitment against a roadmap that has already been approved.
- Filings, if public. Legally binding. The floor of the ladder.
Notice that the sources most teams start with are the four at the top, and the sources that actually predict behaviour are the five at the bottom. That inversion is the whole playbook.
The 90-minute run
Timebox it hard. The value of this exercise collapses after ninety minutes because you start reading instead of counting. Set a timer for each block.
0–10 min — Pricing page
Screenshot it, dated. Then record three things:
- The tiers and the price of each.
- The gating metric — what number do you have to buy more of? Seats, usage, volume, workspaces, records, API calls?
- The fence — which features are held back for the top tier?
The gating metric is the single most revealing artifact a company publishes. It is their theory of where their value comes from, stated in a number they have to defend. A company that charges per seat believes it sells to teams. A company that charges per unit of usage believes it sells to workloads, and is betting its costs scale with yours. When a competitor changes the gating metric, they have changed their entire theory of the business — that is a bigger event than any feature launch, and almost nobody notices it.
10–25 min — Job postings
Pull every open role. Do not read the descriptions yet. First, count them by function: engineering, sales, marketing, support, ops, data. Write the counts down.
The shape of the hiring plan is the roadmap, six to twelve months early. Nine engineers and one salesperson is a company building. Two engineers and nine salespeople is a company harvesting. Four support roles appearing at once is a company with a load problem it has decided to solve with headcount rather than product.
Now read the engineering titles only. Titles name systems. “Senior engineer, billing platform” means the billing platform is being rebuilt, which usually means pricing is about to change. A first-ever “solutions architect” or “implementation” role means they have started closing deals big enough to need hand-holding — they are moving upmarket, whatever the homepage says. Two infrastructure roles and a security engineer usually means a compliance certification is in flight, which means an enterprise motion is coming.
25–40 min — Changelog
Take the last six months of release notes. Tally each shipped item into one of four buckets: new surface (a thing that did not exist), depth (an existing thing got better), integration, or maintenance.
The ratio tells you where their engineering capacity actually goes, which is frequently not where their marketing says it goes. A changelog that is 70% maintenance is a company paying down debt and unable to move. A changelog that is 60% integrations is a company that has decided distribution beats product. A changelog with three new surfaces in six months and no depth is a company chasing.
Also note the cadence. Weekly shipping that goes quiet for six weeks is a re-platforming, an acquisition, or a layoff. All three are worth knowing about.
40–55 min — Reviews, negative first
Go to the review sites. Sort by lowest rating. Filter to the last six months — older complaints may have been fixed and will mislead you.
Do not read for sentiment. Read for repeated nouns. One person complaining about onboarding is noise. Eleven people using the word “export” in ninety days is a structural gap, and it is a gap their own team almost certainly knows about and has deprioritised. That is a durable weakness, not a temporary one.
Then read the five-star reviews for the same thing: which noun keeps appearing in the praise? That is the feature you cannot win on. Route around it.
55–70 min — Paid ads
Most large ad platforms publish a public transparency library of currently-running ads. Look up the competitor and read what they say to cold traffic.
Compare it against their homepage. When the two disagree, the ads are true. The homepage is written by committee for existing customers, investors and recruits. The ads are written by someone whose budget dies if the message does not convert. If the homepage says “the collaboration platform for modern teams” and every ad says “stop paying for seats you don’t use,” you have just learned that their actual wedge is price, and that they know it.
70–85 min — Fill the scorecard
Same nine rows every time, for every competitor. Consistency is what makes these comparable across quarters.
- Gating metric: ______ (changed since last quarter? Y/N)
- Entry price / top price: ______
- Open roles by function: eng __ / sales __ / mktg __ / support __ / ops __
- Hiring shape: building / harvesting / firefighting
- Changelog mix: surface __% / depth __% / integration __% / maintenance __%
- Ship cadence: steady / accelerating / stalled since ______
- Top repeated complaint (6mo): ______ (count: __)
- Top repeated praise (6mo): ______ (count: __)
- Cold-traffic message vs homepage: aligned / diverged — how: ______
85–90 min — The verdict
Three sentences. No more. If it takes more, you have not finished thinking.
