LearnAI in Business Operations

Monthly review: 5 numbers to check

AI in Business Operations2026-09-01

In the previous lesson you drew the line between what your system runs alone and what waits for your approval. After a month of live use you own something more valuable than the settings themselves: a log. Every message that arrived, every reply that went out, every time the system stopped and asked. A business that reads that log once a month keeps improving. One that ignores it repeats the same argument every time something goes wrong. A review is not a hunt for someone to blame, but for a pattern you can turn into an improvement.

A real-world example

A car parts shop takes branch orders over WhatsApp. A branch keeper sends: we need twenty oil filters for this model, and the system replies with availability and price, then records the order. After the first month the owner sat down for half an hour with the log. He did not read the messages one by one. He looked at five numbers only.

What he found surprised him: a third of all approval stops were the same case repeated, a request for a quantity larger than the branch had on hand. That was not a rare exception deserving his decision each time. It was a missing rule. The system should have replied with the available quantity and offered a transfer from the nearest branch instead of freezing the order and waiting for him. One rule change removed dozens of messages from his phone the following month, and made branches faster at the same time.

Note where the problem was: not in the system and not in the staff, but in a rule written before anyone knew how the work actually flows. You cannot spot that from a single message. You spot it by looking at a whole month.

Five numbers are enough

You do not need a complicated dashboard. Five numbers pulled from the month's log tell you where you stand:

NumberWhat it meansWhat to do with it
Requests handled automaticallyThe workload lifted off your staffCompare with last month: a rise means the team trusts it
Stops for approvalHow often the system needed your decisionIf one type repeats, that is a missing rule, not an exception
Requests corrected after executionThe errors that actually slipped throughRead every case yourself; this is the most valuable number
Messages the system did not understandThe gap between field language and your instructionsAdd the wordings your team really uses to the instructions
Your average approval response timeWhether you became the bottleneckIf it is slow, widen the limits or name a backup approver

An orderly half hour

Make the review a fixed appointment at month end, and follow this order:

  1. Pull the five numbers before reading any message, so one dramatic case does not pull you from the wider picture.
  2. Read every request that was corrected after execution and ask: could the system have known? If yes, the rule is at fault. If no, the information it received was.
  3. Sort the stops into two piles: a genuine rare exception that deserves your decision, and a repeated case that deserves a written rule.
  4. Pick two changes only for this month. More than that and you cannot tell which change helped and which hurt.
  5. Write in your procedures file what you changed, the date, and who decided it, so the team has a written reference rather than one person's memory.

The number you cannot postpone is the third one. An error that ran and was never corrected shows up in no statistic, because nobody complained about it. That is why you read those cases yourself instead of delegating them.

Three signs the problem is bigger

Sometimes a rule tweak is not enough. Watch for three signs: misunderstood messages rising month after month despite better instructions, a correction count that never drops, and staff quietly returning to the old manual way even though the automated step works. The third is the most dangerous because it never appears in the log at all: someone who went back to the notebook leaves no trace in the system. So ask your team directly once a month: did you work around the system this month, and why?

Once the numbers settle and corrections fall, you are ready to widen the scope: a new automated step, or connecting orders to stock and invoicing in one flow as in running your business from WhatsApp and WhatsApp inventory management.

In short: review one month's log in half an hour using five numbers: automatic runs, stops, corrections, misunderstood messages, and your own response time. A repeated stop of one type is a missing rule, not an exception. Read the correction cases yourself, then change two things only and write them down in your procedures file.