This is the thesis behind Oh Ventures. It is worth stating directly.
The transactional layer is gone.
Within the next 24 to 36 months, the bookkeeping, reconciliation, and routine reporting layer of the CFO function will be largely automated. Not in theory. In practice, inside live companies. The tools already exist. The integration is the only remaining work.
What gets compressed:
- Bookkeeping and reconciliation. Bank feeds, vendor invoices, and routine journal entries no longer require a human in the loop.
- Variance analysis. An AI agent can compare actuals to plan, flag exceptions, and write the commentary faster and more consistently than a junior analyst.
- Routine reporting. Monthly close packs, board materials, and KPI dashboards can be generated from the underlying data layer on demand.
- Document review. Contract analysis, vendor terms comparison, and routine compliance checks.
- Diligence prep. Q&A response drafting, data room population, and trace-back from financial line items to source documents.
Each of these used to support a full-time role inside a company. Some still do. None of them will five years from now.
What does not compress.
The work that does not get automated is the work where the answer depends on context that does not live in the data.
Capital strategy.
Whether to raise now or in six months. Whether to take a SAFE at a $5M cap or wait for a priced round at $8M. Whether to take the strategic investor or the financial one. None of these have a right answer that can be derived from financial data. They depend on the founder's risk appetite, the team's stability, the market window, and the dynamics of the specific investors in conversation.
Hiring sequencing.
The next five hires define the company. The order they come in matters more than the names. Whether the next hire is a senior engineer, a VP of Sales, or a Head of Finance is a judgment call about which constraint is binding right now. AI can model the financial impact of each. It cannot tell you which constraint is binding.
Investor selection.
An AI can produce a list of 200 plausibly-aligned investors in twenty seconds. It cannot tell you which three you should actually meet with. That requires knowledge of the partners, their portfolios, their recent disappointments, their thesis evolution, and their internal politics.
Cap table architecture.
A tool can model the conversion math. It cannot tell you whether the cap table you are building will support a Series A in eighteen months, given the specific dynamics of how seed funds in your sector behave in priced rounds.
Negotiation.
Term sheets, board composition, anti-dilution provisions, pro rata rights. Every line is negotiated against a counterparty with their own pressure points and constraints. AI can simulate. It does not negotiate.
Where the work moves.
The CFO function does not shrink. It moves up. The role that used to be 70 percent transactional execution and 30 percent strategic judgment inverts. The new CFO function is 80 percent decision-grade work and 20 percent oversight of the automated stack underneath.
This is why fractional and embedded CFOs become more valuable, not less, as AI compresses the bottom of the stack. A founder no longer needs to pay a full-time CFO to run the close. They need a part-time operator who can hold the decisions the AI cannot make.
Two implications for founders.
1. Stop hiring for what AI compresses.
The full-time controller role that used to make sense at $5M ARR no longer makes sense. The work is automatable. Hire for the decision layer instead.
2. Stop paying full-time for fractional work.
If the work the CFO is doing is 20 percent decision-grade and 80 percent oversight, a full-time hire is overpriced. A fractional embedded CFO operating at the decision layer is the right shape.
The operator-built tooling thesis.
The other implication: most generic AI finance tools will lose to operator-built ones. Generic tools optimize for the data layer that is getting commoditized. Operator-built tools optimize for the decision layer that is not.
Oh Ventures OS, the platform we run engagements on, is built around this thesis. It compresses transactional work where compression is cheap. It exposes the decision layer where judgment compounds. The data layer is plumbing. The decision layer is the product.
What this means for the firm.
Oh Ventures is built for a world where the CFO function has moved up. We do not compete with bookkeepers. We do not compete with controllers. We do not compete with AI tools. We compete at the decision layer, where the work cannot be automated and the judgment compounds with experience.
That is the thesis. Everything we install operates from it.