06-reference

secret cfo great unbundling building fpa iv

2026-07-25·reference·source: CFO Secrets·by The Secret CFO (anon)
fp&afinance-org-designai-for-financeseries-building-fpastrategic-finance

"The Great Unbundling" — @The Secret CFO

Why this is in the vault

Part IV — the closing installment — of the four-week Building FP&A series (Parts I-III already filed). This issue turns from "how to design FP&A today" to seven decade-scale predictions about how AI redraws the function's boundaries, and it's the most direct articulation yet of a thesis RDCO already operates on: bundled knowledge-work functions decompose into a data/context layer plus a judgment layer, and the judgment layer is where the durable value sits.

⚠️ Sponsorship

Sponsored by Summation (AI Analyst product, summation.com/ai-analyst) — same sponsor as Parts II and III. This is now a 3-of-4-issue run for the same sponsor within a single series, which breaks the "assume rotating, unknown-until-scanned" pattern the README currently documents for CFO Secrets generally. Placement: narrative ad with an embedded case study (Fanatics, claimed "$10m in growth and savings uncovered" after adopting Summation for automated overnight reporting), plus a "Thank you to our sponsor" footer block. No disclosed author/investor relationship — reads as a clean paid placement, direct-to-CFO positioning. Not a vendor RDCO uses.

The core argument

The author opens by disclaiming prophecy — dismissing LinkedIn AI-hype predictions as mostly written by people who've never run a finance function — then argues FP&A's long-term shape is more uncertain than controllership's, because controllership has a structural moat (audit/compliance/legal accountability) and FP&A doesn't: it's a bundle of activities, not a protected discipline, and AI is "redrawing the lines between functions, not just redrawing work inside them."

Seven numbered predictions carry the piece:

  1. Stack consolidation — the all-in-one ERP/FP&A-suite model is a trap (rigid, expensive, drives teams back to shadow-Excel); AI should instead enable one trusted data foundation with many flexible query/model/report layers on top.
  2. Excel is still the canvas — finance stays a "rows-and-columns discipline"; the future interface is probably chat-controlled but still spreadsheet-shaped, backed by a trusted data layer.
  3. The end of the "data waiter" — AI/self-serve chatbots should kill the low-value work of FP&A analysts fetching and packaging information for executives; what survives is owning the systems, definitions, and judgment behind reliable self-serve.
  4. The great unbundling (title concept) — FP&A slowly comes apart over a decade: some work disappears, some moves to better-suited homes, agents absorb routine pieces, executives self-serve directly. Governance shifts from org charts to "context, permissions, standards, and guardrails given to agents."
  5. Everything is finance, but not everything financializes cleanly — most decisions resist clean ROI math, leading to "false financialization" (forcing an ROI number onto things like leadership development). The fix is "externalizing FP&A" — building financial thinking into business leaders' own decision-making rather than routing everything through a spreadsheet.
  6. Controlling absorbs routine FP&A work — reconciliation-heavy budgeting/forecasting work migrates to a controllership team with an accounting mindset, freeing FP&A for judgment/challenge work. A 2x2 grid frames this as the "central hub" of FP&A getting absorbed into controlling while "support spokes" move closer to the business.
  7. A rebrand — once routine work is stripped out, what remains is judgment-heavy work on high-stakes, low-guardrail decisions (capex, M&A, market entry, capital allocation). The author floats "Strategic Finance," possibly merged with corporate strategy/corp dev, while admitting "I don't love it."

Close: four "no-regrets moves" — consolidate the stack around one trusted data/context layer, aggressively automate routine tasks, push business/commercial context into controlling teams, and develop remaining FP&A people for judgment/ambiguity work. Explicitly framed as planning assumptions, not predictions to bet on. The series recaps Parts I-III by link (activity taxonomy of "35 sub-activities across seven buckets" from Part II is reused directly here) and signals a format change next month.

Mapping against Ray Data Co

Prediction #3 (the "data waiter" dying) is the sharpest direct hit on RDCO's own agent-deployer thesis: the author is describing, from inside a different discipline, the exact same claim RDCO makes about data engineering — that low-value information-fetching and packaging work gets automated to zero value, and durable value concentrates in owning systems, definitions, and judgment. It's independent convergence with the harness-engineering thesis (agents absorb the mechanical layer, humans own guardrails/judgment) from a CFO seat rather than a data-eng one, which is useful corroboration when pitching the thesis to non-technical operators. Prediction #6's controlling-absorbs-routine-work move is a concrete instance of "AI redraws org boundaries, not just work inside them" — worth testing against CAF's own DIE hub-and-spoke model, where the same central-hub-vs-support-spoke tension exists for the "Fabric" governed knowledge graph.

Related