A PR agency sells hours against a fixed fee, which means the margin on an account is invisible until someone adds it up — usually at month end, when the month is already spent. Your time tracker and your books already hold every number that calculation needs. This guide covers how to connect them to Claude or ChatGPT so the retainer math, the invoicing and the chasing arrive drafted.

It covers the money side of our PR agency automation guide. Media lists and pitch follow-up are in automating your PR journalist database; account team workload is in PR agency staff automation.

Connecting AI to your PR agency billing software (QuickBooks and Harvest)

QuickBooks connects to Claude directly through its Small Business package, which also ships an invoice chaser and a month-end prep routine. Time trackers like Harvest and Toggl expose an API, and most agency management platforms will at least export a timesheet on a schedule.

The rule holds here too: the AI reads your records and drafts the action. It does not post entries into your books, and it does not send an invoice or a payment reminder to a client until someone approves it or a job has earned the right to send on its own.

Keeping that line matters most in billing, because an invoice is a client communication. A number sent before a partner has looked at it is a conversation you then have to walk back.

Using AI to track hours against each retainer

You know your retainer totals. Knowing on the fifteenth which accounts are already underwater for the month is a different question, and answering it is arithmetic rather than insight — which makes it the first thing worth handing over.

The mid-month position. Hours logged against each account's scoped total, as a percentage, with the trend from the previous two months beside it. Delivered on a schedule rather than requested, because the useful moment is while there is still a month left to adjust.

Where the hours actually went. Logged time grouped by the work it bought — media relations, content, reporting, client meetings. An account at eighty percent of its hours with most of them in meetings and reporting is a different problem from one at eighty percent with most of them in pitching.

Unbilled and uncategorized time. Entries missing an account code, hours logged against a closed project, and timesheets not filed, surfaced as one short list each week rather than discovered during invoicing.

Using AI to spot overservicing before the month closes

Overservicing is the structural problem in retainer work. It does not arrive as a decision; it accumulates as small, reasonable yeses — one more round of edits, one more call, one more angle for a launch that has already shipped.

Read against the scope you agreed, those yeses are visible while they are still small:

• Scope drift by account. Work being delivered that does not map to anything in the signed scope, named specifically, with the hours attached.
• The accounts trending the wrong way. Where this month's burn rate is running meaningfully above the last two, flagged in week two rather than confirmed in week five.
• Pass-through costs being absorbed. Wire distribution, monitoring subscriptions and clipping services charged to an account but never rebilled, listed before the month closes.
• The scope conversation, prepared. When an account genuinely needs more than it is paying for, the note that opens that conversation comes with the hours, the extra work and the period it covers already laid out.

That last one is the point. The useful output is not a warning; it is a well-evidenced case for a scope increase, drafted while the evidence is fresh. If your retainer margins are a month-end surprise, Unprompted can build this against the books and tracker you already run.

Using AI for invoicing and aging balances

Month-end is the same set of steps every month, which makes it a good candidate for a job that runs on a schedule.

What is ready to invoice, by account, with the hours, the pass-through costs and anything out of scope itemized for a partner to check. Retainer invoices prepared on the dates they are due. Expenses matched to the accounts that incurred them rather than left in a general bucket.

Then the collection side, which agencies are reliably worse at than the work. Aging balances by account and by age, each with a reminder drafted that firms up on the schedule you set — polite at two weeks, direct at sixty days. A balance chased at thirty days is a different conversation from the same balance at ninety, and the thirty-day version is the one that preserves the relationship.

Month-end also answers the question owners actually care about: which accounts carry the agency, which break even, and which cost more in hours than they return.

Why you should layer AI on your existing accounting software

QuickBooks stays your book of record, and your time tracker stays where hours are logged. Neither is replaced and nothing is migrated.

What sits on top is the reading layer: the mid-month position, the drift flag, the invoice assembled and the reminder drafted. The decisions stay where they belong — whether to raise a fee, whether to absorb a month of extra work to protect a relationship, whether an account is worth keeping at the rate it pays. Those are judgments about the business, and no amount of reading your records produces them.

If retainer margins are something you find out about after the month is over, talk to us and we'll show you the mid-month position built from your own numbers.