The Portfolio Conflict Read: Clear, Adjacent, or Direct Competitor

This is the screen to run before partner time goes into a new deal. Does the deal collide with a company the fund already backs? If it does, is the answer go ahead, go ahead with disclosure and a firewall, or pass? It is written for associates, partners, platform and talent partners, and angels who want a VC portfolio conflict check that two people would score the same way. The guide gives five scored dimensions, verdict rules, an eight-step procedure, the ways this check goes wrong, and templates for filing the evidence.

What does the portfolio conflict read decide?

It gives each pairing of a prospect and an existing holding one of three verdicts: Clear, Adjacent, or Direct. Each verdict comes with a fixed next action. The unit is the pair, not the deal. A prospect with three neighbours in the portfolio gets three separate reads.

A few terms come up repeatedly:

• Investors' rights agreement (IRA): the contract that grants investors information rights and often defines who counts as a competitor.
• Information rights: the contractual right to receive a company's financials and other non-public information.
• Firewall: the assignments and restrictions that stop one company's information reaching the other through the fund.
• Trajectory: the direction a company is moving, read from what it hires for and what it newly ships.

The read exists because the usual alternative, a partner's gut feel plus a quick call to the portfolio founder, fails in both directions. The page that currently ranks for this topic is written from the founder's side. It says the check "still kills deals, and it still leaks information." A scored read that relies on public signals cuts down both problems. It kills fewer deals that are only adjacent, and it shares less while you work out which kind of deal you have.

What do the documented conflicts show?

Only three documented outcomes are in the public record, and each one teaches a different lesson. Finix shows that buyer perception outranks product taxonomy. Hinge shows that failing to disclose costs more than the overlap does. Anthropic shows that firm policy, not convention, sets the bar.

Finix: the conflict experts missed

Finix announced a $35 million Series B led by Sequoia. Shortly afterwards, Sequoia told Finix it had a conflict because of its earlier investment in Stripe. Diligence had relied on experienced payments professionals. It had not included the ordinary merchants and software companies that saw Finix and Stripe as rivals. Sequoia cited "a variety of small data points that collectively painted a different picture of the market."

Sequoia gave back its equity, board seat, and information rights, and Finix used the returned equity to raise another $10M. It was reported as the first time in the firm's history that it had cut ties with a newly funded company over a conflict of interest. Stripe had not asked Sequoia to walk away. Finix later became a payments facilitator, a company that processes payments on behalf of the businesses it signs up, which put it in direct competition with Stripe. Experts sort companies by their architecture. Buyers sort them by the problem being solved.

Hinge: the conflict nobody was told about

A Bessemer partner was recused from the Hinge board for not telling CEO Perez about a new investment the founder saw as competitive: New York-based Clearing. Clearing's CEO disputed that the companies competed. The partner lost the seat anyway, and Bessemer's Elliott Robinson moved from board observer to director. The overlap was arguable. The silence was not.

Anthropic: policy overrides the norm

Sequoia moved to back Anthropic while already holding OpenAI and xAI. Altman testified that investors with confidential OpenAI information could lose access if they invested in competitors. Information exposure, not market overlap, was the cost under discussion.

The evidence here is thin. Three cases are not a base rate, and Clearing's stage at the time is not established in the public sources. Use these cases to set the shape of the model, not its numbers.

The five dimensions and how to score each

Score every pair on buyer overlap, product overlap, shared accounts, trajectory, and information exposure, each from 0 to 2. Attach one piece of saved evidence to each score. Any dimension without evidence is scored U (unknown), not 0.

These bands are this guide's house model, not a published standard. No public body has published one. What makes them useful is that they are written down and applied the same way every time.

Buyer overlap

This measures whether the same person, in the same kind of organisation, holds the budget for both products. It proves the companies are competing for the same money. It lies when you match on title alone. The same title can sit in a council, a bank, and a staffing firm, each with a different buying motion. Score 2 only when title, employer type, and employer size all match.

Product overlap

This measures whether a buyer would replace one product with the other. Score it from the buyer's point of view, not the vendor's. It lies when you score it from feature lists, pricing models, or the category a company gives itself. Finix's per-transaction pricing differed from Stripe's, and buyers still saw them as rivals.

Shared accounts

This measures whether named customers use both companies for the same job. It proves the companies are already meeting in the market. It lies when a shared logo hides different departments or different use cases. A big employer can buy two products in the same category for unrelated teams.

Trajectory

This measures movement toward the other company's core business. It is read from new product pages and from the roles each company is hiring for. It proves that today's adjacency may become tomorrow's collision. It lies in both directions. Exploratory job posts can look like a strategy, and a real pivot can arrive with no visible warning. No published study shows which outside-in signals come before a pivot.

Information exposure

This measures what the fund sees at each company and who sees it. It does not change whether the companies compete. It decides whether an Adjacent verdict can be managed. Baker Botts partner Michael Torosian put it plainly: "The real risk is information leakage."

