MBTI Compatibility for Co-Founders: What Type Pairings Can and Cannot Predict

We could not find an MBTI compatibility chart validated against the outcome founders actually care about, and we looked. The grids that rank all 256 type pairs from "ideal match" to "conflict" publish no sample, no method and no measured outcome, and the Myers-Briggs framework was built to describe one person at a time rather than to score a pair. So a chart telling you an INTJ and an ENFP are a 92% match is not reporting a finding. It is reporting a convention.

That does not mean the question is worthless. It means the question is pointed at the wrong thing. The largest peer-reviewed study of founder personality and startup outcomes does not ask whether two founders are compatible. It asks which personality traits the founding team covers between them, and it finds that coverage is what moves the odds. That is a question you can answer honestly about yourself and a co-founder, and the rest of this page is how.

One thing to settle before the evidence: this page is published by a company that sells nothing but does run a founder personality test, and that test compares two people side by side. We have an interest in you taking it. So every claim below is attributed to a source you can open, the alternatives are named with what they actually do, and where our own instrument cannot support a claim we say so.

The strongest evidence in this area is McCarthy and colleagues, published in Scientific Reports in 2023, covering 25,214 founders at 21,187 startups founded from 1990 onward. Two findings from it are worth carrying around. First, firms with three or more founders were more than twice as likely to succeed as solo-founded startups. Second, ten combinations of founders with different personality types correlated with greater odds of success, and three specific trios were each more than twice as likely to succeed as other combinations.

The mechanism the authors propose is the part that matters here, and it is not chemistry. Their team-level analysis maps, for each startup, the highest value any co-founder reaches on each Big Five trait: the team’s coverage, not its average and not its similarity. Their own explanation is that it takes more than one founder to cover all the beneficial personality traits. Their conclusion is that diversity and complementarity of personalities matter in the foundation team.

Now the limits, because they are large and the press coverage of this paper dropped nearly all of them. Nobody in that study took a personality test. Personality was inferred from founders’ public Twitter posts by a commercial text classifier, which the authors acknowledge rests on a potentially biased presentation of a person’s digital identity, and they note that the language founders use online can itself be affected by events such as business failure. The underlying Crunchbase data over-represents technology firms, male founders and, importantly for the team-size finding, multi-founder companies. Companies that failed early are under-represented. The findings are correlations. Read it as the best available map, not as a law.

A number you will see attached to this paper everywhere, including in the marketing of at least one co-founder assessment, is that diverse founder teams are "eight to ten times" more likely to succeed. That figure is not in the paper. It comes from a quote in the press release. The paper’s own strongest team claim, stated twice, is "more than twice as likely". We use the paper’s number.

Two findings outside the startup literature point the same way, and together they are more useful than any compatibility grid. A meta-analysis of team composition by Bell found that in real workplaces, rather than in laboratory studies, the team measures that predicted performance were the team’s minimum agreeableness and the team’s average conscientiousness and openness. Minimum, not spread. That is a weakest-link result: one person low on agreeableness costs the team more than two people who differ from each other do. We read this from the published abstract rather than the full text, so treat it as a direction and not a coefficient.

The second comes from an experiment rather than an observation, which is rare here. Schoss and colleagues randomly composed 133 teams in a new-venture simulation and found that diversity in leadership orientation reduced conflict, for the plain reason that not all team members can lead at the same time. It is a simulation with student-scale stakes and its outcomes were team satisfaction and efficacy rather than company performance, so it cannot tell you a founding team will do better. It does give a mechanism for why two people who both want to be the one deciding will struggle, which is the only version of "incompatibility" with any evidence behind it.

So the question worth asking about you and a potential co-founder is not "are our types a match". It is three concrete things. Which of the five traits does neither of us cover? Where are we both at the low end, since that is the weakest link nobody will catch? And do we both want to be the one who decides? Those are answerable in an afternoon, and none of them needs a grid.

Here is how to turn two MBTI letters into something you can actually compare. The two frameworks do overlap, and the size of the overlap is the point. When McCrae and Costa measured the MBTI against a Big Five inventory in 468 adults in 1989, they concluded that its four indices measure aspects of four of the five major dimensions of normal personality. Four of five. The conventional correspondence is the one you would guess: extraversion with extraversion, intuition versus sensing with openness, thinking versus feeling with agreeableness, judging versus perceiving with conscientiousness.

The fifth dimension is emotional stability, and nothing in a four-letter type reaches it. That matters here more than it looks, because the 21,187-startup study found successful founders differed from its comparison group across all thirty trait facets it measured, and a cluster of those sit in the emotional range. So if you compare two founders by type alone, you are comparing them on four of the five traits the evidence points at, and you are blind on the one that governs how either of you behaves in the month it goes badly. Every compatibility chart built on four letters shares that blind spot, whatever its grid says.

