Building an In-House Expert Network? Sourcing Is the Easy Part

This spring a US family office advertised for a senior primary research associate. The job was to build lists of experts on the companies the investment team cares about, find their contact details on LinkedIn and in contact databases, reach out cold, follow up by phone, interview them, pay them, and write up the transcripts. The same person would also coordinate the calls the firm still books through third-party networks.

That job description is a fair summary of a conversation we keep having with analysts. A fund still uses its expert networks, but it has started sourcing part of its research itself. One team we spoke with recently wanted to reach everyone who had left a target company in the previous six to twelve months. For a mid-sized company that came to somewhere between 20 and 200 people.

None of this is new in principle. Industry analysts were writing about hedge funds building their own expert networks back in 2008, and at least one European long/short fund has kept a 20-person in-house research team for over fifteen years. What has changed is the cost of the first step. Finding 150 former employees of a company used to take a researcher weeks. Today a contact database and a LinkedIn seat get you most of the way in an afternoon.

The trouble starts after the list exists.

Why teams are doing it

The economics are well known to anyone who has signed off on a research budget. Standard calls through the large networks commonly cost $1,000 to $1,500 an hour. In one recent survey of more than 1,300 experts, most said they received less than a fifth of what the client was billed. A few calls per name was always easy to justify on those numbers. A 60-person sweep of former sales staff to check whether a new product is actually taking share is harder.

There is a quality argument too. Network calls draw on the same pool of experts that every other fund is calling, and analysts complain about experts who tell them what they want to hear. Sourcing your own list lets you pick the people who were closest to the question: the regional sales lead who left in March, the distributor who switched suppliers, the customer who churned. Those people are often not registered with any network.

Transcript libraries also cover the most-researched names now, which pushes paid calls toward less obvious questions. Those are often the ones where your own list beats a network's.

The expert network industry is still growing, for what it's worth. Industry estimates put it at roughly $3 billion in 2025. Teams building their own sourcing are adding a second channel, and they are keeping the first.

What the network was quietly doing for you

A network fee pays for more than an introduction. Before a call happens, a network typically checks that the expert isn't currently employed by the company you're asking about, applies a cooling-off period for people who left recently (six months is a common industry standard, though no regulation sets it), gets the expert to agree to terms and attest that they won't share confidential information, and gives your compliance team a way to approve, block, and review calls.

When you source the expert yourself, every one of those jobs comes back to you.

This matters more than it sounds. Every investment adviser, registered or not, has to keep written policies designed to prevent misuse of material non-public information, and those policies have to fit how the firm actually does research. In an April 2022 risk alert, SEC examination staff described advisers that lacked adequate procedures around expert network consultants, including:

Tracking and logging calls with expert network consultants; Reviewing detailed notes from expert network calls; and Reviewing relevant trading activity of supervised persons in the securities of publicly traded companies that are in similar industries as those discussed during calls.

The alert is staff guidance, not a rule, and it is written about network calls. But a fund that pays former employees for their time directly will have a hard time arguing those expectations don't apply. Talking to experts is legal. What your compliance team will want is evidence that each conversation was screened and logged. An analyst's own contacts carry no automatic exemption either. The SEC has settled with an analyst who got inside information from a friend who was an executive, with no network involved.

The usual objection to in-house programs, one the large research vendors make often, is that they end up with no compliance infrastructure and no institutional memory. For a program run on analysts' personal calendars and personal notes, that objection is correct.

(This is not legal advice. Your compliance team decides what your program needs.)

The arithmetic nobody puts in the plan

Say the sourcing works. You build a list of 150 former employees and 40 agree to talk.

At an hour each, that is 40 hours of calls before anyone has scheduled them or written them up. Add that work back and a large share of an analyst's month is gone. One private equity associate on an industry forum described doing seven calls in two days and never getting near the model. This plan asks them to do that for weeks, for one company.

