A full-time CFO picks a stack once and lives with it for years. You inherit four.
One client runs Xero with a bookkeeper in another time zone. One has Pennylane and a French accountant who closes on the 12th. One is German and sends you a DATEV export when the Steuerberater gets to it. The fourth has QuickBooks, Stripe and a forecast in a spreadsheet the founder built in 2023.
According to CFO Connect's 2026 salary benchmark, 51% of fractional CFOs work with 3 or 4 clients at once. That single number changes what "tech stack" means. Most published stack guides are lists of logos for a company that gets to choose. Yours is a question of which parts you carry between clients, and which parts you leave where they are.
Here is the stack, organised by layer.
Start With Ownership, Not Categories
Every layer below falls on one side of a line.
The client owns the ledger, the bank accounts, the billing system, the payroll, and every number inside them. When the engagement ends, all of that stays, and it should.
You own the method. The structure of your forecast, the way you map a chart of accounts to reporting lines, the conventions you apply (sign, materiality, what counts as recurring), the layout of your board pack, the prompts that work.
Get this line wrong in one direction and you rebuild your method from scratch at every new client. Get it wrong in the other and you end up holding client data in your own accounts, which is a confidentiality problem before it is a tooling problem.
Layer 1: The Ledger (the Client's)
You rarely choose it, and the market is fragmented. In the CFO Connect Top CFO Tools Report 2026, a survey of 215 finance leaders fielded in June and July 2026, the most-cited cloud accounting platforms were Xero (14%), NetSuite (10%), QuickBooks (8%), Pennylane (7%) and DATEV (7%), with Odoo, SAP and Sage behind. No platform clears 15%.
Among companies under 50 employees, the segment where most fractional work happens, Xero leads at 30% and QuickBooks follows at 14% (same report). By country, the picture shifts again: Pennylane in France, DATEV in Germany through the Steuerberater, Sage across the UK and continental SMEs.
Your job at this layer is access, not administration. Ask for an advisor or accountant seat on the client's own subscription, never import their books into an account you pay for. What you need out of the ledger is modest: a monthly P&L, a monthly balance sheet and the general ledger underneath, over the longest window available. The QuickBooks version of that routine is in How to Analyze Your QuickBooks Data with AI, and the Xero one in How to Use AI with Xero.
Layer 2: Bank Feeds and Billing
The ledger tells you what was booked. The bank tells you what is there today. For a cash forecast, which is often the first thing a fractional CFO is hired to produce, the gap between the two matters.
Most bank feeds already land in the ledger: Qonto, Mercury, Revolut Business or a traditional bank, synced for the bookkeeper's reconciliation. Billing sits beside it, often Stripe for subscription businesses. Treasury rarely has its own tool at this size. The same CFO Connect report found that 55% of finance teams manage treasury in spreadsheets and 17% have no treasury tool at all.
The practical rule: pull cash from the bank layer, not from the ledger, whenever the close runs more than a few days behind. A 13-week cash view built on last month's booked balance is already 3 weeks stale when you present it.
Layer 3: Close and Reconciliation
This layer is usually not yours. The bookkeeper or the external accountant closes the books, inside the ledger's own reconciliation rules or with a dedicated close tool.
What you own is the calendar and the standard: which day the numbers are final, what "final" means, which accounts you check before you accept the close. Agree it in writing in the first month of the engagement. Most of the fractional CFO's month is scheduled backwards from that date.
AI helps here, and finance teams know it: reconciliations were cited as the best AI use case by 9% of respondents in the CFO Connect survey, behind financial analysis (20%), reporting (12%) and modelling (12%). The method that keeps an audit trail is in How to Use AI for Your Monthly Financial Close.
Layer 4: The Model (Where the Forecast Lives)
This is the layer that decides whether your practice scales.
The market has not moved much. In the CFO Connect report, 76% of finance teams run FP&A in spreadsheets, up from 71% the year before, and another 7% use no tool at all. Among companies under 50 employees, the spreadsheet share is 85%.
Dedicated FP&A platforms exist and are good at what they are built for. Abacum, Pigment, Jirav, Runway, Mosaic and others give a single company a planning system with integrations, workflows and permissions. They fit best when the client has a finance team that will keep running the platform after you leave. If that is the case, build there, and hand over a system rather than a file.
