Primary Research for Hedge Funds: A Practitioner's Guide

The edge problem

Sell-side research is commoditised. Alt-data is widely subscribed. Transcripts are searchable in seconds. LLMs summarise 10-K filings faster than any associate can. The half-life of public information has collapsed, and with it the half-life of consensus.

What's left? Information you originated. Questions only you have asked. Data points only you have collected. That is what primary research delivers — and it is increasingly the only durable source of edge in fundamental investing.

This guide is a practitioner's playbook for hedge fund analysts on how to use primary research — expert interviews, B2B surveys, and channel checks — to build, validate, or kill investment theses. It is opinionated. We will tell you where most analysts go wrong, what the best ones do differently, and how to think about the economics of doing this work yourself versus outsourcing it.

What primary research actually is (and isn't)

Primary research is original, first-hand information gathered directly from market participants — customers, competitors, former executives, channel partners, suppliers, regulators — to inform a specific investment thesis.

It is not:

• Secondary research — sell-side reports, filings, transcripts, news.
• Alt-data — credit card panels, web scraping, app usage. Useful, but not originated by you and increasingly priced into consensus.
• Expert network access — an expert network is a sourcing layer. The research itself is the workflow built on top: scoping, recruiting, instrument design, fielding, synthesis, decision.

The methods that matter for hedge fund analysts:

1. Expert interviews — 30–60 minute structured calls with former execs, customers, channel partners.
2. B2B surveys — quantitative instruments fielded to 30–300 vetted respondents.
3. Channel checks — recurring qualitative pulses on distributors, resellers, customers.
4. Voice-of-customer panels — longitudinal cohort tracking on retention, NPS, vendor displacement.

The thesis-first principle

If you take one thing from this guide: write the thesis before you commission any research. Not after. Not "as you go." Before.

The one-page thesis memo should contain three things:

1. Variant perception. Where exactly does your view differ from consensus? "I'm more positive than the Street" is not variant perception. "Consensus models 8% NRR compression in FY25; my work suggests <2%" is variant perception.
2. KPI tree. Decompose the variant perception into operational drivers your research can directly test — price, volume, mix, attach rate, churn, win rate, share of wallet.
3. Pre-mortem. Write down what would have to be true for you to be wrong. Then design the research to test those failure modes — not the success modes.

Best-in-class analysts run disconfirming research: research designed to break the thesis, not confirm it. Most analysts do the opposite without realising. If you commission five expert calls to "learn more about Company X," you will find five reasons to believe whatever you already believe.

Choosing the right method for the question

A rough decision matrix:

The most common mistake is method-by-default: doing expert calls because that's what your fund's EN subscription is set up for, even when a survey would produce a sharper answer at lower cost.

Running expert calls well

The median expert call is mediocre. Not because experts are bad, but because the analyst showed up unprepared. A few rules:

• Pre-call work is non-negotiable. Read the expert's history. Know the public facts about their former employer. Don't burn 20 of your 45 minutes on context.
• Structured discussion guide — 6–10 anchor questions, with branching follow-ups. Time-box each section.
• Ask the same question three different ways. Experts often give the rehearsed answer first. The third version is where you learn something.
• Score every call. Signal/noise rating, novel insights flagged, contradictions logged. Over time this curates your roster — the top 20% of experts produce 80% of the alpha.
• Triangulate. Never trade off one call. Require two independent sources for any thesis-critical claim.
• Avoid the charismatic-expert trap. One persuasive former CRO can derail a thesis. Weight evidence by triangulation, not by how compelling the speaker was.

And on volume: more calls do not equal better research. Five well-structured calls with the right people beat 20 random ones. Most analysts confuse activity with insight.

Running B2B surveys well

Surveys are systematically underused by hedge fund analysts, especially in software, where customer-level quantification can be devastating to (or supportive of) a thesis. The reasons people avoid surveys — "they're slow," "samples are flaky," "I don't trust the design" — are real but solvable.

What good survey design looks like:

• Tight screening. Pay for fewer, better-qualified respondents. n=50 of verified decision-makers beats n=300 of self-attested "users."
• Forced-choice questions, not open-ended speculation. "Which vendor are you most likely to renew with in the next 12 months?" not "How do you feel about Vendor X?"
• Always include competitive benchmarks. Never ask about Vendor X without asking the same question of Vendors Y and Z. Absolute scores are meaningless; relative scores are tradable.
• Wave it. A single point-in-time survey is worth a fraction of the same instrument repeated quarterly. Trend is the signal.
• Watch non-response bias. Who chose to answer? Are they systematically different from who didn't?
• Right-size the sample. A 10-point NRR delta needs n=30–50. A 2-point pricing delta needs n=150+. Don't draw confident conclusions from n=8.

Channel checks

Channel checks aren't just for retail. Any product that moves through a distribution layer — SaaS resellers, medtech distributors, semiconductor channel partners, QSR franchisees — can be channel-checked.

