A partner at a mid-market private equity fund described the colleague he wished he had. Show that colleague a new company, say an HR software business, and they would know straight away which of its managers someone at the fund had met, and who else the fund could reach through them. "Which of their associates, former colleagues, might we know?"
The answer to his question is already in the fund's records. Over the years it has met thousands of people, and each meeting left a trace: an email thread, a calendar invite, a call note, a CRM entry. Pulling those traces together by hand takes longer than a deal allows. A company brain connects that evidence.
Getting it right matters because most deals start with a relationship. In a survey of 885 venture capitalists, firms said over 30 percent of their deals came through professional networks and another 20 percent through referrals from other investors. Only 10 percent arrived inbound. We cover the sourcing workflow in why funds treat knowledge graphs as infrastructure.
This post covers what a company brain lets a firm do with "who do we know?": answer the four questions inside it, see how strong each relationship is from the data the firm already holds, and find the route to an introduction that someone will actually make.
Four Jobs
One question, four jobs
"Who do we know?" covers four different requests.
- Lookup. Who at the firm knows this person or company, and how well?
- Pathfinding. Nobody at the firm knows the target. Who knows someone who does, and which route will work?
- Gaps. Which sectors, cities or target lists have no warm contacts? Which relationships were warm two years ago and have gone cold?
- Matching. Which person in the network fits this ask: an expert for a diligence call, a buyer for a portfolio company, a candidate for a role, a guest for a dinner?
All four come up in our work.
- A venture investor I know wants to filter her network by city before a trip. She knows a lot of people, and the long list is the problem: she wants to see who in Zurich she has spoken to.
- An executive search firm we work with is opening a forensics practice. Its consultants talk to compliance chiefs and heads of financial crime every week. A partner asked how the specialist leading the new practice would know whom to recontact.
- A fund we work with spends hundreds of thousands of pounds a year on expert networks. Some of those calls are with people the firm already knows.
The data lead at one venture firm described the order his team works in. First decide who the firm should know. Then assign each relationship to someone. Then, when a target matters, ask "who would we know that can get us a warm intro?"
Relationship Strength
What a relationship is made of
The sociologist Mark Granovetter defined the strength of a tie in 1973 as a combination of "the amount of time, the emotional intensity, the intimacy (mutual confiding), and the reciprocal services which characterize the tie." Later researchers added more dimensions. Eric Gilbert and Karrie Karahalios tested seven of them against more than 2,000 friendships their participants had rated. Their model told strong ties from weak ones with over 85 percent accuracy. Intimacy carried the most weight, and duration alone explained a large share.
The data a firm already holds covers most of those dimensions. With a company brain, each person's relationships are drawn from:
- email threads and replies, across every account they connect
- calendar invites and meeting transcripts, counted once when both exist
- LinkedIn connections, messages and group chats
- WhatsApp, when the person connects it
- who introduced whom
- shared work history from public profiles
Anyone asking then sees how strong each relationship is, judged on signals such as:
- Recency. When did they last interact?
- Intensity over time. How many interactions, across how many channels, over how many years?
- Balance. Do both sides reply, or does one side do all the writing?
- Duration. How long have they known each other?
- Shared work. Did they work at the same company, at the same time, in the same team?
Shared work history needs care. Two people who overlapped at a ten-person startup will know each other. Two people who overlapped at a company of 100,000 may never have met. So a shared employer counts for less at a big company, and for more when the two sat in the same department.
Dormant Ties
Old relationships keep their value
Most scoring systems let a relationship decay without contact. That makes sense up to a point. A friend you met every Friday for three years still counts after he changes jobs and the Fridays stop.
The research supports keeping these relationships in view. Daniel Levin, Jorge Walter and Keith Murnighan asked executives to consult contacts they had fallen out of touch with about a current work project. The advice compared well with what their current contacts gave. Reconnected strong ties offered both the new information people expect from acquaintances and the trust they expect from close colleagues.