For the other clients, the forecast lives in a spreadsheet. That is portable, which is why we all keep using it. It is also where the rebuild happens: a new client, a new chart of accounts, a model reassembled by hand or regenerated by an AI that re-decides the structure every session. Two companion pieces go deeper: taking over a client's existing model, and a template you can reuse across clients.
The test for this layer is simple. Can you open a new client on Monday and have their actuals sitting in your standard structure by Wednesday, without rebuilding the formulas?
Layer 5: Reporting and the Board Pack
The board pack is the part the client sees, so it is the part they judge you on.
Tooling varies: Excel and PowerPoint or Google Slides for the pack itself, a reporting add-on on the ledger (Fathom and Syft are common on Xero and QuickBooks), a BI tool such as Power BI, Looker Studio or Metabase once the client wants live dashboards.
Two rules hold whatever you use. The pack draws from the model, not directly from the ledger, otherwise you end up with two versions of revenue in the same meeting. And the layout is yours: the same structure, the same order, the same commentary format across every client. That consistency is the cheapest leverage in the job. Your fourth board pack of the month should take a fraction of the first.
Layer 6: The AI Assistant, and How It Connects
The assistant layer changed fastest. The CFO Connect report found that 67% of finance teams use AI tools in 2026, against 31% in 2024. Asked which tool, 41% named Claude, 12% Gemini, 10% ChatGPT and 9% Copilot, with Claude Code at a further 5%. Dedicated finance AI platforms were each named by 1% or less.
There are three ways an assistant reaches a client's numbers, and each has a different profile.
- Upload an export. You download the P&L, drop it into the chat, ask the question. Nothing to set up, and nothing persists after the session.
- Connect the ledger over MCP. The Model Context Protocol is an open standard for letting an assistant call external tools. Xero publishes an official MCP server, and so does Intuit for QuickBooks Online. The assistant can query the ledger directly instead of waiting for a CSV. Note that these servers expose create and update operations as well as reads. On a ledger you do not own, read-only is the right scope for analysis.
- Connect a model layer over MCP. The assistant works on a structured model rather than on raw transactions, so the mapping and the forecast it builds are still there next month.
Then comes the question that matters more than which assistant: whose account does it run on, and has the client agreed? Some client agreements now prohibit putting financials into an LLM outright, others only require consent or specific retention terms. What those clauses actually forbid, and the workflows that stay inside them, is covered in Your Client's Contract Says No AI. What happens to the data on each plan is in AI and Confidential Financial Data. Keep a separate project or workspace per client, and never let one client's numbers sit in a conversation where another client's are.
The Portability Test
Run it on your own stack. For each item, ask whether it goes with you to the next client or stays where it is.
Goes with you:
- the model structure (P&L, cash flow and balance sheet, and how they connect)
- the mapping templates per chart of accounts: French PCG, SKR03 and SKR04, a QuickBooks default chart
- your conventions, written down once, so an assistant reads them instead of guessing
- the board pack layout and the commentary format
- the prompts that have survived more than one client
Stays with the client:
- the ledger, the bank, the billing data, the credentials
- the actuals and the forecast values
- a copy of the model, in a format they can open without you
The last line is the one that gets forgotten. When you leave, the client should keep a working model in Excel, with formulas intact, that their next CFO can read. Anything else is lock-in, even if you did not intend it.
Where Layerz Fits
Layerz sits in layer 4. It is a structured spreadsheet, ready for AI: the model structure is kept separate from each client's data, so the same structure can be reused from one client to the next while the values stay in each client's own model.
Claude drives it over MCP. It pulls actuals from Pennylane, Qonto and Stripe through native connectors, and takes FEC and DATEV files, or any ledger report exported to Excel or CSV. There is no native QuickBooks or Xero connector today: you export the reports, and Claude maps them into the model. It also imports an existing Excel model, which is usually where a new engagement starts. Every model exports to clean Excel with the formulas intact, so the client keeps a file their next CFO can open.
It is free for one person working alone, with unlimited models, which is the normal shape of a fractional practice. If a client runs a full FP&A platform with a team behind it, build there instead. Layerz is for the clients whose forecast would otherwise live in a spreadsheet you rebuild every quarter.
The Takeaway
You will not get to standardise your clients' ledgers, banks or accountants, and you should not try. That half of the stack belongs to them.
The other half is yours: the model structure, the mappings, the conventions, the board pack. Pick tools for that half on one criterion. Does it make the fifth client cheaper than the first?