Start with one workflow that costs the most manual time. Prove value there before expanding.
Step 1: Decide what the report needs to answer
Before automating the report you already build, ask what decision it is actually meant to support. Reports built manually tend to accumulate sections that made sense once but that nobody reads anymore. Automating a bloated report just makes the bloat arrive faster.
Strip the report down to the numbers that genuinely drive a decision each week, then automate that leaner version.
A useful exercise is to ask whoever receives the report what they actually do after reading it. If there is a section they skip every single week, that is a strong signal it does not belong in the automated version either.
This step often takes longer than the technical setup that follows it, and that is fine. Getting clear on what the report should say is more valuable than getting it to arrive faster while still saying the wrong things.
number somewhere everyone can see it, not just in the head of whoever built the automation.
Step 2: Identify where each number actually lives today
For each figure in the report, trace where it comes from: which system, which field, and whether it requires any manual calculation right now. This step usually reveals inconsistencies, the same metric calculated slightly differently in two systems, that need to be resolved before automating, not after.
It is common to discover that a number everyone has trusted for years is actually calculated two different ways depending on who built the spreadsheet that week. Automating the report forces this kind of inconsistency into the open, which is uncomfortable but genuinely useful.
Document the agreed definition for each number somewhere everyone can see it, not just in the head of whoever built the automation. This single piece of documentation prevents the same disagreement from resurfacing every few months.
Working on something similar?
Let's talk →Step 3: Connect the sources and set the schedule
Working on something similar?
Let's talk →Once every number has a clear, single source, connect each system so the data can be pulled automatically on a schedule that matches when the report is actually needed. This is usually the most technical step, and it is also the one that pays back the most time once it is running.
Match the schedule to when the number is actually used, not to a convenient time of day. A report generated at midnight is only useful if someone reads it before the decision it supports needs to be made.
This is also the point to decide what the report looks like when a source system is temporarily unavailable. A good design shows a clear note that one section could not be updated rather than quietly showing last week’s number as if it were current, since a stale number presented with full confidence is worse than no number at all.
Step 4: Build in a review step before it reaches anyone’s inbox
For at least the first few cycles, have someone check the automated report against the manual version before it goes out. This catches data mapping errors early, and it builds trust in the new process before you rely on it fully.
Once the automated version has proven itself reliable over a few cycles, retire the manual process entirely rather than running both side by side indefinitely. Keeping the old process alive as a safety net past this point usually just means nobody ever fully commits to the new one, and you end up with two competing versions of the truth circulating at the same time.
Your pre-automation checklist
Before you move forward, confirm:
- You know exactly what decision the report is meant to support each week.
- Every number in the report has one clear, documented source.
- Inconsistent calculations across systems have been resolved before automating.
- The schedule matches when the report is actually needed, not just when it is convenient.
- Someone reviews the automated output against the manual version for the first cycles.
- You have a way to be alerted if a number changes by an unusual amount.
- You have a documented plan for what the report shows if a source system is unavailable.
If you want a clear next step after reading this, start with an AI readiness assessment to map where automation fits your operations.