How do you turn five scores into a verdict?

Apply the rules below in order and stop at the first one that matches. Buyer overlap sets the ceiling, product and accounts decide between Adjacent and Direct, trajectory can promote a pair, and information exposure decides whether an Adjacent verdict can be managed.

1. Direct if buyer overlap is 2 and either product overlap or shared accounts is 2.
2. Direct if trajectory is 2 and buyer overlap is 1 or higher.
3. Clear if buyer overlap, shared accounts, and trajectory are all 0, product overlap is 1 or lower, and the buyer score rests on end-buyer evidence.
4. Adjacent for everything else.

Two overrides apply after that:

• Exposure override: an Adjacent pair with information exposure at 2 is treated as Direct for action until a documented firewall brings exposure down to 1.
• Unknown override: a U on buyer overlap can never produce a Clear verdict. At best it produces a provisional Adjacent, pending buyer validation.

For two partners to grade the same pair the same way, they need the same evidence and the same rules. Have two people score independently from the saved evidence, then compare. Settle any disagreement by adding evidence, not by debate. If they still disagree on buyer overlap after that, the higher score stands.

Prospect:
Portfolio company:            Vehicle:
Scored by:                    Second scorer:              Date:

Dimension             Score (0/1/2/U)   Evidence (link or file name)
Buyer overlap         _                 _
Product overlap       _                 _
Shared accounts       _                 _
Trajectory            _                 _
Information exposure  _                 _

End buyers consulted (role, employer type, what they said):
1.
2.
3.

Rule matched (1-4):           Overrides applied:
Verdict: Clear / Adjacent / Direct
Firm policy reference:
IRA obligations (from counsel):
Founder notified (date, by whom):
Firewall memo (link):
Next re-score trigger:

How do you read buyer overlap from public signals?

Treat each public signal as a lead to check, not a verdict. The weakest evidence is how the vendor describes itself. The strongest is how the buyer describes the two companies. Rank what you collect by how close it comes to the buyer's own words.

No published false-positive rates exist for logo pages, case studies, reviewer titles on review sites, buyer titles in job posts, or integration directories. Nobody can tell you how often each one wrongly suggests overlap. The only documented case, Finix, is a false negative: product-level differences hid buyer-level overlap. So treat the outer layers as a way to find candidates, never as a way to rule a pair out.

Why a title match is not a shared account

Here is a worked example of the second layer misleading. Suppose a prospect and a portfolio company both sell to benefits leaders. The title matches. In Refolk's index of professional profiles, though, the people holding that title fall into two very different pools.

The top UK employers for these titles include Lewisham, Tamworth, and Merton councils, Barclays, and Monzo. The US pool includes two PEOs (professional employer organisations, firms that run benefits and payroll for many smaller employers): Vensure and ProService Hawaii. A product sold to a US PEO's benefits director and one sold to a London council's head of benefits share a title. They do not share a buyer.

The UK count probably undercounts, because UK practice often uses "Reward" titles and those were not queried. That has not been checked. If your pair turns on UK buyers, query the Reward titles too before you score.

The hard part of this step is getting past the title to the employer type and size, and then finding real buyers to ask. Keyword search returns the title and the noise together. Refolk takes a plain-English description of the buyer, including employer type, size, and the programme they run, and returns named people. That gives you an interview list for step three instead of a headcount.

What do investor agreements say about competitors?

Usually less than people assume. Competitor clauses in filed investors' rights agreements are mostly written to let funds back competitors, so the contract rarely forces a pass. Your firm's own policy decides, and counsel still needs to find the obligations the contract does create.

Thresholds run from 2% to 20%, and some agreements exempt VC firms entirely. What contracts do reliably create is a disclosure duty and a lever on information. One filed agreement makes investors disclose any new investment they reasonably believe may be deemed competitive. Contract terms also let a company withhold information rights from investors who are competitors. Those two clauses are the ones counsel should look for in step five.

Industry guidance does not fill the gap. ILPA says all appropriate investment opportunities should go to the fund during its investment period, and that the allocation policy should be given to LPs on request. It says GPs should limit overlapping investments between their own funds and disclose them. It also says a GP should not invest in opportunities appropriate for the fund through other vehicles, except pro rata under pre-disclosed co-investment agreements. Review by the LPAC (the fund's limited partner advisory committee) covers cross-fund investments. That means two of the firm's own funds in the same company, not two competing companies. No ILPA text addresses competing portfolio companies directly. Hunter Walk argues that multi-stage firms are likely to rewrite the existing etiquette across vehicles.

The procedure, step by step

Run the eight steps in order. Steps one to four are internal and leak nothing. Steps five to seven create obligations and contact. Step eight keeps the verdict true after the money moves. No source documents this whole sequence end to end, and none gives timings, so set your own durations and record them.

Notes on the hard steps

Step three is where Finix went wrong. Experts told Sequoia the companies differed, and the merchants and software platforms that bought from both were never asked. Pick buyers from the segment where your title check found overlap, not from your network of operators.