- What they are betting on. One sentence, in your words, not theirs.
- Where they are structurally weak. Not “their UI is dated” — something the scorecard supports.
- What we do about it in the next 30 days. One action, owned, dated. If there is no action, the teardown was entertainment.
Worked example
Say you sell a mid-market analytics tool and you are tearing down a rival — call it Sightline — that everyone on your sales team is scared of.
Pricing. Three tiers: $99, $399, “contact us.” Gating metric moved from seats to monitored data sources at some point in the last quarter — the archived version of the page still says per-seat. Fence: SSO, audit log and role permissions are all top-tier only.
Jobs. Eleven open roles. Engineering 3, sales 5, marketing 1, support 2. Two of the sales roles are “Enterprise Account Executive.” One engineering role is “Senior Engineer, Billing.”
Changelog. Six months: 4 new surfaces, 3 depth items, 9 integrations, 14 maintenance. Cadence steady until eight weeks ago, then thin.
Reviews. Fourteen mentions of “slow” on large datasets in six months. Praise clusters hard on “setup” — people repeatedly say they were live in an afternoon.
Ads. Homepage says “enterprise-grade analytics.” Every running ad says “live in 20 minutes.”
The verdict writes itself. Sightline is betting that fast setup wins the deal and that they can hold the account by moving billing to data sources so the bill grows without a renegotiation. They are structurally weak on performance at scale — fourteen complaints in six months, and a changelog with almost no depth work says they are not fixing it. Next 30 days: we build a large-dataset benchmark, publish it, and arm sales with a two-question qualifier about data volume — owned by Priya, due the 30th.
Notice what happened. The scary competitor turned out to have a specific, documented, durable weakness, and your sales team now has a question to ask on the first call instead of a feeling to be scared of.
Our take: The single highest-yield line in this whole process is the gating metric. Everything else tells you what a competitor is doing; the gating metric tells you what they believe. And because changing it is expensive and public, a change there is the closest thing to a leaked strategy memo you will ever get legally, for free, from a page they published themselves.
Cadence
Once a quarter, per competitor, one page each. Keep the pages. The value is not in any single teardown — it is in the diff. Four quarters of scorecards will show you a trajectory that no single snapshot can: hiring shifting from engineering to sales, the changelog sliding into maintenance, the entry price quietly rising. That is how you see a competitor turning before they announce it.
Two or three competitors is the right number. Five is a research department you do not have.
Seven ways this goes wrong
- You read their marketing first. It anchors everything after it. Start at the pricing page, and do not open the homepage until you have filled in the scorecard.
- You mistake motion for strategy. Shipping a lot is not the same as shipping toward something. Four unrelated new surfaces in six months is a company with no thesis, not a company beating you.
- You count features instead of weighting them. A competitor with sixty features and yours has forty tells you nothing. One feature that removes a reason to buy you outweighs twenty that do not.
- You do the teardown once, in a panic. Teardowns done the week you lost a big deal are the least useful ones, because you already know the conclusion you want. Do them on a schedule, when nothing is on fire.
- You skip the negative reviews because they feel like gossip. They are the only place customers describe the product without either company’s marketing in the room.
- You produce a document instead of a decision. No owner and no date means it did not happen. The three-sentence verdict is the deliverable; the rest is working notes.
- You start copying. The most common outcome of a good teardown is a roadmap that slowly turns into theirs. The scorecard is for finding the gap they are not defending — not for closing the gaps they are.
The 20-minute version
If ninety minutes is not going to happen, do this: pricing page, job postings, negative reviews. Twenty minutes, three sources, all near the expensive end of the ladder. You will get most of the signal and none of the fatigue — and a short teardown you actually run every quarter beats a thorough one you run once.