Most CI reports are read once, filed, and never acted on. Not because the research was weak, but because the report stopped one or two levels short of being useful. The analyst gathered facts, maybe spotted a pattern, and then handed the "so what" over to the reader. Readers do not do that work. They skim, nod, and move on.
The CI Report Pyramid is a simple test for whether a report will drive a decision or decorate a shared drive. It has four layers, and every layer has to be earned from the one below it.
Data > Insight > Implication > Action.
Each layer answers a different question. If you cannot answer the question, you have not reached that layer, no matter what your slide title says.
Layer 1: Data. "What happened?"
Data is the raw, verifiable material: filings, job postings, pricing pages, import records, press releases, customer reviews, conference remarks. It is the only layer where the standard is factual accuracy, and the standard is absolute. Every item here should be sourceable, dated, and labeled as fact.
This is also where most CI reports live and die. A 30-page competitor profile that lists everything a rival did last quarter is a data dump wearing a report's clothing. It feels rigorous because it is dense. It is actually unfinished work.
Running example. You cover the industrial pumps market. Over six weeks you log the following about a competitor, Veltrax Flow Systems: Veltrax posted 14 openings for field service technicians across the US Gulf Coast; it announced a distribution agreement with a regional MRO supplier in Texas; its latest price list shows a 6 percent reduction on mid-range centrifugal pumps; and its CEO mentioned "aftermarket revenue" four times in the last earnings call, up from zero mentions a year ago.
Four facts. All checkable. None of them, on their own, tell your leadership anything worth a meeting.
Layer 2: Insight. "What changed, and what pattern connects it?"
Insight is where separate facts become one finding. The test: can you state, in one sentence, a pattern that a reasonable reader could not have seen from any single data point alone?
This is also where intellectual honesty starts to matter. An insight is an inference, not a fact, and it should be labeled as one. The discipline of writing "we infer" instead of "Veltrax is" keeps you credible when you are wrong, and you will sometimes be wrong.
Running example. The four facts converge: We infer Veltrax is shifting from selling pumps to selling uptime, building a Gulf Coast service footprint and using price cuts on new units to seed an installed base it can monetize through aftermarket contracts.
That is a real finding. Hiring alone could mean growth. Price cuts alone could mean weakness. Together, with the earnings language, they point at a strategy change. Notice the sentence is falsifiable; future evidence can confirm or kill it. That is what separates insight from narrative.
Layer 3: Implication. "What does this mean for us, specifically?"
An insight about a competitor is still about the competitor. The implication layer translates it into consequences for your company: your accounts, your margins, your roadmap, your pricing corridor. The test: does the sentence contain your company as the subject or object? If it could appear unchanged in a rival's report, it is not yet an implication.
This layer is where analysts get timid, because implications require judgment and expose you to disagreement. Push through. A report without implications forces every reader to derive their own, and ten readers will derive ten different ones, most of them casually.
Running example. If Veltrax converts even a quarter of its new installed base to service contracts, our Gulf Coast aftermarket revenue, roughly 30 percent of regional margin, comes under direct attack within 18 months. Our three largest regional accounts all have Veltrax units on site already, making them natural first targets for bundled service offers.
Now the reader is awake. The competitor's move has a location, a timeline, and a number attached to your own P&L. Label your assumptions here too; the "quarter of the installed base" figure is a scenario, not a forecast, and saying so costs you nothing.
Layer 4: Action. "What do we do, who does it, and by when?"
The top of the pyramid is a recommendation concrete enough to be accepted or rejected in the meeting where it is presented. The test is brutal and simple: does it name an owner, a move, and a deadline? "Monitor the situation" fails. "Consider strategic options" fails. Both are the analyst handing the work back.
Actions do not need to be grand. Often the strongest recommendation is a cheap, fast, reversible move that buys information.
Running example. Recommend: (1) Commercial team offers multi-year service agreements to our three exposed Gulf Coast accounts before end of quarter, before Veltrax's field team is fully staffed. (2) Product team prices a service-plus-hardware bundle for mid-range pumps within 60 days. (3) CI re-tests the Veltrax thesis in 90 days against two indicators: technician headcount actually onboarded, and any published service contract wins.
Point 3 matters. Building a feedback loop into the action layer is what makes CI a system rather than a sequence of one-off memos.
The two ways reports fail the pyramid
Failure mode 1: The truncated pyramid. The report stops at data or insight. It is accurate, thorough, and inert. You can diagnose it by scanning for the words "we recommend"; if they never appear, the pyramid was never finished. This failure is comfortable because it carries no risk of being wrong. It also carries no chance of being useful.
Failure mode 2: The floating peak. The opposite disease: bold recommendations resting on thin or unlabeled evidence. The action layer exists, but you cannot trace it down through implication and insight to specific facts. These reports feel decisive and age badly. When one confident-but-unsupported call goes wrong, every future report from that analyst gets discounted.
The pyramid's real function is forcing traceability in both directions. Every action should trace down to named facts. Every fact worth including should trace up toward a possible action. Facts that trace to nothing are scope creep; cut them or move them to an appendix.
The 30-second self-audit
Take the last CI report you wrote or received and ask four questions:
- Can every factual claim be sourced and dated? (Data)
- Is there at least one sentence that connects multiple facts into a labeled inference? (Insight)
- Does the report name consequences with your own company as the subject? (Implication)
- Is there a recommendation with an owner and a deadline, plus a date to re-test the thesis? (Action)
Four yeses: the report can drive a decision. Three or fewer: you know exactly which layer to build next.
The pyramid will not make your judgment better. It will make it visible, and visible judgment is what separates intelligence from expensive news summarization.