Most of the work that eats a week is not hard. It is repeated. The same four questions get asked on the same four days, each one costing a fresh hour of exports and reconciliation. Automating them is not about building a system. It is about writing the question down once and running it on a schedule, so the answer arrives before anyone thinks to ask for it.
The rule for choosing what to automate
Automate a ritual you already perform badly and late. Do not automate something you have never done, because you have no way to judge whether the output is right.
A ritual qualifies when three things are true: it happens on a schedule, it reads from two or more systems, and the answer changes a decision. Missing any one of the three and the effort does not pay back. A weekly report that reads one system is a lookup. A monthly report nobody acts on is decoration.
Ritual 1. The Monday cash check
Ask: "What is my cash position this morning, what payroll and bills are due in the next fourteen days, and which invoices are overdue by more than thirty days?"
Good output is three numbers and a list. Cash available now. Committed outflow inside the window. Overdue receivables with customer names. You are looking for the case where the second number is larger than the first, and you want to know that on Monday rather than on the Friday it becomes urgent.
Run it weekly, same morning. The value is in the comparison week to week, not in any single reading.
Ritual 2. Month end close
Ask: "Close last month: revenue by source, expenses by category, the five largest variances against the prior month, and anything in the payment processor that does not match the books."
Good output ends with a discrepancy list, not a summary paragraph. A close that reports revenue is a report. A close that names the four transactions the books and the processor disagree on is work getting done.
This is the single highest-value ritual in the set, because the discrepancy list is what a bookkeeper would otherwise spend two days assembling.
Ritual 3. Mid-month ad pacing review
Ask: "On day fifteen, at the current spend rate, what will each campaign spend by month end, and which ones will come in over budget?"
Good output is a table by campaign with month to date spend, daily rate, projected month end, and budget. The decision it forces is a reallocation while there is still half a month to move, instead of a post mortem in the first week of the next one.
Review it twice a month rather than monthly. A pacing check with two days left is an autopsy.
Ritual 4. The weekly retention read
Ask: "Which customers ordered in the last ninety days and not in the last thirty, and what is the total revenue at risk if none of them return?"
Good output is a count, a value and a list short enough to act on. A list of two thousand names is not actionable and tells you the segment is defined too broadly. Narrow it until the list is something a person can work through in a week.
All four rituals run from 150+ pre-built skills, so the sequence and the source list are already written. You supply the schedule and the judgment.
Keeping the answers comparable
A recurring question is only useful if the answer is measured the same way every time. That is the reason to put definitions into memory rather than restating them in each prompt: revenue net of refunds, fiscal month, which store is primary.
Log the decision, not just the answer. One line per ritual: what the number was, what you decided, and what you expected to happen. Three months of that is a better record of how the business actually runs than any dashboard built afterwards to explain it.
When a ritual stops changing a decision, retire it. A report nobody reads is a small permanent tax on attention.
Common questions
Do I need to schedule these inside the assistant?
No. The questions can be saved and run whenever you run them, which is enough for most teams: Monday morning, twice a month, and at close. What matters is the fixed shape of the question, not the mechanism that triggers it.
Which ritual should I automate first?
Month end close, if you have an accounting system and a payment processor connected. It carries the largest manual cost and produces the clearest output: a discrepancy list you can act on the same day.
How do I know the automated answer is right?
Verify the first three runs by hand against the source systems. Once the definition is in memory and the sources are stable, the number stops moving between runs for no reason, which is the signal that it is measuring what you think it is measuring.
Can these run across several stores or accounts?
Yes, and you should say which one in the question. When more than one account is connected, name it: the US store, the ad account ending 4471, the entity that closes on the fiscal calendar. Unnamed accounts are where recurring reports silently drift.
What if a source is disconnected mid-cycle?
The ritual reports the gap rather than a partial total presented as complete. Reconnect the source and re-run. A report that quietly omits a source is worse than no report.
Try CorpusIQ Free
Write the question down once and stop re-asking it
30-day free trial. Cancel anytime. All 40+ connectors included.
Start free trial →