Are QuickBooks' AI Agents Enough for Your Finance Team?

When a company employs a few hundred people, its finance team is often a controller and a handful of clerks, and their month splits in two.

One half is keeping the books right inside QuickBooks: categorizing every bank transaction, matching deposits to invoices, entering supplier bills, closing the month. The other half is getting into QuickBooks what happened somewhere else: the jobs finished, the shipments delivered and the hours approved that have to become customer invoices.

In our example, the second half is the larger bill: a company raising 1,500 invoice lines a month from its operations records spends about 75 hours of a billing clerk's month finding each record, matching the customer and the item, checking the price and typing it in: $31,563 a year at a clerk's pay with benefits.

Intuit's AI for that month is Intuit Intelligence, built into QuickBooks and released in beta with the Fall 2026 update, the latest form of the AI agents Intuit launched in 2025.

The owner's question is whether it covers the finance team's month or only part of it, so we weighed each AI feature against that month.

Inside the books, it is a useful first pass that a person still checks, and the plan you are on decides how much of it you get. It drafts invoices from documents someone hands it, so the billing that starts in your operations records still waits for a person.

At Qvery, we help growing companies cut what billing costs and put their clerks back on the accounts that need attention, with custom AI agents that collect your finished work where it is recorded, match every line to QuickBooks and prepare the invoices your billing lead releases.

Which QuickBooks You Run Decides Which AI You Get

The plans that fit a business with a few hundred staff are Advanced, the top QuickBooks Online plan and the only one with more than five users, and IES, the system Intuit sells above Advanced for larger companies and groups of companies.

The AI is spread across the plans below them, and each step up adds to it. Intuit shows each price at half off for the first three months.

Two lines in Intuit's own help center matter more than the table. The chat, where most of the asking happens, is for admin users only, with a monthly allowance that resets each billing cycle; a $10 add-on raises it, and Intuit says no plan offers unlimited prompts.

And Intuit Intelligence is available on QuickBooks Online and on Intuit's suite for accountants only: on IES, the step up for a growing company, it is early access, and only for customers with a single company.

IES publishes no price; Intuit agrees one per customer after a call. Moving up keeps the books where they are, since every plan from the free one to IES runs on one platform.

Advanced's 25 users are room for the finance team itself. They get tight once the people who approve bills and raise jobs across the company each need a login.

Categorizing Bank Transactions Is the Strongest Part, and It Still Needs a Reviewer

On the bank page, QuickBooks suggests a match or a category for every transaction from the company's own history and the full bank description, and marks each suggestion green, blue or orange for how much evidence sits behind it.

It finds partial and combined matches, and warns when a line looks like one whose record has not arrived yet.

From Essentials up, the lines it is most sure of are grouped as ready to post, so a person can post them in one step after a look, and a line it cannot place can go as a question to whoever made the purchase.

From Plus up, categorizing runs on its own as transactions arrive, and the AI reconciles, matching the books to the bank and flagging what does not tie out.

This is where the AI earns its place.

The bookkeepers who describe their routine mostly describe it the same way: the confident matches are quick to check, the guesses on new vendors are not, and they still look at a line before it posts.

Intuit says much the same in its terms: its assistants may be wrong, and the customer answers for what they do.

Verdict for this feature: worth turning on for a team whose transactions repeat, with a person posting what it suggests. A team with many new vendors each month gets more suggestions to correct.

Invoices Draft From the Documents You Hand It

A person uploads statements of work, sales orders or contracts, or points the chat at a Google Drive folder, and invoicing on autopilot, in beta from Essentials up, reads them, matches each to an existing customer and product, and lines up draft invoices for review. Nothing goes to a customer until someone selects send.

Intuit's own help article draws the edges. It takes five documents a message and works with Google Drive and no other cloud store, and its note on duplicates is the one to read before it bills anyone.

Reminders are the stronger half. A saved automation flags overdue invoices on the rules you set, a person picks which reminders go, and Intuit reports that customers using its AI-drafted reminders are paid 4 days sooner on average than customers sending standard ones, by its own comparison of US customers from March 2025 to March 2026.

Verdict for this feature: good for a company that bills from a statement of work or a sales order a few times a month. For one that raises hundreds of invoice lines from what its operations recorded, someone still gathers the documents, five at a time, and checks each draft against what was already billed.

The Chat Answers Questions About the Books, Within a Monthly Allowance

Ask it how the month went, which invoices are overdue or why a cost moved, and it answers from the company's own numbers. It builds reports on request, forecasts cash, and on Plus and Advanced runs what-if scenarios on a price change or a new hire. On Advanced it also writes the summary for a management report.

