The Portfolio Departure Watch: Catching a Key-Person Exit Early

You need to know when a founder or key executive leaves one of your portfolio companies before it surfaces in a board deck or the press, so you can intervene while it still matters. This playbook is for early-stage investors, platform and talent partners, and angels who sit between board meetings and want a concrete, named-seat watch they can build themselves. It gives you the seats to track, the public signals that move first, a weekly procedure, and the triage rule that separates a routine single exit from a departure cluster that warrants a same-week call.

Published investing guides tend to score distress after the fact or size a market. Almost none give the operational procedure for detecting a key-person exit from public signals in the days between board meetings. This is that procedure. It is a watch a fund runs on its own, not a dashboard the founder has to feed.

Why the board gap is the whole problem

The signal window is defined by the board calendar, not by the news cycle. Early-stage boards meet roughly every six to eight weeks, which works out to six or seven meetings a year, so a departure you do not catch publicly can stay invisible to you for up to two months.

That gap is the entire reason this playbook beats a quarterly monitoring dashboard. If your only view of a key-person exit is the next deck, you learn about it after the runway conversation has already happened inside the company. A weekly public-signal sweep collapses the two-month blind spot to days. You want to be the investor who calls the founder about the CTO leaving, not the one who reads about it in slide fourteen.

Board cadence itself scales with stage. Optimal frequency runs monthly at seed to quarterly by Series B and beyond. A monthly cadence is widely considered too fast for real operational change, which is useful to you for a different reason: it means seed founders and boards are looking at the numbers more often but the company is also moving faster, so you tune your sweep up toward daily for the earliest companies and keep it weekly for the rest.

The stakes are not abstract. In one dataset of more than 150 co-founder departures, roughly one in four became a significant financial event: 18 percent ended in litigation or settlement and 28 percent cost more than $50,000. Board-driven founder-CEO replacement in the early-scale years is estimated at 10 to 20 percent. Catching the signal early does not prevent these outcomes, but it buys you the time to shape them instead of reacting to them.

Which seats to name, and why finance and technical go first

Watch four seats per company: CEO or founder, CTO, CFO, and COO. These are the officers practitioners consistently flag as highest-signal, because when higher-ups resign in quick succession for personal reasons or exciting new opportunities, they usually have the clearest view of the company's real situation, especially the financials.

That mechanism matters. A CFO or CTO leaving ahead of the CEO's public brave face is information asymmetry made visible. The people with the numbers and the people with the codebase see the trajectory first, so their exit is a leading indicator of what the CEO will admit two board meetings later. When the finance or technical seat goes first, weight it more heavily than the same title moving at a healthy company.

Stage-specific named-seat counts and thresholds are not publicly established, so treat these four as the load-bearing minimum. Add a functional head - a VP Sales, a Head of Product - only where a specific company is unusually dependent on one person. The point of naming individuals rather than titles is that titles do not have LinkedIn URLs or GitHub handles. You cannot diff a role. You can diff a person.

Coverage scales unevenly by geography, and you should plan for it. In Refolk's index, the United States basket of CTO, VP Engineering, and Head of Engineering profiles holds 37,468 records, while the United Kingdom holds 7,183 profiles titled Chief Technology Officer. That is a denser public matching surface for a US-heavy portfolio. A non-US portfolio has thinner profile coverage, so you weight the team-page and GitHub sources more heavily there to avoid blind spots.

The public signals, ranked by speed and reliability

Four sources carry departure signal, and they differ sharply in how fast they move, whether they are dated and hard, and who they cover. Rank them before you wire anything, because the fastest source for a private company is not the same as the fastest source for a public one.

The SEC Form 8-K is the fastest hard signal, but only for public companies. An officer or director departure triggers an Item 5.02 filing within four business days of the event, and the named-executive-officer definition covers the CEO, CFO, and the three most highly compensated other executives. If the event falls on a weekend or holiday, the clock starts the next business day. For the pre-seed-to-Series-B portfolio this watch targets, that filing does not exist, which is precisely why a private-company method has no published competitor.

For private companies, the standout early signal is GitHub contribution decay. Because org access is cut at departure, a contributor's commits to a private organization can vanish from their profile once they are removed. One documented case had a contributor working on private repos from February to June, then losing all of those contributions from their profile after leaving and being removed from the organization. That decay can precede a LinkedIn edit by weeks, since most professionals wait one to four weeks after starting a new job to update LinkedIn and some never do.

The lesson of the funnel is that public disclosure only helps you once it is too late to be early. Wire each seat to at least two of these sources so a miss on one is caught by another.