In practice the list gets cut. The analyst calls the 8 people who look most senior, and the other 32 who agreed to talk never hear back. The sweep turns back into a handful of calls, which is exactly the network model the team was trying to get past, minus the compliance support.

There is also a comparability problem. By the fifteenth call on the same question, the interviewer has changed. The questions have drifted, the follow-ups depend on how tired the analyst was, and the notes from call 3 and call 15 are not written the same way. You cannot count how many of 40 people said the new product was winning deals if the question was worded six different ways.

Where AI-moderated calls fit

This is the gap AI-moderated calls fill for in-house programs. Several of the largest networks have launched their own versions in the past year or so, for much the same reason. The interviewer is the constrained resource. So take the interviewer off the critical path for the calls that don't need one.

In practice, the team keeps doing what it is good at: choosing the question and finding the people. Each expert gets a link and takes the interview in the browser whenever it suits them, or picks up a phone call at a time you set. The agent runs the same interview with each of them.

A few things change as a result.

Screening comes first. Your conflict questions go first: current employer, when they left, any restrictions they are under. They work as a gate rather than a checklist step. If an expert doesn't meet your criteria, the call ends at screening. Recording consent is captured the same way, before the substantive conversation starts.

Every call is on the record. Each interview produces a recording, a transcript, and a summary tied to that expert. That is the call log and the detailed notes the SEC alert describes, and it didn't depend on anyone remembering to write them.

Forty calls become one dataset. The questions are asked the same way each time, with follow-ups when an answer is vague or interesting. You can line the answers up and count them. For a channel check, that is the point of the exercise.

The evidence travels. Most analysts are building a case for a PM or an investment committee that wasn't on the calls. A quote with a timestamp and a transcript behind it makes that case better than a paraphrase in a memo does.

The record outlives the analyst. Transcripts stay with the firm, which answers the institutional memory objection. That matters most in a program like the one in that job posting, where one person holds most of the relationships.

A sensible way to start is with screening calls: a short conversation that confirms each expert actually held the role and knows the topic. That step costs the most analyst time and needs the least judgment. Once the screening questions are tuned, run full interviews with the people who pass.

To be clear about scope: we run the interview. Finding the experts, deciding who to pay and how much, and making the compliance calls stay with your team.

What should stay with the analyst

AI-moderated calls are not a good fit for every conversation, and there is real skepticism worth taking seriously. User research practitioners warn against AI interviewers where the conversation depends on deep domain knowledge. In one expert survey, most of those who had been on an AI-moderated call rated it below a human-led one. That survey was run by a company selling human-led research, and the sample was small, but the finding deserves weight.

Our view is that the split is fairly clean. The former CFO who can explain why margins really fell deserves an hour of your best analyst, who can pick up on what isn't being said and push through the sixth and seventh follow-up question. So does the call that decides whether the position gets sized up. Those are conviction calls, and a busy analyst should protect time for them.

The 40-person sweep is different. Its value is in the breadth, and asking an analyst to sit through all 40 is how it gets cut to 8. AI-moderated calls work best as the first pass. They tell you which three of the 40 are worth an hour of your time.

One more limit. A good interview can't fix a bad list. If the people you sourced were never close to the question, a consistent transcript will just record that consistently.

Networks stay in the picture

None of this means dropping your networks. They remain the fastest way to reach people you can't find yourself: senior executives at private companies, say, or specialists in a market where your contact data is thin. The teams we talk to keep both channels, and the family office in that job posting does too.

What changes is the division of labor. The network handles hard-to-reach experts. Your own sourcing covers the long list of former employees, customers, and suppliers that no network would ever sell you by the dozen. And the interview is set up once and runs the same way whichever channel the expert came from.

If you are building this inside your firm, spend less of the plan on the list. Spend it on who runs the interviews, how each expert gets screened, and where the record lives.

Planning an in-house expert program? Read how direct investors use InsightAgent on live diligence, or try a short interview with the agent yourself.

Talk to the agent — live