“AI CFO” is a label five different products now wear, and they automate five different things: one runs the analysis a CFO would do, one wraps a human CFO in an AI platform, one sells CFO hours with an AI assistant attached, and two are bookkeeping tools with a question box. The label tells you almost nothing. What each one actually does, as described on its own site in September 2026, is below. The list is in no particular order, and the right answer depends on whether your problem is that nobody closes the books, that nobody reads them, or that nobody can say which product is losing money.
1. Zeni: a human finance team with AI underneath
What it automates: bookkeeping data entry, categorization, journal entries and bank reconciliation, per its FAQ, with a dedicated controller, bookkeeping manager and analyst reviewing the result. Zeni’s AI Bookkeeping plans start at $549 a month (Starter, pre-revenue) and $799 (Growth), with Enterprise custom. The CFO part is separate and human: Fractional CFO Lite at $1,599 a month plus a $2,000 setup fee covers budgeting, budget-versus-actuals and a 12-month cash forecast with a monthly one-hour call; Premium at $2,990 plus $6,000 setup adds revenue and COGS modeling and bi-weekly calls; Enterprise from $4,990 adds board participation and fundraising support.
What it does not automate: the CFO judgment itself. That is a person on a call. The FAQ also states QuickBooks Online Plus is required, services are US-only, and the integrations listed are QBO Plus, Avalara, Plaid, Stripe, PayPal and Square. No marketplaces, no inventory.
Fits: a venture-backed startup that wants to hand over the finance function and get a named CFO.
2. Pilot: CFO services with an AI answer box
What it automates: on the $99-a-month Essentials bookkeeping tier, AI categorizes, reconciles and closes cash-basis books and answers questions about past transactions and cash flow, including scenario runs. The CFO offering is separate, human and priced by tier on Pilot’s pricing page: Basic at $1,750 a month (custom model, budget versus actuals, scenario planning, monthly call), Essentials at $3,150 (cash and runway work, board reporting, bi-weekly call), Custom at $5,250 (13-week cash flow, investor updates, weekly calls), all billed annually.
What it does not automate: anything strategic. The AI answers questions about what already happened. The model, the forecast and the recommendation come from the CFO on the call. Pilot’s page also notes its CFO services do not include a corporate officer.
Fits: a startup or small business that wants an established firm, a person to call, and the option to grow from AI-only books to full service without changing vendors.
3. Digits: a ledger that answers questions
What it automates: categorization, reconciliation, live financial statements, invoicing and bill pay across all plans ($65, $100 and $250 a month), plus an assistant called Ask Digits that answers plain-language questions about the books. The $250 Pro tier automates accrual, depreciation and amortization schedules and runs an “Agentic Close” that gathers statements and flags anomalies. Digits’ FAQ describes an AI quality review that checks the books for errors.
What it does not automate: recommendations. Ask Digits is a query layer; it will tell you what runway looks like, not whether to cut a product. Digits’ page also lists CFO services as partner-provided through an accountant directory, not as a Digits function.
Fits: a service business or software startup whose revenue comes through banks, cards and Stripe, and which wants near-real-time books with a question box on top.
4. Puzzle: agents that run the close and write the narrative
What it automates: more of the close than the other four. Puzzle is a general ledger (its pricing page calls the Core plan “your QuickBooks replacement”) with three families of agents that consume credits: categorization agents the page says can reach “up to 98 percent no-touch,” close agents that run reconciliations, accuracy checks and variance explanations, and insights agents that generate financial narrative and cash runway models. Plans run from $25 a month billed annually to $300; the $100 Complete tier is where AI reconciliation and accuracy review turn on, with a money-back guarantee if the close does not get 50 percent faster under stated conditions.
What it does not automate: anything outside its own ledger. Puzzle’s agents work on Puzzle’s books, so adopting them means migrating off QuickBooks or Xero. And “insights agents” produce narrative and runway models, which is reporting in prose form, not a decision.
Fits: a startup that is willing to change ledgers to get the close automated.
5. ConnectBooks Crunch: analysis over settlement-level books
What it automates: the investigation. ConnectBooks is accounting software for marketplace sellers; it posts Amazon, Shopify, Walmart, TikTok Shop and eBay settlements into QuickBooks Online, QuickBooks Desktop Enterprise or Xero with FIFO cost of goods sold per unit and profit by SKU and channel. Crunch is the AI CFO built into it, in active beta. Its page describes a five-step chain for a question like “why am I down in profit this month”: find what changed across sales, pricing, ads, fees, returns and storage; compare the right periods; find the SKUs responsible; explain why; recommend what to do. The worked example on the page attributes 71 percent of an 18.4 percent profit decline to six SKUs, driven by ad spend and a fulfillment-fee band change. A second example ranks inventory for Q4 by cash contribution and returns a hold, discount, liquidate or remove call per SKU.
The company’s own writing on how an AI reading settlement-level books catches margin erosion early is candid about the limit: it is a detection and analysis layer, not a replacement for judgment, and it only works on books that are already correct at the transaction level.
What it does not automate: the bookkeeping close (a human still reviews the ledger), tax, payroll, or anything for a business without marketplace revenue. The underlying software starts at $149 a month per its pricing page and the FAQ notes there is no open API.
Fits: a multi-marketplace seller with enough SKUs that “which products” is a real question. Nobody else.
What the five have in common
None of them makes a decision. Zeni and Pilot put a human on the call to make it. Digits and Puzzle answer questions and write summaries so a human can make it faster. Crunch goes one step further by ranking options, and then hands the ranking to the seller. The Bureau of Labor Statistics’ Occupational Outlook Handbook entry for accountants and auditors, updated August 2026, projects 5 percent employment growth through 2035 and expects automation to make advisory and analytical duties more prominent rather than reduce demand. That is the honest frame for this category: the tools move the work up a level, they do not remove the person.
Choosing
If the books are not closing, buy the close: Zeni, Pilot Core or Puzzle, depending on how much human you want and whether you will change ledgers. If the books close but nobody reads them, buy the question box: Digits, or Pilot Essentials for a smaller business. If the books close, someone reads them, and the open question is which SKU or channel is bleeding, the only tool on this list built for that is Crunch, and it requires ConnectBooks underneath it.
Whatever you pick, the IRS guidance on recordkeeping still puts the burden of proof for every entry on the business, and the Small Business Administration’s guide to managing your business is a neutral primer on what a finance function needs to cover before any AI is bolted on. Prices and tier contents on all five sites change; verify before buying.