The limits are the ones above: admins only, a monthly allowance, and a beta that Intuit says can make mistakes. For a finance team where five people want to ask it things, the allowance and the admin rule decide more than the answers do.

The same answers now reach ChatGPT, Claude, Gemini, Perplexity and Meta's Muse through Intuit's own connectors, in beta, with invoicing in ChatGPT and Claude generally available. Nothing sends until a person approves it, and Intuit says the data stays in its systems and is not used to train its partners' models.

Read the terms before you let any assistant near the books. Intuit's terms say that what you give its assistants may be used to improve its products, and that you alone answer for what results from using them.

They also bar scraping and unauthorized access, so an outside agent belongs on Intuit's connectors or its API rather than on a person's login in a browser.

The Rest of the AI, One Line Each

• Bills from email: vendors send invoices to a QuickBooks address, and Bill Pay's AI turns them into bills ready to approve, then posts the payment
• Payroll: prepares a pay run from your phone, and runs it only when someone submits it (QuickBooks Workforce, sold separately)
• Sales tax: checks the return for errors before you file (Plus and up)
• Customers: sources leads and drafts follow-ups (Plus and up)
• Projects: creates projects and allocates their costs (Advanced)
• Agent Studio: a pilot, shown at Intuit's October conference, for accounting firms to build their own agents; not open to a company's own team

Turn It On for the Bookkeeping, and Keep a Person on What It Posts

QuickBooks' AI is a capable assistant for the work already in your books. The work that never reached them is still yours.

Turn it on if your finance team's month is mostly bank transactions, supplier bills and a modest number of invoices. On Plus or Advanced it takes the first pass of categorizing and matching off your team, and the reviewer it needs is already on it.

Do not count on it to carry the month if your invoices come from an operations system, a project tracker or a shared sheet, or if you are moving to IES for several companies and expect the newest AI on day one. The first is outside its reach, and the second is early access.

Where Billing Starts Outside QuickBooks, Qvery's Agents Take Over

A field job closes in the service app, a delivery is confirmed in the warehouse system, a consultant's hours are approved in the time tracker, a milestone is signed off in a spreadsheet. None of it is in QuickBooks yet.

Someone finds each one, matches it to the right customer and item in QuickBooks, checks the price and the purchase order, makes sure it was not billed last week, and types it.

In our example that is 1,500 invoice lines a month at about 3 minutes each, 75 hours of a billing clerk's month. Federal wage data put billing clerks at a mean $24.55 an hour in May 2025 (Bureau of Labor Statistics), and pay is about 70% of what a private employer spends on an hour of work, so the hour costs $35.07 and the job costs $31,563 a year.

The agents we build for this job read each finished job, delivery or approved timesheet as it closes, wherever your company records it. They match the customer and the item to your QuickBooks lists, check each line against the customer's price, the purchase order and every invoice already raised, and group the lines into invoices.

A line they cannot settle, a new customer, a price that differs, a job that looks billed already, goes to your billing lead with the reason beside it. The rest wait as one batch for a person to release, and once it is released the agents write the invoices into QuickBooks through Intuit's own API, ready to send.

With the agents finding, matching and checking, the billing lead reads the batch and settles what was flagged, about half a minute a line in our example: 12.5 hours a month instead of 75, or 750 hours a year handed back.

The agents' running cost is an estimated $400 a month at this volume, so each line costs $0.56 to bill instead of $1.75, the job costs $21,503 a year less, and a $10,000 build pays for itself in 5.6 months.

The hours are not the whole cost. A line typed twice bills a customer twice, and a finished job nobody types is never billed. The check against every invoice already raised is the one Intuit's own note leaves to you.

Put in your own volume and your clerk's pay to see your number.

The agents touch customer invoices, so they start narrow.

Before they connect to anything, the scope is put in writing, with where their data is processed: they read the records your invoices come from and your customer, product and invoice lists in QuickBooks, and they create invoices; they never send one, take a payment or change a posted entry.

At first nothing reaches a customer until your billing lead releases it, every line is logged with the record it came from, and you decide later which routine invoices go out on their own. The contract includes a data processing agreement signed by Qvery AI OÜ, which is registered in the EU.

Working with us on this starts with your billing clerk's month. We learn how the clerk bills today, connect the agents to your operations system and to QuickBooks, and within two weeks of having access they are preparing your invoices, with a weekly count of the hours handed back and the lines caught before they went out wrong.