So read your pair this way. Take the two types and mark, for each of the five traits, whether either of you is clearly at the high end. A trait neither of you covers is a gap to hire for or to build a process around, not a flaw. Then mark the traits where you are both at the low end, because that is the weakest link the Bell result points at. Then set the letters aside and talk about the fifth trait, the one types cannot see, by describing how each of you actually behaved the last time something you had built failed.

• Extraversion (E/I): who is in the room with customers and investors, and who runs out of road doing it.
• Openness (N/S): who generates the options, and who notices when the options stopped being grounded.
• Conscientiousness (J/P): who closes things, and who keeps the plan from hardening too early.
• Agreeableness (T/F): who will say the unpopular thing, and who will keep the team together after it is said.
• Emotional stability: not in your type at all. Ask about a specific bad month instead.

There is a reason advice about complementary co-founders so rarely survives contact with how teams actually form. Ruef, Aldrich and Carter studied 830 nascent entrepreneurs in a nationally representative US sample and found that team composition is driven by similarity, not differences. Net of romantic relationships, gender-homogeneous teams were about five times more likely than mixed ones. Strong social ties, family and close friends, shaped who joined; business acquaintances had no measurable effect.

Their phrase for what this costs is a "competency discount": founders appeared more concerned with trust and familiarity at the earliest stage than with functional competence. Note carefully what this study does and does not show. It measured how teams form, not how they perform. It is not evidence that similar teams do worse. It is evidence that the people you are most likely to found with are the people most like you, which is exactly the condition under which a gap goes unnoticed.

That is the practical case for comparing profiles at all. Not to grade the partnership, which no instrument can do, but to make the overlap visible early, while it is still cheap to design around.

If you went looking for a co-founder compatibility test, these are what you will find. We have listed ours in the same table on the same terms. The column that separates them is the last one: none of these tools, ours included, publishes validation data showing that its score predicts whether a founding team works.

Two of them deserve a specific note. The Oxford Science of Startups quiz is run by the same research group that published the 21,187-startup study above, which makes it the most credible instrument in the list, and it assesses one person rather than a pair. CofounderFit is the one built properly for two people, with an invite link, a joint report and a compatibility score, and it is honest enough to describe itself as a decision aid rather than a predictive instrument. If you want a pair score and nothing else, it is the direct answer to the query.

Almost every page about co-founder conflict opens with the claim that 65% of startups fail because of co-founder conflict, usually credited to Harvard Business School or to Noam Wasserman. We went looking for the primary source and could not establish one, so we no longer use the number, and we think the reason is worth writing down.

The figure traces to Wasserman’s 2012 book The Founder’s Dilemmas rather than to a paper with a stated method and sample. In Wasserman’s own framing the claim is about "people problems" broadly, which includes tension with early employees and investors, not co-founder conflict specifically, and it is a share of failures among high-potential startups rather than of all startups. Each retelling has narrowed it. At least one co-founder assessment now presents it as a Harvard Business School finding about co-founder conflict, which is two steps away from what the source supports.

The honest version of the claim is weaker and still useful: when a promising startup dies, the founding team is among the most commonly cited causes, and that is from a book-length study rather than a controlled one. If a page quotes you a precise percentage for how often co-founders sink companies, ask where it came from. We could not find out, and we looked.

• The number appears in a 2012 book, not in a peer-reviewed paper with a published method.
• Its original scope is "people problems", not co-founder conflict alone.
• Its denominator is failed high-potential startups, not all startups.
• It is widely re-attributed to Harvard Business School as a finding about co-founders. That attribution is not supported.

A profile comparison is an agenda, not a verdict. The value is that it puts specific, uncomfortable topics on the table before there is equity and a payroll riding on them, and it does that better than a conversation starting from nothing because it tells you which topics are yours.

Work through the gaps in order. Start with any trait neither of you covers and decide whether you are hiring for it, building a process for it, or accepting it. Then take the traits where you are both low, which is the weakest link, and name who is responsible for catching it. Then take any trait where you are both high, because two people equally determined to own the same thing is the conflict the simulation evidence actually predicts. Only then talk about the things no instrument measures: what each of you wants in five years, what you would each do if an acquisition offer arrived next month, and how you will decide when you disagree and neither of you is persuaded.

If one of you ends up with a lower score on something, that is not a finding about who is the better founder. Our own test scores free text with a language model and our sample is small; treat any individual number as a prompt for a conversation rather than a measurement. The pairs who get value from this are the ones who use it to design how they will work, and the ones who use it to settle an argument about who is right get nothing.