Step four depends on hiring direction, and talent flow is one of the clearest trajectory signals there is. If engineers or sellers are moving from the portfolio company to the prospect, or the reverse, look at what roles they took. A query like "People who left Stripe for Finix in the last two years" is a single plain-English search in Refolk, and it turns a hunch about drift into a list you can read.

Pivoting is normal, so do not treat a stable product today as a reason to skip this step. Startup Genome, using data from more than 650 startups, found that startups that pivot once or twice raise 2.5x more money. No primary source gives a pivot rate by stage.

Step eight has no agreed position in the sequence. Sequoia found its evidence after it had invested, and sources disagree on whether re-checking belongs before or after the founder notice. I put it last as a standing obligation, and I also re-run step four immediately before step six whenever the gap between the first read and the term sheet is long.

How does the conflict read go wrong?

It goes wrong in seven recurring ways. Most of them produce a false Clear, which is the costly direction: you close, then find the conflict, then unwind it in public. Each one below comes with the check that catches it.

The overlap was arguable. The silence was not. Hinge cost a board seat over a conflict its own CEO disputed.

1. Product overlap scored on the vendor's taxonomy. This produces a false Clear. Finix priced per transaction differently from Stripe, and buyers still saw rivals. Check: score product overlap only from what end customers say.
2. A matching buyer title read as shared accounts. Same title, different segment: UK councils and banks against US PEOs and retailers. Check: compare employer type and size, never the title alone.
3. Keyword title matching. "Director of Client Success" came back from a benefits query, one of 25 in the US sample. Check: dedupe and read every title before you count or score.
4. Trajectory ignored. Finix became a payments facilitator and moved into Stripe's core business after the split. Check: re-score after each round of either company.
5. No notice to the portfolio founder. Hinge removed the partner from its board even though Clearing's CEO disputed the overlap. Check: written notice before the term sheet for any pair scored 1 or higher.
6. Vehicle separation taken as cover. ILPA rules govern allocation, not competition. Check: apply firm policy and the IRA competitor clauses to every vehicle.
7. The conflict call leaks positioning. A founder told about a prospect's pricing or traction now has competitive intelligence they did not earn. Check: share the category and the reason for the call, never pricing. No published standard covers this, so treat the rule as a precaution, not a proven norm.

Two more patterns are worth naming because the scorecard can hide them:

• Unknowns recorded as zeros. A blank buyer score that turns into 0 by default is how a hidden Direct gets filed as Clear. Keep U as U.
• The verdict outliving the evidence. A Clear that was right at seed can be wrong by Series A. In published cap-table cohort data, 30.6% of an earlier seed cohort converted to Series A within 24 months, against 15.4% of a more recent one. The median gap between seed and Series A is about 18 months. Many seed holdings sit for a long time with no round to prompt a re-score, so put seed pairs on a calendar review as well.

How do you notify, firewall, and keep the verdict current?

Notify in writing before the term sheet, firewall by assigning people rather than relying on promises, and re-score whenever either company raises. No published protocol says what a portfolio founder should be told, or what makes an Adjacent verdict defensible to both founders. The practice below is assembled from what the documented cases show.

The notice

The call comes from the partner who holds the portfolio company's relationship. The written follow-up is the record. Hinge shows what skipping it costs, and the filed disclosure clauses show that some IRAs require it anyway.

Subject: Heads-up on a company we are looking at

Hi [first name],

Following our call today: we are in early diligence on a company in [category]. It sells to [buyer function] at [employer type]. We scored it as [Adjacent / Direct] relative to [portfolio company] on our conflict read, mainly because of [one-line reason, e.g. shared buyer function, different segment].

What this means for you:
- [Partner name] stays on your board. If we proceed, [other partner name] would represent us at the other company.
- Nothing about your company's plans, pricing, or numbers will go to the other team, and the reverse also holds.
- If board discussion turns to the other company's market, [partner name] will recuse.

If you see the overlap differently, I want to hear it before we go further. Your view is part of the read.

[Partner name]

The firewall

Documented practice is to have different partners represent the fund on the two boards, and to have them recuse when board discussion turns competitive. Information rights are the second lever. Where the IRA lets the company withhold rights from competitor investors, decide in advance whether the fund will waive them. Fenwick's guidance for directors appointed by corporate venture arms covers the same mechanics and is worth reading for the recusal language.

Write the firewall down. It should list who sits on which board, who receives which information package, who recuses on which topics, and who checks that this is happening. An unwritten firewall is a promise, and promises fail at the worst possible moment.

Keeping it current

Every Adjacent or Direct pair gets a re-score trigger written on its scorecard. Re-run the read when either company raises, when either launches a new product page, when either posts roles that sell into the other's buyer, and at least on a calendar cycle for seed holdings. When a verdict moves up, from Clear to Adjacent or from Adjacent to Direct, treat it as a new conflict. Run steps five to seven again before anyone says anything at a board meeting.