The Follow-On Conviction Score: Double Down, Hold, or Pass

When a portfolio company opens its next round, the pro-rata notice starts a short clock and forces one judgement: does this specific company earn more of my capital right now. This guide is for early-stage investors, platform and talent partners, and angels who have to answer that between the notice and the IC meeting. It gives you a scoring framework built from public trajectory signals you can pull in a day or two, so you land on double-down, take-pro-rata, or pass with a rationale a committee will accept.

Most public writing on follow-ons teaches fund-level reserve modeling and portfolio construction math. That work matters, but it never touches the actual per-company call. This guide assumes you already know your reserve pool and instead scores the company in front of you.

What the follow-on decision actually is

The follow-on decision is a per-company judgement layered on top of fund-level reserve math: given that I have some reserve capacity, does this company's public trajectory justify spending it here rather than elsewhere. Three outcomes exist. Double down means exercise pro-rata and buy any available over-allotment. Take pro-rata means hold your ownership and no more. Pass means let the allocation lapse.

The reserve math sets the outer bound. Funds typically reserve 40-60% of the fund for follow-on, and a common per-company rule of thumb is 2-3x the initial check, with USV historically reserving up to 2x. But reserve capacity is not conviction. A 2.4:1 follow-on-to-first-money ratio implies a roughly 71% reserves level that few funds actually hold, and only a small portion of any portfolio reaches an exit. So even a genuine winner can draw a rational pass when the capital is committed elsewhere. That is the core tension this framework manages: keep company quality and capital availability in separate columns.

Reserve capacity sets the ceiling. Public trajectory sets the score. Never let one masquerade as the other.

The base-rate prior: what a next round is worth by default

Start every follow-on from the current graduation base rate, because the prior has structurally halved and folklore will overstate your odds. The historical prior most sources cite is roughly one-in-three seed-to-A. That is no longer the operating reality for recent cohorts.

The decline continues in the most recent data. Of companies that raised a $1 million-plus seed in 2023, only 24% have progressed further; for the 2024 cohort the figure is 16%. The Series A-to-B transition tells the same story: 9% of the Q3 2022 Series A cohort reached B within two years, against 25% for the Q3 2018 cohort. Maturity recovers some of this - 2018-2020 A cohorts reach 40-55%+ by Year 4 - but a company one quarter past its A shows only 1-4% progression, so early reads are noisy.

Read the table this way: a company that would have looked "typical and fundable" at a 1-in-3 prior is now, mechanically, a 1-in-6 bet unless its own signals lift it above the cohort. The A bar rose and capital concentrated in fewer, larger deals. Top funds still graduate 50-75% of their seeds, so if your own realized base rate is that high, weight toward it - but prove it with your own portfolio, not a fund-reported headline that carries survivorship bias.

The dimensions that move the score

Score a company on five public dimensions, each of which either lifts it above the cohort prior or drags it below. For each, know what it proves and what it looks like when it lies.

Engineering telemetry. Commit velocity, measured as the rate of change in 14-day commit counts, is the earliest publicly available signal of startup momentum. When engineering acceleration doubles in a two-week window, it typically precedes a fundraise announcement by three to six weeks. This is the one signal that exists before the narrative does, and it fits inside the pro-rata window. It lies when repos are padded, so check commit authorship distribution rather than raw counts.

Hiring and headcount. An engineering hiring burst - a contributor count jump of 50% or more in a short window - usually means the company just closed a round and is scaling. That cuts against a pre-round read. If the burst follows a financing you missed, you may be doubling down on a company that already raised elsewhere. Ghost jobs inflate this signal, so check whether roles actually stay open past 60 days.

Departures. Departures cut the other way from hiring. Annual voluntary turnover above 15-20%, and key early employees leaving in clusters, is a sign the internal belief is gone. Founder loss is causal: startups are about 50% less likely to change or pivot after losing a founder, per a study of 33,000 startups. Cluster departures often precede the press, so this is where public sourcing earns its keep.

Round dynamics. Valuation direction matters less than insider behavior. Down rounds signal distress to the market, and the priced-round mix in Cooley's Q1 2026 data ran 86% up, 2.6% flat, 11.4% down. But the sharper read is on bridges. A bridge led by existing investors at a flat valuation reads as conviction; a bridge with no insiders reads as a flag. Bridges are now a base case, with 35-40% of seed-to-A companies raising one, median $750K, closing in four to six weeks. Carta's Q4 2023 data even showed a $48.1M median Series A bridge valuation, above the primary A median - so a bridge is not automatically weakness.

Traction against the bar. The Series B investor bar in current data is $3-10M ARR, net revenue retention above 110%, YoY growth above 80%, a repeatable go-to-market motion, and a clear path to $50M ARR within 24-30 months. Compare the company to that, not to its own last deck. Median private B2B SaaS NRR sits around 101%, so 110%+ is genuinely above market - if it is real.

kind: matrix
title: The double-down quadrant
caption: Trajectory strength and reserve capacity are independent axes; only the top-right earns extra capital.
x: Trajectory below cohort :: Trajectory above cohort
y: Reserves tight :: Reserves ample
quadrant: Hold or pass, document the business concern :: Take pro-rata, note you would go further with room
quadrant: Pass, this is a capital call you regret :: Double down, exercise plus over-allotment

The round-dynamics read, in detail

Insider participation is the single highest-signal round dynamic, so read who is in the round before you read the valuation. The mechanism is information leakage: new investors treat insider abstention as private information escaping, so an insider-led round signals that the people closest to the company are still betting. When you see a bridge, do not classify it as distress by reflex. Ask two questions: who is leading it, and is a specific milestone named. An insider-led bridge with a named milestone (a revenue target, a specific hire, a product ship) is conviction with a checkpoint. A bridge with no insiders and no milestone is the flag.

The same logic applies to your own decision. Because new investors read insider behavior, your choice not to exercise carries information you may not intend. That is why the memo has to state plainly whether a pass is about the business or about your reserves.

Reserve and pro-rata parameters you are working inside

Know the capital math cold before diligence, because it defines the ceiling on any double-down. Pro-rata amount is simply your ownership percent times the total size of the new round. Whether you can exceed it depends on side-letter or term-sheet over-allotment rights.

The exercise window is the constraint that turns this from an analysis into a race. CRV documents 20 days, other sources 15-30, use-it-or-lose-it clauses 10-15, and Fred Wilson argues for about 10. Assume the short end and schedule the IC meeting backward from it.

Can the company actually staff what the round promises

Check whether the company can hire the specialists its raise assumes, because talent supply is a real and checkable execution constraint few investors examine. A Series B story built on scaling an ML team is only as good as the pool it can recruit from.

number: 7.8%
label: Share of US software engineers who are ML-skilled, in Refolk's index
note: US ML-skilled engineers number 27,121 against a US pool of 346,288, so specialist hiring is a narrow funnel.