The Automation Autopsy: The $45,000 Excel Spreadsheet Holding Your Operations Together

Every business has one. A weekly report that three people build by hand, from five different systems, every single Monday.

It works. Mostly. Until the one person who knows how it fits together takes two weeks off, and the whole operation discovers it has been running on a single Excel file and a lot of hope.

This is the first Lumeio Automation Autopsy, where we take one painful process, lay it on the table and work out what killed it. Today’s patient is that report. If you want to automate Excel reports, this is the place to start, because the answer is rarely “more spreadsheet”.

[Puts on sunglasses.] I have watched more crime shows than is healthy for someone who works mostly in spreadsheets. You’ve been warned.

What does a manual Excel report really cost?

That’s an indicative number, not a bill. Your own number depends on your hours and your wages.

The formula is simple enough to run on a napkin:

Fourteen hours a week is 37% of a full-time person. Spent on copying, pasting and making headings bold.

The symptoms: how do you know a spreadsheet is holding your operations together?

Our patient is the Monday Ops Pack. It’s a six-tab workbook that tells the managers what got done last week, what it cost and what’s late.

Here’s what the patient presented with:

• One builder. Priya in operations knows the steps. Nobody else has run it start to finish.
• The yellow cell. Tab three has a cell highlighted yellow that says “DO NOT TOUCH”. Nobody remembers why. Everybody obeys.
• Version fog. There are “Ops Pack v3”, “Ops Pack v3 FINAL” and “Ops Pack v3 FINAL (Dave’s edits)”, all in the same folder.
• Monday dread. The report is due at 10am, and is ready at about 2pm.
• Untraceable numbers. When a manager asks “where does this figure come from?”, the honest answer is “a VLOOKUP, probably”.
• Five exports. Every week, by hand, from five different systems.

If you read that list and said “Oh no”, you’re in the right place. [Gently pats your shoulder.]

The crime scene: what the weekly report workflow actually looks like

[Adjusts sunglasses. Stares into the middle distance.] Let’s walk the scene.

Every Monday, the Ops Pack gets built like this:

1. Five exportsJob system CSV, accounting system Excel file, supplier portal download, a shared inbox of supplier emails and PDF delivery notes, and a timesheet CSV.Moving data
2. Copy and pastePriya pastes all five into the master workbook.Moving data
3. FormattingFix headings, delete columns, redo lookups, rebuild pivots.Making it look like last week
4. ReconciliationFinance checks the totals against the accounting system.Checking
5. The mismatch loopEmail Priya. Wait. Fix. Re-check.Waiting
6. ReviewThe operations manager reads it, asks questions and sends corrections.The only decision
7. DistributionThe final version is emailed to six managers as an attachment.Moving data

Look closely at the middle of that diagram. Nobody in it is making a decision. They’re moving data from one place to another and making it look the same as last week.

The only genuinely human moment is near the end, when the operations manager reads the numbers and asks “why is this one weird?” Everything before that is a person doing a job a machine would do faster, and without needing coffee.

The human tax: how much does a 14-hour weekly report cost?

Here’s the working, so you can check it.

The cost of one full-time person. The ABS puts full-time adult average weekly ordinary time earnings at $2,083.70 for May 2026. Over 52 weeks that’s $108,352. Add the 12% super guarantee that has applied since 1 July 2025 and you get about $121,355.

The hours. These are the scenario’s assumptions, per week:

• Ops coordinator, pulling five exports and pasting them in: 3 hours
• Ops coordinator, formatting, lookups and pivots: 4 hours
• Finance officer, reconciling totals and chasing gaps: 2.5 hours
• Operations manager, reviewing, querying and sending corrections: 2 hours
• Anyone, fixing what broke (a moved column, a broken link, the wrong version): 2.5 hours

$44,700That’s 14 hours. 14 ÷ 38 × $121,355 comes to about $44,700 a year, for one weekly report.

Two caveats pull in opposite directions. The ABS figure is an all-industry average, so it probably overstates an admin wage.

The formula also leaves out payroll tax, WorkCover and the cost of a decision made on a wrong number, so it understates the true cost.

If your hours are different, here’s how the number moves:

We’ve written separately about the hidden cost of manual data entry, including a method for timing your own process properly.

Why do manual spreadsheets break?

This isn’t a criticism of whoever builds your report. Humans make small errors at a steady rate, and spreadsheets don’t complain when we do.

86%Of spreadsheets examined in the more rigorous field audits contained errors, according to Raymond Panko’s review of spreadsheet error research, presented to the European Spreadsheet Risks Interest Group in 2000.

The research is old but blunt. The same paper notes that the people who build spreadsheets tend to be overconfident about their accuracy.

Big organisations aren’t immune either. JPMorgan’s own 2013 task force report into its 2012 trading losses listed, among several operational issues, that one risk model operated through Excel spreadsheets completed manually, by copying and pasting data from one spreadsheet to another.

