What a firm can charge when the close no longer takes a week.

If the close took a week because a person coded the feed, you were selling categorization. When the close takes a sitting because Claude drafted it and a person approved it, that price is the wrong number.

I run AG Accounting. I built Numbers Game on our own book because we were tired of pricing the first part. Median close on a 200-transaction month at AG Accounting is about 18 minutes. That is the proving ground, not a promise your book will match it. We did not become an AI bookkeeping service. We stopped pretending the first part was the product.

Sell a monthly close, a narrative the client will actually read, and an advisory conversation on top, at the same headcount. Package it as a retainer with a stoplight report, not as hours and not as “we categorized your bank feed.” Expect a new per-connection software cost while the team is still learning the prompts, and do not hide it. Volume discounts exist. One-off custom dashboards do not, unless they belong in the product. Solo accountants price three to seven files. Fractional CFOs sell audit-before-you-model, 13-week cash, budget-to-actual, WIP by project, and a board or stoplight pack, and they have to decide who pays for the connector when the client already has Ramp, Brex, and QBO. The report is the product. Assembly is cheap. The sentence that explains the number is what you invoice.

Read branded client reporting for what the artifact is. What follows is what to charge for it.

Bookkeeping firms spent a decade getting paid to type, and throwing the P&L in as a courtesy. Clients learned to shop that work on price. Software then made the typing cheaper. Firms who kept the old invoice found themselves in a race with every other firm who also bought a categorizer. That race does not have a winner you want to be.

The move is not “hire the AI so we can take on more $300-a-month files.” The move is: the close is finished, the exceptions are listed, the package is in the folder, and you still have time this month to say why gross margin moved. That sentence is billable. The feed coding is not, not at the rate you wish it were.

Published plans, because this math needs the actual line item: Starter $149 per month for 3 connections, Growth $299 for 7, Scale $799 for 30, Enterprise custom. Seats are not the unit. Connections are. QBO and Xero both count. You bring Claude or ChatGPT; there is no second token bill from us. Those numbers are on the homepage. They are a firm cost, not a client SKU, unless you choose to pass them through.

Name them whatever you already name them. The content has to change.

Close (stoplight) For the cash-basis and modified-cash book. Books closed on a published calendar. P&L, balance sheet, cash flow, from the live ledger, branded. A stoplight: green if no flags, yellow if there is a watch item, red if there is a conversation (negative balance-sheet account, three months of losses, debt spike, owner draw). One page of exceptions, or a sentence that says there were none under the written tests.

You are not selling “AI.” You are selling a close that arrives on time with a color the owner can see on a phone.

Close and narrative Adds the letter. Flux on the lines that moved. Vendor or class commentary where that client has been trained to look. This is the CAS default.Client reporting is the artifact. Your voice is the product. Claude can draft. You still sign.

Close, narrative, and advisory The retainer that includes a working session. Fractional-CFO shape, even when you are a CPA firm. You do not show up to recode Stripe. You show up because the stoplight was red, or because the board pack is next week.

Do not bury advisory inside a cheap close to win the file. You will never climb out. If they only want the cheap close, sell them that, and do not throw in a strategy call “because we have time now.” Time is the inventory you just created. Price it.

Nonprofit and board clients need the same three rungs, with 990/ASC 958 language in the narrative, not a fourth product. Construction and job-costing clients need WIP and budget-to-actual in the pack, not a custom app. If you cannot standardize the pack, you do not have a package. You have a project.

Be honest with the partner table. This part is uncomfortable, and skipping it is how you get a surprised managing partner in month two.

A firm that puts 60 companies on Scale-class volume is looking at a software line they did not have last year, on the order of a low five figures annually at list, while the team is still learning the prompts. They are not at full value on day ten. That is not a reason to stay on a hosted bookkeeper who will not let you add your own clients. It is a reason to roll out in waves (30, then 60, then the rest), as in the rollout post. Do not buy 170 connections to pre-commit a discount on prompts you have not written.

Treat prompt work as implementation, not as proof the product “doesn’t work yet.” Close time tracks prompt quality. Fifteen minutes and forty minutes are both normal in month one.

Decide, in writing, whether the connection cost sits in firm overhead or on the invoice. Clients already juggle Ramp, Brex, Bill, and QBO. Adding a mystery “AI fee” they do not control is a hard sell. Many firms keep it internal and raise the retainer once the package is visibly better. Some allocate a small line per company. Both are coherent. Surprising the client is not.

Ask for volume and a mutual exit if you are making a big commitment. Do not ask for a one-off dashboard “just for us.” If the dashboard is a who’s-done board every CPA firm needs, it should be product. Firms have told us that directly.

Starter and Growth exist because a seven-file book should not buy thirty seats of ambition. Solos: start there.

The offer is not “I will keep your books cheaper.” The offer is: the day I connect, I audit the file. Then I model. I will not forecast on a P&L I have not tested.

Audit the books before you model. Connect QBO or Xero. Run the failure list: unexplained equity, basis mismatches, rec plugs, trailing numbers that do not match last month’s issued pack, unidentified card payoffs. A person still approves any write. You often did not keep these books. You still should not let a model mutate a file you just agreed to trust. Charge a kickoff audit, even if the monthly retainer comes later. Cleanup that used to eat the first two billed weeks is now a scoped first deliverable.

13-week cash on a closed file. A cash forecast is a sellable artifact, not a spreadsheet hobby. It only holds if the close underneath it is real. If you skip the audit, you are modeling the plug. Put “13-week cash forecast from the closed ledger” on the invoice, not “financial model.”

Budget-to-actual, and WIP by project. Construction and job-costing clients do not need a custom app. They need WIP by project and budget-to-actual in the pack you already send. If you cannot standardize that pack, you do not have a package. You have a project.

Board pack and stoplight. Fractional CFO client reporting is a color the owner can see on a phone, plus the two sentences you would say in the room. Client reporting is the artifact. Claude can draft. You still walk into the board meeting.

Who pays when the client already has Ramp, Brex, and QBO. The client already bought the stack. They will not buy “AI access to QuickBooks” on top of it. They will buy the audit, the 13-week, the board pack, and the session. Decide, in writing, whether Numbers Game sits in your overhead or on their invoice as an allocated close-connector line.

Many fractional CFOs keep it internal. Growth at $299 for 7 connections, or Starter at $149 for 3, is a practice cost. Scale at $799 for 30 covers most of an 8-to-40 file book. Pass-through only if you can explain it in one sentence and take it off if you disconnect the file. Do not hide a new line inside “technology” next to their existing QBO bill.

Nicole Jordan-Dahdal at PrismEdge is the published version of retrieval ceasing to be the job: a three-entity opex audit that used to take days, done in under two hours. Read that study. Your invoice should describe the audit and the advisory, not “AI access to QuickBooks.”

If the client already has a bookkeeper, you are not there to recode the feed. Price the session. Do not donate it because the retrieval got fast. A concentrated book of eight to forty files is a full practice. Do not price it like a $300 categorizer with a nicer title.