Competitive intelligence: frameworks & methodology

Your biggest competitor cut its enterprise list price by 20% on a Tuesday. Three renewal calls sit on the calendar for Wednesday, and the reps want to know whether to match, hold, or reframe the conversation. If assembling an answer takes a week, the answer is worthless. This lesson is the production line behind that answer: where signal comes from, how to weight it, how to turn a pile of observations into a prediction, and how often to redo the work. The persistent competitor profile and the loss-review habit that the foundations lesson describes are outputs of this machine, not substitutes for it.

Framework 1: Tiering, and the refresh cadence that keeps it honest

Tier by evidence, not by brand fame. A rival earns Tier 1 when it appears in more than roughly one deal in ten across the last two quarters of competitive opportunities, or in any single deal above your top decile contract value. Tier 2 is adjacent: same buyer, overlapping problem, different primary use case. Tier 3 is trajectory-based, a collision you can argue for within 12 to 24 months. Everything else goes in a list you read once a quarter and otherwise ignore.

One edge case breaks naive tiering: the competitor that is also your channel. A hotel group competes with Booking.com for direct bookings while depending on it for a large share of demand. The same company needs two entries in your system, one tracked as a rival on direct-share and conversion, one tracked as a partner whose terms and ranking changes are a supply risk. Teams that force it into a single row end up writing competitive messaging they cannot publish without damaging the distribution relationship.

Cadence, tied to tier rather than to the planning calendar:

• Weekly: a signal digest covering Tier 1 only, three items maximum per competitor, each with a recommended action or an explicit "no action".
• Monthly: capture and diff Tier 1 pricing, packaging and homepage copy. Diffs, not screenshots in a folder.
• Quarterly: re-tier from closed-deal data, and recode the win/loss batch.
• Twice a year: rewrite the four-corners predictions and grade the previous set.
• Event triggers that override all of the above: a funding round, a pricing change, a product or sales leadership change, a regulatory decision, an acquisition.

Every claim in a profile carries a date and a source. Anything older than two quarters gets marked stale rather than quietly inherited. The common failure is a profile in which four fifths of the claims are undated, so nobody can tell which parts still hold and the whole thing gets treated as equally true.

Framework 2: The signal-to-source matrix

Every piece of intelligence has a source type and a signal type. Most programmes collect from two or three sources and treat all signals as equally reliable.

Source categories:

• Primary: buyer interviews, loss and win calls, rep debriefs, partner and analyst conversations, conference floor talk.
• Secondary: review sites, job postings, changelogs and developer docs, status pages, filings, patent applications.
• Continuous monitoring: alerts, intent data, and traffic or keyword estimators. Semrush, which sells competitive research tools and so has an interest in this category existing, is the usual choice for keyword gap and paid-search overlap work.

Treat estimator output as trend, not level. Panel-modelled traffic and keyword numbers are inferences from samples, and the error on any single competitor can be large. Use them to detect a 40% swing in a rival's paid coverage of your category terms; do not quote the absolute figure to your board as though it came from their analytics.

An underused source: regulatory and compliance documents. When the European Commission designated Booking.com a gatekeeper under the Digital Markets Act in May 2024, it created a public schedule of obligations the company had to meet, with deadlines. For anyone competing with or distributing through it, that is a calendar of forced product and contract changes visible months ahead of the changes themselves. Antitrust filings, prospectuses and consultation responses work the same way: rivals describe their own strategy in language they cannot walk back.

Signal types worth tracking: messaging shifts, product signals (changelog entries, beta invitations, new APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → endpoints), go-to-marketgo-to-marketThe strategy defining how you'll launch a product: target segments, channels, value proposition and coordinated action plan.View full definition → signals (SDR hiring by region, partner announcements, sponsorship changes) and financial signals.

Score each source on two properties: latency, meaning the gap between the event and the moment you could observe it, and falsifiability, meaning whether a second person could check it. A rep anecdote has near-zero latency and near-zero falsifiability. A filing has high falsifiability and months of latency. The working rule: no single low-falsifiability signal changes positioning or reaches a battlecard. Require two independent sources of different types.

Reciprocity is the part teams forget. Your own postings, docs site and pricing page are read exactly the way you read theirs. If you would infer a roadmap from a rival hiring a billing platform engineer, assume someone infers yours.

Framework 3: Four-corners analysis

Michael Porter's four-corners model, from Competitive Strategy (1980), exists to answer a question that a feature comparison never touches: what will this competitor do next, and how will they respond if we move? Four boxes.