It wasn’t the whole story of those losses. It’s still a striking thing to find in a major bank’s post-mortem.

Then there’s the risk nobody puts in a budget. If Priya resigns, the Ops Pack goes with her, because the real instructions live in her head. The yellow cell is just the part you can see.

How do you automate Excel reports?

This is the “What we’d automate” part of the autopsy. Almost all of it is ordinary, rule-based automation. No AI required.

1. Map it, then shrink itList every tab, column and chart, and ask who actually reads each one. Reports collect clutter like a kitchen drawer collects takeaway menus. Delete before you automate.People
2. Replace the exportsPull data straight from each system through its API, a database connection or a scheduled export to a shared folder. No more downloading CSVs by hand.Automation
3. Move the formatting into rulesColumn clean-up, lookups and joins become a defined set of steps that runs the same way every time. Nobody re-deletes the same three columns on Monday.Rules
4. Add the checks a human used to doRow counts, totals that must match the accounting system, and missing jobs get flagged automatically. Finance only sees the mismatches.Rules
5. Schedule itThe report refreshes overnight and lands in inboxes, or on a dashboard, before anyone’s first coffee.Automation
6. Keep a review gateA person reads it before it’s treated as the truth. More on that below.Human

You may already own the first step up. Excel’s built-in Power Query saves your clean-up steps as a query you can refresh whenever new data arrives. It can also combine files that share the same structure from a folder or SharePoint into a single table.

For a report like the Ops Pack, Power Query alone can remove most of the copy and paste. Its limit is that someone still has to open the file and press Refresh, unless the refresh is scheduled in another product such as Power BI.

The checks, the alerts and the inbox source usually need a proper workflow around it.

That’s the point where a designed pipeline earns its keep: source connections, a transformation step, validation rules, a scheduled run and a delivery step. It’s also the part of the job Lumeio does day to day.

Where does AI actually help?

Four of the five sources are tidy exports. They don’t need AI. They need a connection and a rule.

The shared inbox is different. Supplier emails arrive in every shape imaginable. Some have the delivery date in the subject line, some in a PDF, and one supplier apparently types it in the signature.

That’s where AI is the right tool:

• Extraction. Reading supplier, date, quantity and job reference out of messy emails and PDFs.
• Classification. Sorting delivery notes from invoices from “just checking in” emails.
• First-draft commentary. Writing a plain-English summary of what changed this week, for the manager to edit, not to publish unread.

Everything AI extracts still goes through the same validation rules as the tidy data. If it can’t match a delivery to a real job, it gets flagged for a person.

If the same inbox also carries supplier invoices, we’ve covered how to stop copying invoices from Outlook to Xero separately.

What should stay human?

In the Ops Pack, the most valuable hour of the week is the operations manager asking “why is this job 30% over?” That hour should get longer, not shorter. Automation just stops it starting at 2pm.

This is what human-in-the-loop review means in practice. The machine does the gathering, and a person stays accountable for what it says.

Before vs after: the Monday Ops Pack

The estimated opportunity

In this scenario, automation could return roughly 11 to 12 hours a week. At the same rate used above, that’s an indicative value of about $35,000 to $38,000 a year in staff time.

That’s an estimate based on the scenario’s assumptions, not a promise. It doesn’t include the cost of building and running the automation, and actual results depend on how clean your sources are and how complex the report really is.

The bigger gains are harder to put a number on.

The report is ready on time. The numbers can be traced. And the business no longer stops when one person takes a holiday.

Common mistakes when automating a spreadsheet

The most common mistakes are honest ones. Most come from moving too fast on the build and too slow on the thinking.

Automating the mess exactly as it is. If six tabs exist because someone asked for them in 2019, automating all six just makes the clutter faster. Map it first, then cut.

Building a bigger spreadsheet. Adding macros and more formulas to the same workbook moves the risk around. It doesn’t remove it, and it usually makes the yellow cell harder to understand.

Losing the builder’s knowledge. The person who runs the report knows the exceptions: the supplier who always sends late, the job code that changed last year. Sit with them before you build anything.

Removing the review. A report that refreshes itself at 3am can also be wrong at 3am, with nobody watching. Keep a person on sign-off.

Automating something tiny. If the report takes 30 minutes a month and nobody suffers, leave it alone. That’s a perfectly good answer.

The lesson (and our recommendation)

[Takes off sunglasses, for dramatic effect.] The patient didn’t die of Excel. It died of being the only copy of a process nobody had written down.

Excel is a good tool. The problem is using it as the plumbing between five systems, with a person carrying the data by hand every Monday.

Our recommendation:

• Map the report and delete what nobody reads.
• Automate the movement and formatting with ordinary rules first.
• Use AI only for the messy source, and check its output like any other data.
• Keep a human on review, with exceptions flagged rather than hidden.

If your business has a report like the Ops Pack, it probably has three more. We see this pattern across the industries we work with, from construction to accounting.

Frequently asked questions