What channel checks predict well:

• Near-term sell-through and inventory levels
• Competitive displacement at the point of sale
• Discounting and promotional pressure
• New-product reception

What they predict poorly:

• Anything happening upstream of the channel (e.g., R&D pipeline)
• Quarterly print precision in noisy categories
• Long-cycle enterprise deals that don't touch the channel

The biggest mistake: trading a single channel-check data point as if it were the quarter. One reseller saying "demand is soft" is anecdote. Twenty resellers saying it across two waves is signal.

Synthesis and decision

After 10–15 expert calls and a survey, you have 8+ hours of transcripts and a cross-tab deck. The synthesis is where most projects fail — not because the data was bad, but because the analyst never converted it into a position-size decision.

A good synthesis memo:

• One-sentence headline up top. The variant perception, confirmed/refuted/modified. PMs read the first line.
• Confidence-tag every claim. High / Medium / Low based on sample size and triangulation.
• Distinguish fact / opinion / prediction. "Vendor X lost the renewal" is fact. "Their product is weaker" is opinion. "They'll lose more next year" is prediction. Treat each accordingly.
• Quote precisely, sparingly. One sharp quote beats five paraphrases.
• End with a decision. Build / size up / size down / kill. If the research doesn't change anything, the research wasn't worth doing.

Compliance and MNPI

The post-SAC, post-Visium environment is unforgiving. Every multi-manager and most single-managers now require recorded calls, pre-call screening, restricted-list checks, and documented compliance attestations. Treat this as table stakes, not friction.

Practical do's and don'ts:

• Pre-screen every expert for current employment status, cooling-off compliance, and MNPI exposure.
• Record and retain every call. Non-negotiable.
• End the call the moment an expert says "I probably shouldn't tell you this," shares forward guidance, or references unannounced numbers.
• Mosaic theory is your friend — combining many non-material public and non-public pieces into an actionable thesis is legal and defensible. Trading on a single material non-public data point is not.
• Integrate restricted lists with every research vendor.

Properly structured workflows don't slow research down — they speed legal review and protect the analyst.

Build vs. buy vs. done-for-you

The honest economics. A pod-shop analyst covering 20 names, fully loaded at $500K–$1M annual cost, has roughly 10 working weeks per name per year. Of that, maybe 4 weeks per year on deep primary work on each high-conviction name. Time spent scheduling expert calls, writing discussion guides, fielding surveys, and synthesising transcripts is time not spent generating ideas, building models, or talking to the PM.

Three operating models exist:

1. DIY via expert network — buy a GLG / AlphaSights / Third Bridge / Guidepoint subscription, schedule calls yourself, synthesise yourself. Default for most pod shops. Works when your time is cheap and the calls are simple. Breaks down on multi-call projects and surveys.
2. In-house primary research team — larger funds (Citadel, Point72, Coatue) staff dedicated research-ops teams. High fixed cost; works only at scale.
3. Done-for-you / managed primary research — brief a provider on the thesis questions; they recruit, interview, survey, and deliver a synthesised output. Project-priced. Best fit for multi-call projects, surveys, and channel checks where the analyst's time-to-decision matters more than per-call cost.

Most funds run a hybrid. The right question is not "which model," it's "which model for which kind of question." Quick single-call validations: expert network. Multi-call thesis builds, surveys, channel-check programs, catalyst sprints: done-for-you generally wins on fully-loaded analyst time.

AI in the research workflow

What AI does well today:

• Drafting discussion guides from a thesis memo
• Summarising and tagging transcripts
• Extracting themes across 10+ calls
• Searching prior research libraries
• Sentiment and contradiction flagging

What still requires humans:

• Recruiting the right former VP of Pricing at the target company
• Asking the unscripted follow-up question that breaks open a call
• Judging expert credibility
• Reconciling contradictory data points into a position-size decision
• Compliance judgment in real time

AI flattens the cost of synthesising public information. It does not originate primary data. The moat shifts further toward whoever asks the best questions of the best-recruited respondents — not whoever owns the best LLM.

A one-page playbook for your next thesis

1. Write the thesis first. Variant perception, KPI tree, pre-mortem. One page.
2. Map the ecosystem. Customers, competitors, channel, ex-employees, regulators. Prioritise by signal-per-call.
3. Choose methods by question type. Don't default to calls.
4. Sequence the work. 3–5 scoping calls → survey to quantify → 2–3 deep dives on outliers.
5. Triangulate everything. Two independent sources per thesis-critical claim.
6. Score and curate. Track expert quality over time.
7. Synthesise to a decision. One-sentence headline. Confidence-tagged claims. Build / size / kill.
8. Compliance discipline. Pre-screen, record, document. Every call. Every time.

Closing thought

Primary research is no longer a nice-to-have for fundamental hedge funds — it is the discipline that separates funds that generate edge from funds that pay for consensus. The funds that win over the next decade will be the ones that ask sharper questions, recruit better respondents, design tighter surveys, and convert it all into decisions faster than the next desk.

If you'd like to talk through how we run primary research projects for hedge fund analysts — expert interviews, B2B surveys, and channel checks delivered as finished outputs rather than raw calls — get in touch with the Woozle team.