Building and diffing those baselines by hand across twenty or forty companies is the friction. Refolk turns each seat into a plain-English standing query - the current officers at a company, the people who changed their title away from it in the last sixty days, the contributors who have gone quiet - so the weekly sweep is a set of saved asks across GitHub, LinkedIn, and the open web rather than a spreadsheet you maintain by hand.

The procedure: from watchlist to intervention call

The end-to-end method is seven steps: name the seats, wire each to its fastest source, sweep weekly, de-duplicate, classify, triage the cluster, and call. Run it start to finish and a distress cluster reaches you days before it reaches a board deck.

The two steps that carry the most weight are de-duplication and classification. Skip de-duplication and one person across three sources looks like three exits, and you escalate on a phantom cluster. Skip classification and you treat a planned, well-managed transition as a crisis, which burns the founder trust the whole watch is meant to protect.

Routine or distress: the triage rule

The distress signal is the cluster, not the single exit. A single explained exit with a named successor is common and survivable; escalation value comes from counting exits in quick succession against the turnover line, not from reacting to any one name.

Use two dimensions to make the call: whether the exit is planned, and how many are happening at once. A planned exit shows a named successor and transition language on the same day, sometimes with the officer staying on to assist. A distress exit shows no successor, a vague reason, a finance or technical seat going first, and clustering.

The bottom-right quadrant is the one that earns the same-week call. Two or more exec exits in quick succession with no successor, especially when the finance or technical seat moves first, is the pattern that precedes the runway conversation. The WARN Act sits alongside this as a company-level overlay: the federal act requires employers with 100 or more full-time staff to give 60 days' notice of a mass layoff, New York's mini-WARN uses a longer 90-day notice and a lower 50-employee trigger, and state trackers publish company, location, and effective date weekly. Use WARN only where a portfolio company is large enough to file, never as a signal about a single founder.

How this goes wrong: failure modes and false positives

Every source in this watch lies in a specific way, and the value of the standard is knowing how. Below are the seven failure modes to guard against, each with the check that catches it.

• LinkedIn-only watch misses the exit. Some execs never update their profile, so single-source monitoring produces false negatives and you learn from the board deck. Check: pair every seat with a second source, either the team page or GitHub.
• GitHub decay false positive. A commit drop can mean vacation, parental leave, or a shift into management, not departure. Check: confirm removal from the org or team page before escalating.
• 8-K reliance on private companies. Early-stage portfolio companies file no 8-K, so treating no filing as no departure is a false negative. Check: never apply the public-company path to a private company.
• Team-page lag. Marketing sites update slowly, so a stale page reads as still employed. Check: timestamp and diff the page rather than trusting its current state.
• Cluster mis-count from duplicates. The same person across three sources looks like three exits and triggers a false distress cluster. Check: de-duplicate to one person-event before counting.
• Planned exit misread as distress. A named successor plus transition language is routine, and escalating it burns founder trust. Check: apply the routine-versus-distress rule before booking the call.
• WARN as an individual-exec signal. WARN covers mass layoffs at firms with 100 or more employees, not single founder exits, and most seed companies fall below the threshold. Check: use WARN only as a company-level distress overlay.

The through-line is that no single source is both fast and reliable. The 8-K is hard but absent for private companies. GitHub is fast but noisy. LinkedIn is broad but voluntary and lagged. Team pages are direct but stale. The watch works because you cross them, not because any one of them is trustworthy alone.

Verify before you call it done

Before you treat the watch as live, confirm each of the following. A gap in any one produces the false negative or false positive that the whole procedure exists to prevent.

Keeping the watch current

A departure watch decays the moment a company raises, reorganises, or hires the seats you named. Re-baseline on a fixed cadence and on every trigger event, or the diff quietly stops meaning anything.

Re-run the named-seat build whenever a portfolio company closes a round, since a new raise usually adds a CFO or a COO you were not tracking and sometimes moves a founder out of the CEO chair. Refresh the LinkedIn and GitHub handles quarterly, because people change handles and companies rename their GitHub org. When you add a company to the portfolio, it enters step one of the procedure the same week, not at the next portfolio review.

Two facts should keep you honest about scope. First, a precise published median lag from actual last day to public signal for founders and executives is not established, so treat the source-speed table as a ranking of relative speed, not a guarantee of days. Second, a workable published false-positive rate for exec-departure monitoring is not established either, which is why the de-duplication and classification steps are not optional polish - they are the parts of the method that keep your false-positive rate low enough that partners trust the escalation flag.

The measure of a working watch is a single outcome: the next time a key person leaves a portfolio company under distress, you booked the founder call before that departure reached a board deck or the press. Everything in this playbook exists to move that call earlier by weeks.