Unbundling the ERP: Tackling CFO Tech Legacy IV
Why this is in the vault
Part 4 (capstone) of the Tech Legacy arc. The Secret CFO's final architectural prescription: shrink the ERP back to its core strength (GL + sub-ledgers), build a 4-layer modern stack around it, let AI structure the messy data instead of forcing humans to clean upstream. The four layers - (1) Transaction Layer (best-of-breed tools), (2) Accounting System of Record (AI-standardized metadata), (3) Data Integration Layer (AI structures messy inputs), (4) Intelligence Canvas (natural language queries on trusted data) - are an architecture diagram for the harness-engineering thesis. The 5-step prioritization framework (Eliminate → Map → RAG → Grade data provenance impact → Grade switching cost) is operationally portable. The decision rule on whether to replace core ERP: unbundle around it first, only replace if the legacy system lacks modern APIs.
Mapping against Ray Data Co
This is the architectural capstone of the entire CFO Secrets corpus relative to RDCO. Specific maps:
- The 4-layer modern stack IS the harness-engineering architecture stated explicitly. Layer 1 is the best-of-breed tooling (the data-engineer's choice of warehouse, transformation, observability), Layer 2-3 are the harness assembly that makes the heterogeneous tools work together, Layer 4 is the consumption surface. This is the strongest external articulation of the [[06-reference/concepts/2026-05-10-harness-moat-two-layers-portability]] thesis to date - and it comes from a CFO buyer, not an AI builder, which is the validation that matters.
- MAC's product framing: this 4-layer stack IS MAC's TOC. Each layer can be a content cluster: transaction-layer choice (warehouse + transformation), system-of-record discipline (data contracts + metadata), integration layer (the harness itself - this is the load-bearing chapter), intelligence canvas (the consumption-side discipline). MAC ships a 4-cluster info-product organized around this architecture.
- The 5-step prioritization framework (Eliminate → Map → RAG → Grade provenance → Grade switching cost) is portable straight into how MAC's buyer should sequence their modernization. MAC sells the framework as a one-page diagnostic.
- The "kill the term ERP altogether" provocation is the kind of sharp re-frame Sanity Check could mirror in the data-engineering space: "kill the term Modern Data Stack altogether" - the unbundled-around-warehouse pattern is mature enough that the unifying-noun is now the trap. SC piece in this.
- RDCO's own infra: the unbundle-first-replace-second decision rule applies to my own stack. I shouldn't replace existing tools (Notion, vault filesystem, Anthropic API) until I've stripped non-core functions and assessed whether each tool has the modern APIs to participate in the harness. So far they do, so the discipline is unbundling, not replacement.
- The "next 5 years belong to CFOs prepared to stake reputation on transformation" close is the founder-level framing of why the L5 north star spend on agent capability is worth the asymmetric career risk Tech Legacy I called out.
⚠️ Sponsorship
Sponsored by Maximor AI (recurring across all four parts). Topic-matched - Maximor's product positioning slots into Layer 2-3 of the proposed stack. Disclosure clean. The architecture argument is editorially independent but the sponsor positioning aligns with the recommended stack shape.
Related
- [[06-reference/2026-03-07-cfosecrets-shelfware-shenanigans-tech-legacy-i]] - Tech Legacy I, the diagnosis
- [[06-reference/2026-03-14-cfosecrets-how-you-got-here-tech-legacy-ii]] - Tech Legacy II, the 7-eras history
- [[06-reference/2026-03-21-cfosecrets-bad-data-where-to-start-tech-legacy-iii]] - Tech Legacy III, the data-quality 2x2 (data layer of this architecture)
- [[06-reference/concepts/2026-05-10-harness-moat-two-layers-portability]] - the harness-moat thesis; this article is the architectural diagram of it
- [[06-reference/2026-04-28-cfosecrets-finance-stack-of-the-future-unbundled-erp]] - the Mailbag-format unbundled-ERP companion; this part is the deeper version
- [[06-reference/2026-04-30-mac-bet-architecture-audit]] - MAC bet architecture; the 4-layer stack is MAC's TOC
- [[06-reference/2026-03-29-cfosecrets-financeos-datarails-six-agent-orchestration]] - Datarails Six-Agent; one productized instantiation of Layer 4 (Intelligence Canvas)