Adam Kleinbaum studied a reorganisation that broke people's routine contact. People chose to keep the ties that were valuable, reciprocated and embedded in shared circles.
So in a company brain, an old relationship fades slowly and then holds steady. A colleague you worked with for five years and haven't emailed this year still shows up as a good route. Those dormant strong ties also make a list of their own: people worth a catch-up, ranked by what they could help with now.
Strength And Reach
Close contacts make introductions. Loose contacts reach new places.
LinkedIn ran a five-year experiment on about 20 million users, changing which connections its "People You May Know" feature suggested. Researchers from LinkedIn, MIT, Harvard and Stanford then traced which new connections led to jobs. In their words, "the stronger the newly added ties were, the less likely they were to lead to a job transmission." The most useful connections sat just above the weakest, at about ten mutual contacts.
Ronald Burt found a similar pattern inside one company. Among 673 supply-chain managers, those who linked otherwise separate groups produced better-rated ideas, earned more and were promoted more.
Close contacts are the people most likely to agree to make an introduction. Bridges are the people who can get you somewhere new. Your team should be able to see both.
My own graph shows the difference. A friend who is a venture investor links my working circle to an angel group I would never have met otherwise. On strength alone she is one close friend among many. Looked at for reach, she is the single link between two clusters, and that is why she matters.
Routes
The best route runs through people who will act
In 2003, Peter Dodds, Roby Muhamad and Duncan Watts ran the largest test of "six degrees of separation". More than 60,000 people tried to reach one of 18 targets in 13 countries by forwarding an email to someone they knew. Most chains died because someone along the way didn't pass the message on. The chains that succeeded ran through professional relationships and intermediate to weak ties, and they didn't need well-connected hubs. The authors concluded that success "depends sensitively on individual incentives."
So when your team asks for a route, the options come back ranked on five things:
- The weakest link. A route is as strong as its weakest relationship.
- Willingness. Does the connector reply to the person asking? Have they made introductions before? Does a favour run either way?
- Freshness. Has the connector spoken to the target this year?
- Length. One or two hops for a real ask. Three hops is research.
- A fallback. If the best connector is away, who comes next?
Fred Wilson set the etiquette in 2009: ask both people before making an introduction. He makes half a dozen or more a day, and when people skip the opt-in he ends up taking meetings he doesn't want.
So pathfinding should end in the ask itself: who to approach, why them, the evidence, and a draft note asking the connector whether they would be comfortable making the introduction. A mock example:
Target: Maria Okafor, CFO, Northfield Logistics (mock data)
Best route: Priya (our team) → James Hart → Maria
- Priya and James: 14 meetings since 2023, the last one three weeks ago. Both reply.
- James and Maria: four years in the same 60-person finance team. James commented on Maria's post last month.
Fallback: Tom (our team) → Maria. One email thread in 2024, with no reply from Maria. Weaker.
Draft note to James: "James, we're looking at Northfield and I'd value 20 minutes with Maria on their freight data. Would you be comfortable asking whether she's open to a call? I can send a short blurb you can forward."
Blind Spots
What the logs miss
Scores built from messages have a known weakness. Jason Wiese and colleagues at Carnegie Mellon titled their 2015 paper "Call and SMS Logs Do Not Always Indicate Tie Strength". Some close relationships leave little trace in any one channel. Partners build relationships over dinner, on the phone and in WhatsApp groups that never touch the company's email.
Three things help:
- Combine channels. A relationship that lives in WhatsApp and meetings still shows up.
- Show the evidence. Every score lists what it rests on, so people can judge it.
- Let people correct it. One click says "I know her better than this", and the next answer reflects it.
Firms that can only read metadata can still do most of this. One global engineering consultancy we have spoken to can't read email content, for legal reasons. It can use who wrote to whom and when, and the job titles in signatures. That covers recency, intensity, balance and duration.
