The Co-Founder Breakup-Risk Read From Public History

Before you wire a seed check, you grade the founders. Almost every published playbook grades them one at a time: this person's track record, that person's domain depth. This guide grades the thing those playbooks skip - the durability of the relationship between two named co-founders - using only what you can reconstruct from public history. It is for early-stage investors, platform and talent partners, and angels who want a relationship risk read, not another individual-quality checklist.

The output is a score across named dimensions and a defensible verdict: back it, dig deeper, or pass. The spine is public: shared prior employer, relationship tenure, role-lane clarity, and serial co-founder churn are all reconstructable from LinkedIn, Crunchbase, and filings. The parts that are not public - equity split, vesting, board control - I will tell you plainly to leave as unknown rather than invent.

Why the pairing deserves its own diligence pass

Co-founder conflict is a top killer of otherwise fundable companies, and the pairing itself carries risk that individual-quality diligence never surfaces. The most-cited figure is Wasserman's: 65% of high-potential startups fail as a result of conflict among co-founders, from his study of nearly 10,000 founders and 3,600 startups. Treat that headline with care - it traces back to a 1989 Gorman and Sahlman survey of 96 at-risk companies, of which 61 ranked team issues among their top three failure reasons, which is 63.5% rounded to 65%. It is a rounded, decades-old derivation, not a live measurement.

The live measurement comes from Carta. Across 22,352 founders in 2-3 person teams who raised VC between Q1 2016 and Q2 2023, roughly 23% of co-founders had left by the 3-year mark. Steve Blank puts the pre-funding attrition even higher, estimating that one third to one half of startups melt down over team dynamics before they get funded.

Two things make this a distinct diligence job. First, the risk is back-loaded: departure at three years rose from roughly 19-20% in the 2018-2019 cohorts to past 23% in the 2021 cohorts, and it keeps climbing after. A seed-stage pair that looks harmonious is inside the low-risk window, not past the danger zone. Second, Wasserman found that founding-team turnover increases dramatically when the startup raises its first round - the very event you are about to trigger. Your check does not just observe the risk; it advances the clock on it.

The dimensions that actually move the read

Five public dimensions carry the score, and their weights are counterintuitive. Prior operating experience matters far more than prior relationship. Complementarity matters more than comfort. Tenure is real but soft. Here is how each dimension behaves and what it looks like when it lies to you.

The evidence that reorders these weights is the surprising part. In the Outcast Ventures Billion-Dollar Founder Study of just over 344 U.S. venture-backed companies exiting at $1B or more, founders who had worked together before had a median exit valuation of $2.3B, lower than the $2.9B median for those who had not - a 26% gap. School ties repeat the pattern: founders who went to school together showed a median of $2.5B versus $2.7B for those who did not. Familiarity, in the data, is a mild negative for scale.

What did predict scale was operating history, not togetherness. Founders with a previous exit achieved 91% higher valuations, and founders who had previously worked at a startup produced 41% higher exit valuations than first-timers. So when the dimensions conflict, prior-exit experience wins.

Score operating history, not togetherness. A previous exit swings valuations far harder than any tenure signal.

There is a genuine tension in the source material, and an honest framework holds both sides. Wasserman's earlier work treats prior joint work as a green flag, arguing the most successful teams tend to be people who worked together in the past, while married couples, family members and friends failed most often because they avoided tough conversations. Outcast's data treats familiarity as a yellow flag. My resolution: familiarity is neutral-to-slightly-negative for outcome size, but the mechanism Wasserman identified - conflict avoidance among the close - is real, which is why serial churn and adversity history do the heavy lifting, not tenure.

What is public and what is not

Draw a hard line between what public profiles support and what only private documents can. Crossing it is the fastest way to write an indefensible memo.

Publicly reconstructable, with dates:

• Shared prior employer, confirmed by overlapping employment dates
• Shared school, confirmed by overlapping years
• Prior co-founding, from company histories
• Length of relationship before founding, computed from timelines
• Role-lane signals from titles, such as Co-Founder & CEO versus Co-Founder & CTO

Largely private and only inferable:

• Exact equity split
• Vesting schedule
• Board and voting control

The equity picture is a common trap. Wasserman found 73% of founding teams split equity within a month of founding, often equally and irrevocably, but that is an aggregate. The exact split for a specific pre-seed pair is not established publicly - it lives in cap tables and founder agreements. Skill complementarity is partly inferable from titles: complementary skills are best practice, which is why you often see a non-technical and technical pair, each in their appropriate swim lanes.

The procedure

Run these seven steps in order. The first five are desk work an analyst can complete in about two hours; the sixth is partner-level and takes hours to days. The method basis is documented: the Outcast team assembled most founder-level data manually by reviewing LinkedIn profiles, podcast interviews, SEC filings, and historical articles to reconstruct how founding teams formed.

For step 3, keep a real benchmark in mind: Mercury's co-founders worked together for six years at Heyzap before founding Mercury. That is deep, documented, dated overlap - the kind you can defend, and the kind most pairs do not have.

