Stop 8 of 9 · Daily
Staff, quality and capacity
The review sample chosen by risk rather than convenience, the capacity model built from time data, and the training note that stops an error class repeating.
What it is
Farida stopped doing the books two years ago and started being responsible for them, which is a harder job with worse tools. Rana and Tom post the work. Farida owns whether it is right, whether there is enough of it, and whether next month has more of it than the practice can carry.
Three things run daily or close to it, and none of them touch a client file. Claude reads posted work, time data and past findings. Nothing posts, reclasses or changes a bank rule.
The review sample
The point of a sample is that it is chosen before you look. Weight it toward where errors are, not where they are convenient.
From [client code] posted transactions for last month in QuickBooks Online, read only, select a review sample of forty rows weighted to: payees appearing for the first time, accounts used fewer than three times in the last twelve months, amounts over 750 dollars, anything reclassed after it was first posted, and anything posted within two days of the close date. Fill the remainder with a random draw. Give me the forty rows with date, payee, amount, account, who posted it, and the reason it was selected. Change nothing.
Every finding gets an error class, not a description. Wrong account, wrong period, missing receipt, rule applied where a rule should not exist, personal spend posted as business. Five classes cover most of it, and classes are what make three months of findings mean something.
The capacity model
Practice management gives hours by client by person. QuickBooks Online gives transaction counts. Together they say what a client costs to run, and what next month looks like before it arrives.
Here is the time export by client and person for the last six months, and the monthly transaction counts per client: [paste both exports]. Build a table with one row per client: average hours a month, hours per hundred transactions, the trend over six months, and which bookkeeper carries it. Flag every client whose hours per hundred transactions is more than double the median and say what in the data could explain it. Then project next month total hours per person against their available hours and tell me who is over.
The output is a question list, not a pricing decision. Twice the hours usually means a stale rule list, a feed that never connected, or a client who sends receipts as photographs at midnight. All three are cheaper to fix than to bill for.
Here are three months of review findings with their error classes for both bookkeepers: [paste the findings log]. Find the class with the most findings and the class costing the most rework hours. Write one page on the bigger of the two: what the error looks like, three real examples from the log with client codes only, why it happens, the check that catches it before posting, and the rule line that would prevent it. No blame, no names, and nothing longer than a page.