06-reference

mostlymetrics annual planning five inputs

2026-09-15·reference·source: Mostly Metrics·by CJ Gustafson
annual-planningfpandacontinuous-forecastingheadcount-planningmostlymetricsbrex

Why this is in the vault

CJ Gustafson names the five clean data inputs (current headcount, open headcount, trailing P&L, sales-rep attainment, trailing-12-month costs) every annual plan is built on, and argues that once those feeds are clean and AI does the assembly instead of an analyst doing it by hand, the once-a-year "foraging" exercise turns into a standing monthly refresh — the September anchor stays, but the plan stops being a frozen artifact.

The core argument

Gustafson's traditional cadence: lock the baseline through September, spend Q4 "turning the screws" on the final three months of the current year plus all twelve of the next (15 months total), and walk into the Q3 board meeting with an approved number around Thanksgiving. He argues against the common alternative — waiting until Q4 actuals land in February/March to approve — because delaying decisions in a hypergrowth company costs more in lost momentum than it buys in accuracy. His update to the model: nothing about the September cutoff is load-bearing anymore except tradition. The cutoff existed only because gathering and scrubbing five messy inputs by hand took months; if headcount, cost, and attainment data are clean and piped in and AI (not "an analyst named Chester") does the assembly, those five inputs become a feed you can refresh monthly instead of forage once a year — so the plan still gets approved on September numbers, but the last-three-months blind spot shrinks continuously instead of sitting frozen until January. On the individual inputs: open headcount should split into "actively recruiting" vs. "approved but not worked" with different fill-probability discounting, and budget envelopes should be framed as a "theoretical max not to exceed" rather than "additional headcount," because "additional" framing is what causes roles to get double-counted or silently dropped. He deliberately excludes attrition from the five inputs — treating any gap between people leaving and joining as unbudgeted upside to find later, not a given to bake in.

Mapping against Ray Data Co

Two concrete connections. First, the household cash-flow-visibility project (home-rebuild-2027) is mid-build on exactly the sequencing problem Gustafson describes: the next step there is a three-scenario page, and his rule — pull the five inputs clean and trustworthy before you model a single scenario — is a direct corrective, since Michelle's income and childcare costs sit outside Monarch entirely (the real finding from the 2026-08 cash-flow thread) the way open headcount sits outside a naive headcount export if you don't deliberately segment for it. Second, his "continuous planning" thesis — clean piped data plus AI assembly turns a once-a-year exercise into a standing monthly feed — is functionally the same bet the OI (Organizational Intelligence) umbrella is making with Pulse: replacing a point-in-time planning artifact with a continuously-refreshed one. It's a sharper, borrowed vocabulary ("theoretical max not to exceed" vs. "additional headcount") for a framing problem RDCO already has in both places.

⚠️ Sponsorship

Single sponsor block at the top of the issue: "Mostly metrics is proudly powered by Brex," an intelligent finance/expense platform ad ("Agentic Finance... expenses handled automatically, policy enforced before the spend happens, books closed in minutes") with a partner-tagged link (brex.com/grow?partnerId=metrics). Brex is an already-known recurring paid sponsor per the process-newsletter README. No other paid or self-promo sponsor block in this issue — the closing "Run the Numbers Podcast" segment (interview with RBC analyst Matt Hedberg) is the sender's own owned-media cross-promo, not a paid third party or curated external link, consistent with how the README already tracks that podcast as part of the regular Mostly Metrics content mix.

Related