• Drivers: what management is rewarded for. Growth, margin, a specific segment, an exit.
• Current strategy: what they are actually doing now, read from pricing, hiring and shipping rather than from their keynote.
• Assumptions: what they believe about the market and about you. This is where the exploitable errors live.
• Capabilities: what they can execute, including data, distribution and balance sheet.

Worked as an exercise on Intercom: its current strategy is legible from pricing, where the Fin AI agentAI agentSoftware that pursues a goal on its own: it plans steps, uses tools and takes actions with limited human input.View full definition → is billed per resolved conversation, around a dollar a resolution, rather than per seat. Capabilities include a large corpus of support conversations and an installed base already routing volume through its inbox. The implied assumption is that buyers will accept paying for outcomes on support volume that used to be priced by headcount. From that you write predictions with dates and disconfirmers: if the assumption holds, expect the outcome-priced motion to spread to adjacent surfaces; if it fails, expect a retreat to seat-based bundling. Both are checkable within two quarters.

The failure mode is filling in four boxes and stopping. A four-corners with no prediction attached is a book report. Each corner has to end in a sentence of the form "we expect X by Y, and we would know we were wrong if Z". Grade them at the six-month refresh. Analysts whose predictions never resolve are not doing analysis.

Framework 4: Coding win/loss so it aggregates

Interviews only become intelligence when they are coded consistently. Build a controlled vocabulary of ten to fifteen loss reasons, no more, and force one primary reason per deal with up to two secondary. Free text alongside, never instead. Fixed fields: competitor named, deal size band, segment, stage at which the deal turned, who introduced the competitor into the process, and who conducted the interview.

Reliability matters more than volume. Double-code a tenth of interviews with a second person and compare; below roughly 80% agreement on the primary code, the codebook is ambiguous rather than the coders careless. Fix the definitions before you analyse anything.

Two sampling traps. First, if you interview only competitive losses, you will over-detect feature gaps and under-detect the status quo, because losses to "no decision" and to deferred budget rarely get scheduled. Include won deals and no-decisions in the sample. Second, debriefs run by the account executive who lost the deal skew hard toward price, which is the explanation that costs the rep the least.

Volume sets the cadence, not the other way round. If you close 40 competitive deals a quarter, a quarterly batch will not distinguish a pattern from noise; move to a rolling four-quarter window and refresh it monthly. A second-order finding surfaces almost immediately in most programmes: the competitor logged in the CRM is often wrong, because reps record the incumbent they know rather than the vendor that actually won. Coding catches that mismatch, and correcting it changes which battlecards get maintained.

Real-world cases

Case 1: Booking.com and the limits of copying what you can see. Booking.com has long run experimentation at industrial scale, with engineers describing over a thousand concurrent tests. Because those tests are live in production, a rival can observe any variant it happens to load. What it cannot observe is which arm won. Copying a competitor's visible interface copies a hypothesis, not a result, and roughly half of tested variants at any serious experimentation programme lose. Treat observed UI as evidence of what a competitor is questioning, not of what they have concluded.

Case 2: Semrush and the analyst's blind spot. A vendor selling competitive research runs on the same estimator data it sells, which makes its own category read sharp on search visibility and weaker on everything that never touches a search engine: procurement shortlists, partner-led deals, private pricing. Any team leaning on one instrumented channel inherits its shape. Pair every tool-derived read with primary interviews, or you will confidently describe the part of the market your tooling can see.

CMO action items

• Name one owner in product marketing for the weekly Tier 1 digest, delivered the same morning each week to the CMO, CROCROConversion Rate Optimization (CRO) is the systematic practice of increasing the percentage of users who complete a desired action, using data, testing, and user research.View full definition → and CPO, with an explicit action or a stated "no action" per item.
• Put a date and a source on every claim in every competitor profile, and run a staleness sweep each quarter that deletes rather than archives.
• Grade last cycle's four-corners predictions in the same session where you write the new ones, and record the hit rate.
• Bring coded win/loss patterns to sales leadership with the specific messaging and battlecard changes they imply, not the raw quotes.

Common mistakes that kill CI programs

• Treating CI as a deliverable: a landscape report circulated once and filed. A static report is worse than none, because it creates confidence in claims nobody dated.
• Collecting without a decision in mind: dashboards full of competitor data that inform nothing. Write down the three to five decisions the data is meant to improve before you collect anything.
• Letting sales own CI: reps are paid to win the deal in front of them, so their read is deal-specific and tends to confirm the pitch they already give. Product marketing owns the method, sales supplies structured input.
• Measuring the programme by output volume. Digests sent is not a metric. Predictions graded, battlecard claims retired as stale, and win-rate movement against a named Tier 1 competitor are.