A fractional CFO is the right hire for judgment. It is the wrong fix for a split stack. If you already pay a bookkeeper and a fraction of a finance leader and still do not have a close you trust, adding hours will not create a source of truth. The work you are buying is reconstruction: Stripe export, billing CSV, contract PDF, QuickBooks, spreadsheet.
This is the founder decision, including teams setting books up for the first time. The question is not “software vs a human.” It is whether books, billing, and investor-ready numbers can live in one system — or whether you will keep paying people to stitch three tools together.
What a fractional CFO is actually for
Hire one when you need:
- Fundraising narrative, board packs, and cash planning
- Judgment on pricing, hiring, and runway
- A grown-up to challenge the forecast
That work is not categorising payouts. It is not rebuilding deferred revenue from a billing export. It is not explaining why the processor settlement does not match invoices.
If most of the retainer is spent on the second list, you do not have a CFO gap. You have an architecture gap. The person is expensive glue.
The labor stack founders accidentally buy
The default path is familiar:
- Open QuickBooks or Xero because every accountant knows it
- Add Stripe, a billing tool, or both
- Hire a bookkeeper when categorisation slips
- Hire a fractional CFO when the board starts asking questions the books cannot answer
Each step feels rational. Together they are a standing reconstruction process. Usage, compute, and credits make it worse: the commercial terms never lived in the ledger, so every close is an interpretation of last month’s activity.
Outsourced bookkeeping firms sit on the same foundation. They can be excellent at keeping a small-business file clean. They cannot make a seat-based ledger understand metered revenue. You still own the gap between what customers consumed and what the reports say.
When to replace the stack instead of hiring into it
Replace the stack when you can name how you make money as usage, compute, or commitments — and the current file cannot hold that without a side model.
Signals:
- You have a billing tool and an accounting tool and a spreadsheet that is more trusted than either
- Close is late because someone is matching payouts to invoices by hand
- The fractional CFO’s first request is “send me the Stripe report and the contract folder”
- Investor questions take a week because revenue, cash, and bookings are three different stories
A modern financial operating system is not “AI bookkeeping” on top of QuickBooks. It is one system that bills customers, keeps books, and produces numbers you can take to a board without a cleanup sprint. Humans still review exceptions. They should not be the integration layer.
If you have no ledger yet
Greenfield is the cheaper version of the same decision.
If you are at incorporation, first bank account, or first invoices, and you already know revenue will be usage, compute, credits, or commitments, do not start in QuickBooks “for now.” Starting wrong is a 12–18 month migration: chart of accounts built for a simple services business, billing bolted on later, then a reconstruction when the first serious investor or controller shows up.
The wedge is not “we help you open the books.” It is: set up one ledger before you outgrow three tools. Spreadsheets plus manual workflows are the greenfield signal — as long as you can name usage, compute, or commitments. If you cannot, a simple accounting file may be enough until the model is clear.
A practical split
Keep (or hire) a fractional CFO for judgment, fundraising, and operating cadence — once the system produces current numbers.
Do not hire one to be your billing-to-books pipeline. That is software and a ledger that maintains state as activity happens.
Do not start in a generic ledger if you already know the revenue model will not fit it.
Lucius is built for that founder job: one system instead of QuickBooks plus a billing tool plus a human close. If you are still deciding between another contractor and a different architecture, the test is simple. Can the person you want to hire name usage, compute, or commitments as the reason the current books fail — or are they proposing more hours on the same file?