"Tactics for Budgeting at Hyperscale"

Why this is in the vault

CJ's interview with Vercel's CFO Marten Abrahamsen lays out a trigger-based multi-scenario planning framework that replaces the single-forecast-defended-to-death model — directly applicable to how the RDCO AI-COO should track competing bets and revenue scenarios rather than maintaining one brittle projection.

⚠️ Sponsorship

Two promotional blocks:

Brex — paid third-party sponsor. CJ runs "Mostly Media on Brex" and plugs Brex's AI-powered agentic finance platform with a referral link (brex.com/grow?partnerId=metrics). Full commercial relationship, disclosed upfront.
Mostly Talent (mostlytalent.com) — CJ's own recruiting arm for finance/accounting placements. Self-promo/house ad, not a third-party paid sponsor. Appears twice (mid-body and footer). Set sponsor_entity to reflect both.

The core argument

CJ opens with his own hypergrowth FP&A trauma — running financials at a company scaling $10M → $150M in under 3 years, where every plan was obsolete before the quarter ended. The fix he learned too late, and that Vercel's CFO has operationalized, is the three-rail plan:

Framework: Three Plans, No Default Pick

Build three parallel annual operating plans at year-start: base, average, ambitious. Each has its own revenue trajectory AND its own resource/headcount/spend allocation pre-attached.
Don't choose one in January. Give the whole org (and the board) visibility into all three, then let actual business performance tell you which rail you're tracking to.
Monthly job shifts: instead of rebuilding the forecast, you're replotting position — which percentile growth curve is the business on right now?

Trigger-based spending unlock

Each plan has pre-negotiated trigger points. When the business hits a trigger, the corresponding spending envelope is already approved and expected to fire.
This eliminates the approval-deck cycle and the "hurry-up-and-wait" bottleneck. If you're on the ambitious rail and don't spend into it, you've created a P&L gap — you're underfunding the demand gen for the revenue target you just qualified for.
Marten's framing: "The number is less a target, and more a trigger."

Radical board transparency

Marten operates "full kimono" with the board: explicitly puts the uncertainty on the table and shows all three rails rather than defending one number. Easier than a point estimate defended for 12 months.

The "good problem to have" trap

Critical caveat: trigger-based fast-spending only works if the underlying infrastructure can absorb it. Manual processes deferred because "that's a good problem to have" become live grenades. Marten's hard rule: solve the infrastructure problem now, never punt it.
CJ's war story: sales commissions in a single Excel tab blew up into a six-figure quarterly true-up that fouled projections and blew up the comp team when headcount scaled.

Issue contents

Main piece: Deep-dive on Vercel's three-rail annual operating plan framework, drawn from CJ's in-person interview with Marten Abrahamsen at the NYSE. ~2,000 words, analytical + narrative.
Run the Numbers podcast: Episode cross-promo — same Marten Abrahamsen interview available on YouTube/Apple/Spotify. Topics: AI in finance (rev rec, forecasting, KPI dashboards), PLG + consumption pricing, "speeding tickets vs. parking tickets" risk framework.
Mostly Talent: Two recruiting arm plugs (self-promo, not third-party paid).

Mapping against Ray Data Co

The three-rail trigger-based model maps directly to how the AI-COO's planning loop should be restructured: instead of a single RDCO revenue/pipeline forecast, maintain three parallel scenario tracks (conservative: phData W2 + minimal RDCO product revenue; base: phData + 1-2 product bets traction; ambitious: phData + meaningful RDCO SaaS/plugin revenue) with pre-defined monthly trigger points the autonomous loop checks. When a trigger fires — e.g., plugin marketplace hits X installs, a phData deal hits a certain ARR threshold — the spending/investment allocation for that rail activates without a re-approval cycle.

Secondary mapping: CJ's "good problem to have" anecdote is a direct audit prompt for RDCO. What processes are currently manual because volume is low enough to hand-crank? The iMessage/Discord channel routing, newsletter processing, and Notion board updates are all single-Excel-tab equivalents at current scale. As the autonomous agent loop increases throughput, these become live grenades. The Vercel lesson is to solve the infrastructure constraint before hitting the rail that stresses it.

Third application: the "replotting not rebuilding" principle applies to the AI-COO's phData pipeline tracking — monthly job is to assess which scenario track the phData opportunity pipeline is on, not to rebuild the projection from scratch each cycle.

[[2026-07-02-mostly-metrics-rolling-13-week-cash-flow-claude-excel]] — cash management at speed; the rolling-13-week model is the tactical complement to this issue's strategic multi-plan framework [[2026-06-16-mostly-metrics-rivian-capacity-planning]] — capacity planning under uncertainty; Rivian's resource-against-scenarios approach is structurally parallel to Vercel's three-rail spending triggers [[2026-06-04-mostly-metrics-consumption-based-arr]] — Vercel's business model context; consumption-based ARR is precisely why forecasting is so hard there and why multi-plan is the only viable approach