The distinction is simple and constantly muddled in practice. Secondary intelligence comes from material that already exists. Primary intelligence comes from asking people. Both are legitimate, both are necessary, and the sequencing between them is where most of the money is wasted.
What secondary sources are good for
Establishing the factual skeleton, cheaply and repeatably. Almost everything structural about a competitive landscape is in public documents if you are willing to read them.
| Source | Answers well | Cannot answer |
|---|---|---|
| Trial registries ClinicalTrials.gov, CTIS, ICTRP |
Design, endpoints, enrolment criteria, site count, timing, and the amendment history | Whether enrolment is actually on track, or why a criterion changed |
| Congress abstracts | Top-line efficacy against an older data cut | Safety detail, subgroup consistency, how the field received it |
| Regulatory documents Review packages, CHMP opinions, adcom materials |
What regulators questioned and how sponsors answered — often the richest single source | Anything not yet filed |
| Earnings calls and filings | Stated priorities, spend direction, hedged language about timelines | What was deliberately not said |
| Patents | Formulation and method-of-use intent, exclusivity runway | Commercial intent — filing is cheap relative to developing |
| Job posts, sponsorships, site activations | Where capability is being built, ahead of announcements | Scale or seriousness |
Underused: the amendment history. A registry entry's revision log is one of the highest-signal, lowest-cost sources available. A sample size increase, a widened enrolment criterion, a moved primary completion date, a changed primary endpoint — each of those is a decision a sponsor made under pressure, visible for free, and most teams only ever read the current version of the record.
What only primary can tell you
Four things, reliably:
Whether a result is clinically meaningful. A statistically significant progression-free survival benefit of six weeks may or may not change what an oncologist does on Monday morning. The published paper cannot tell you which; a treating physician can, in one sentence.
What the trial looked like from inside. Investigators know things the manuscript does not carry — how much dose reduction was really needed, how the toxicity presented, which patients they stopped enrolling and why.
How a claim will be received by payers. Formulary decision-makers will tell you which comparator they consider relevant and what evidence gap they will cite. That is not derivable from a label.
Expectation. What informed people think is going to happen next, and why — the input that turns a landscape into a forecast.
The sequencing that saves money
Primary research is expensive per unit of information and slow. Secondary is cheap and fast. So the order matters, and the common failure is running them the wrong way round: commissioning fifteen expert interviews and spending a third of them establishing facts that were sitting in a registry.
- Exhaust secondary first. Build the factual skeleton. Know the trial design, the amendment history, what the regulator asked, what the sponsor said on the last three earnings calls.
- Write down what you still cannot answer. Specifically. "Will this change practice in second line, and what would have to be true for it to" is a primary question. "What is the primary endpoint" is not.
- Design the interview around those gaps. An hour with a trial investigator is a scarce resource. Spending twenty minutes of it on background is a waste of both parties' time and it signals you did not prepare.
- Feed findings back into the record. Otherwise the same question gets commissioned again next year.
The compliance line, stated plainly
Primary research in pharma CI sits close to a boundary that has to be handled explicitly rather than by instinct.
Gathering public information, and asking knowledgeable people for their opinions and their own experience, is legitimate and routine. Inducing someone to disclose information they are contractually obliged to keep confidential is not — and that does not become acceptable because an intermediary did the asking, or because the interviewee volunteered it, or because the topic was approached obliquely.
What that requires in practice: a written elicitation policy, interviewers trained on it, screening that excludes current employees of the competitor in question, explicit instructions to interviewees not to disclose confidential information, and a documented process for what happens when someone starts to anyway. Functions that treat this as an ethics slide rather than an operating procedure are carrying real risk.
Where the two are hardest to combine
Oncology, again — because the gap between the published result and the clinical reality is unusually wide, so the secondary record understates how much interpretation is needed. A landscape built purely from registries and abstracts in a competitive tumour type will look complete and be misleading.
Maintaining both halves at indication level across a portfolio is a real staffing problem. A congress analyst briefing is one place the combined form is visible — the ASCO 2026 analyst briefing, for instance, sits on top of the presented data rather than replacing it. Whether you build or buy, the test is the same: can the function tell you not just what the data said, but whether the people who generated it think it matters.