Privacy
Private by default
Relationship data is personal, and people share it when they trust what happens to it.
Each person's relationships stay private to them. The firm can't read them. People then share in layers: with a named colleague, with their co-founders, or with the whole firm. What they share is the existence and strength of a relationship. A colleague can see that Sarah knows the target well. They never see Sarah's emails.
Before a route reveals a private relationship, its owner approves it. That is the double opt-in applied inside the firm, and nobody receives an intro request they didn't agree to. Agents follow the same rule. They can draft an introduction, and a person decides whether to send it.
UK guidance points the same way. The Information Commissioner's Office tells employers that monitor staff to use the least intrusive means, to complete a data protection impact assessment, and to avoid "function creep". Relationship scores exist to help people find each other. They never measure individual performance.
We cover how permissions follow each person through the brain in A company brain that knows who is asking.
Agents
Agents need the graph too
Ask a general assistant who to invite to a dinner for founders in your sector, and it has no idea who you know. Ask it who at your firm knows a target's CFO, and it can search your inbox for the name. It can't rank your colleagues, find a route through someone else, or tell a close relationship from one cold email.
With a company brain, the agents your team uses can answer those questions too.
- An agent preparing a trip lists the people worth seeing in that city.
- An agent screening a target attaches the firm's routes to its founders and managers.
- An agent planning an event builds the guest list from the gaps.
In each case the relationship evidence comes with the answer, and a person decides what to send. For a fund, this sits next to the sourcing work we described in Why private equity firms treat knowledge graphs as infrastructure.
Getting Started
Start with one question
Pick one of the four questions. Most firms start with lookup or pathfinding against a live target list. Within two weeks, one team can ask it and get ranked answers drawn from the firm's history, with personal sources private to their owner and sharing rules the team has agreed.
You choose where it runs:
- 60x Cloud, hosted by us with EU or US data residency
- Your Cloud, inside your own Azure, AWS or GCP cloud environment
Both can run on open-weight models.
Measure outcomes: meetings that came through a warm route, deals sourced from existing relationships, expert calls replaced by people the firm already knew. Counting introductions alone gives people a reason to make more of them.
Book a demo and bring a target list. We'll show you who your firm already knows.
Sources
Sources
- Paul Gompers, William Gornall, Steven Kaplan and Ilya Strebulaev, How Do Venture Capitalists Make Decisions?, NBER Working Paper 22587, 2016.
- Mark Granovetter, "The Strength of Weak Ties", American Journal of Sociology 78(6), 1973, as quoted in Gilbert and Karahalios.
- Eric Gilbert and Karrie Karahalios, Predicting Tie Strength With Social Media, CHI 2009.
- Daniel Levin, Jorge Walter and J. Keith Murnighan, "Dormant Ties: The Value of Reconnecting", Organization Science 22(4), 2011.
- Adam Kleinbaum, "Reorganization and Tie Decay Choices", Management Science 64(5), 2018.
- Karthik Rajkumar, Guillaume Saint-Jacques, Iavor Bojinov, Erik Brynjolfsson and Sinan Aral, "A causal test of the strength of weak ties", Science, 2022. Summary: MIT News.
- Ronald Burt, "Structural Holes and Good Ideas", American Journal of Sociology 110(2), 2004.
- Peter Dodds, Roby Muhamad and Duncan Watts, An experimental study of search in global social networks, Science 301, 2003.
- Fred Wilson, The Double Opt-In Introduction, AVC, 2009.
- Jason Wiese, Jun-Ki Min, Jason Hong and John Zimmerman, "You Never Call, You Never Write": Call and SMS Logs Do Not Always Indicate Tie Strength, CSCW 2015.
- Information Commissioner's Office, Employment practices and data protection: monitoring workers, October 2023.
Client and contact examples are anonymised. The route brief uses mock data. 60x installs company brains for investment firms and professional services teams, in your cloud or ours.