The reconstruction in steps 1, 2, and 6 is where a plain-English search saves the most time. Instead of manually crawling two profiles and guessing who worked where, I let you describe the pattern you are testing for and return the people who match it across public LinkedIn, GitHub, and the open web.

That is the one place I fit naturally into this workflow. Use Refolk to surface the pairs and the overlap points; use your own judgement on the dimensions.

The step-6 read investors run in the room

The most decisive evidence is not on a profile at all, which is why backchannel is a step and not a footnote. Before a VC wires a dollar they will talk to 15 to 20 people - former managers, ex-colleagues, early customers, and founders who have worked with or competed against the pair. Firms like Sequoia, Andreessen Horowitz, and Benchmark run 10 to 20-plus reference calls per deal, targeting people the founder names and people the founder does not name.

CRV documents the cross-referencing technique that makes off-list calls possible: map the portfolio on a website or Crunchbase, cross-reference founders on LinkedIn to identify mutual connections, and trace career history to surface people from prior firms. That is how you reach references outside the curated list.

There is one read you simply cannot reconstruct from public data, so name it as a gap and close it in person. In-room, investors watch whether each founder has a clear lane, whether the person closest to a topic answers it, and whether teammates sharpen each other. Public titles give you a lane hypothesis; the room confirms it.

Hi [name] - I understand you overlapped with [Founder A] and [Founder B] at [prior employer] around [years]. I am doing diligence on the company they started together and would value 15 minutes on how they worked as a pair back then. Three things I care about: who took which decisions, how they handled a disagreement, and whether either has a habit of moving on from partners. Happy to keep it entirely off the record.

How this read goes wrong

This is the section to reread before you write the memo. Every dimension has a false positive, and most bad co-founder calls come from one of these eight.

• Same-logo illusion. Two founders list the same employer but never overlapped. It reads as "they worked together." Check start and end dates, not company names.
• Familiarity read as strength. Treating shared history as a green flag inflates the score, yet worked-together teams exited 26% lower. Weight complementarity and prior-exit experience above tenure.
• Tenure without adversity. A long, easy friendship tells you little; failed projects can be a better signal than an easy success because you have seen how they handle adversity. Look for a documented hard pivot or failure they went through together.
• Lane collision missed. Two founders in the same functional lane read as a strong team but lack coverage. Confirm distinct titles and skills; flag duplicate lanes.
• Equity inferred as public fact. Stating a pair's split or vesting from public data is fabrication - it is not established publicly. Label equity, vesting, and control as unknown unless it appears in an SEC filing.
• Serial-founder halo. A repeat founder looks safe, but repeated co-founder departures are the churn signal. Count how many prior co-founders left, not how many companies were started.
• Backchannel curation. Founder-supplied references are curated; the references the founder never knew about are where deals die quietly. Source at least one contact the pair did not name.
• Stale-stat anchoring. The 65% headline is a rounded 1989-derived figure, not a live measurement. Pair it with the dated Carta and Outcast numbers rather than citing it alone.

Keep the relationship read in proportion to the other failure reasons. Co-founder conflict is real, but CB Insights' analysis of 431 VC-backed shutdowns since 2023 found capital and product-market fit dominate the visible causes.

Team dynamics often sit upstream of these - a fractured pair burns capital and misses PMF - but do not let a clean relationship read paper over a broken business.

Turning the scores into a verdict

Map the dimension scores to one of three actions, and write a one-line rationale for each dimension so the verdict survives a partner meeting. The matrix below is the fast version: it plots the two dimensions that carry the most weight against each other.

Read the verdicts this way:

• Back it when lanes are distinct, at least one founder has a prior exit or startup operating history, and backchannel raises no serious churn or trust flag. Familiarity is fine here; it is not carrying the score.
• Dig deeper when the shape is promising but one high-weight dimension is soft - two strong operators in the same lane, or a complementary first-time pair. Resolve it with off-list calls and the in-room lane test.
• Pass when founders are first-timers in the same lane, when a repeat founder shows a pattern of co-founders departing, or when the shared history is thin and the backchannel is worse.

The funnel below is how a batch of pairs typically narrows across the procedure. The volumes are illustrative of shape, not measured counts.

Before you commit the verdict to a memo, run the checklist.

Keeping the read current

The relationship read is not a one-time gate; it is a signal you re-check as the company moves through its first financing and beyond. The mechanism to watch is fixed even as the numbers drift: founding-team turnover increases dramatically after the first round, and departure risk keeps climbing well past the three-year mark. Re-run steps 4 through 6 after the round closes and again around the two-to-three-year window, when the Carta data says attrition accelerates.

One honesty note on sourcing. I attempted to pull an exclusive aggregate from Refolk's index for co-founder pairs across four filter variants, and each returned zero matching records, so there is no proprietary figure in this guide - the quantitative spine is Carta, Outcast, and CB Insights, cited plainly. Where a number here is dated or derived, I have said so. When you re-check a pair, re-verify the public timeline first, then close the private gaps in the room, and let complementarity and operating history outrank comfort every time.