QuickBooks

Where does your data go?

Your financial data goes to Intuit's cloud — and the story here is about platform, not training:

• Intuit pools your data across its products. By design, your QuickBooks books, your Mailchimp customer lists, and your Credit Karma profile "live together" — Intuit's own description. The benefit: pre-filled loan applications, cross-product recommendations. The cost: your bookkeeping feeds marketing and credit offers across the family.
• The one carve-out: TurboTax preparation data stays in TurboTax and goes nowhere but the IRS without your explicit permission.
• Your data won't train partner AI models. Intuit's agreements with OpenAI and Anthropic explicitly prohibit it — a written, contractual commitment that puts QuickBooks ahead of the consumer chatbots on this point.
• The AI agents shipped without an opt-out. When the agent team reached all QuickBooks Online users (mid-2025), there was no way to decline — categorizing, invoice chasing, and drafted outreach run in your books whether you asked or not.

What happens when it's wrong?

This is the sharpest gap between marketing and reality in the accounting category:

• The claim: Intuit states 95% categorization accuracy, 12 hours saved monthly, invoices paid ~5 days faster. No independent audit backs any of it — all vendor-stated launch figures.
• The field reality: bookkeeper testing measures Intuit Assist at roughly 50% accuracy on novel transactions — the ones bank rules don't cover. On a 600-transaction month with 200 novel entries, that's ~100 corrections after the AI weighed in. Half-right is what no AI gets you.
• The failure mechanism is documented: Intuit blends your company's history with a population model of everyone else's books — and the population half can win. One bookkeeper's client paid Delta Dentistry; the AI kept coding it as travel, because everyone else's Delta is an airline. The reviewer never sees which layer decided.
• Two postures in one product: the chat assistant refuses to invent figures on empty books — names the report, gives a row count. But receipt extraction categorizes confidently without being asked; one test had a "Pallet delivery" invoice filed as Equipment leasing, no doubt flagged anywhere.

What's it actually bad at?

• The agents are tier-gated, and the tiers moved. Simple Start gets suggestions only. The agents people actually want — Accounting and Payments — start at Essentials. Customer and Sales Tax AI need Plus. Finance and Project Management AI need Advanced. The "team of agents" pitch is an Advanced pitch.
• The chat is metered. 25 prompts per month even on the $340 plan. The $10/month add-on buys 100 more — no unlimited option exists at any price.
• No automated month-end close. Accruals and adjusting entries stay manual at every tier. When a miscategorization surfaces at tax time, you're liable.
• The AI panels slow the platform down. Multiple reviewers report the 2025 redesign — AI embedded into every window — dragging performance.
• Support got worse as AI went in. Below Advanced: chatbots and offshore queues. The top tier gets humans.

What does it cost really?

The traps: every tier advertises "50% off for 3 months" — $19, $37.50, $57.50, $137.50 — which resets to the new, higher price in month four. Third price increase in two years, announced 30 days before each renewal. And one company per subscription: run three LLCs, buy three subscriptions, consolidate in a spreadsheet. Payroll, human CPA review (QuickBooks Live), and Bill Pay below Advanced are separate add-ons.

What does it replace — and what do you still do?

The Payments Agent's invoice chasing and receipt capture genuinely remove work — expense matching, payment reminders, reconciliation flagging. You still own: reviewing every categorization, the month-end close, all adjusting entries, and final judgment on your numbers. Nothing here replaces your accountant. The 13-hours-saved claim is marketing; the defensible claim is "fewer keystrokes, same review."

What the ecosystem offers

• Conversational answers about your numbers (Aug 2026): ask "why did margins drop" and get the likely causes with context, not just a report. You can also edit financial forecasts in plain language and test scenarios.
• Invoicing on autopilot (beta): create invoices by describing them, upload a document and get a batch of invoices drafted, with payment reminders that send only after you approve.
• Build your own automations: describe a routine — a weekly report, a payment reminder — and schedule it to run itself, with a log of everything it did.
• The chatbots can now touch your money: QuickBooks Payments works inside Claude and ChatGPT — draft and send invoices and payment links from the chat window, with your approval before anything sends. A real convergence between this review and the chatbot listings.
• Advanced-tier extras: an AI service that keeps books current all month (flagging only what needs a human), Bill Pay Elite now included at no extra cost, KPI dashboards with drill-downs, and industry workflows for construction, manufacturing, and nonprofits.
• The unchanged caveats: the chat stays admin-only and beta, with limited monthly capacity — and the new depth lives in the $340 tier and above.

If I stop paying, what do I lose?

• Export: yes, self-serve — Settings → Export Data pulls reports and lists (General Ledger, Balance Sheet, P&L, customers, vendors) to Excel, or to QuickBooks Desktop.
• Read-only for exactly one year. After cancellation you can view, print, and export for 12 months — then the data needs a resubscription to reach. Trials get only 90 days. Miss the window and your books sit behind a paywall.
• The better exit is a downgrade, not a cancel — switching plans keeps data, settings, and history active.
• Export before you cancel: reconcile through the cutoff, lock the period, pull the reports. The cancellation flow itself is easy, with retention upsells alongside.
• No prorated refunds on monthly plans — the 60-day money-back guarantee is annual subscriptions only.

Who is this NOT for?

Anyone on a sub-$85 budget who wanted the agents — that's the floor where the real AI starts. Businesses needing multi-entity consolidation, non-US entities, or an automated close. And anyone who expected the AI to be right most of the time on unfamiliar transactions — the data says plan to review half.