Finance Automation (Where To Start And What To Skip)

I automated the wrong thing first. A whole Saturday building a Power Query flow to clean a file I opened four times a year, while the reconciliation I ran every Monday morning stayed manual for another eight months.

That is the trap in finance automation. The interesting process and the expensive process are usually different processes, and nothing about sitting at your desk tells you which is which.

So I built a scoring sheet. AUTO, short for Automate Out Time-Wasting Operations, rates every process you own on four things and hands you an order to work in. Fifteen minutes, a spreadsheet, done.

Below is the framework, then the list of what to automate underneath it, grouped by where the work sits in your month. Close, forecasting, variance, payables, data prep. Everything here runs inside a working finance function, which is mine.

What finance automation covers now

For most of the last decade this meant one thing. You got software to repeat a task on a schedule so a person did not have to. Macros, Power Query, a Power Automate flow that drops a refreshed report into an inbox at 6am. That still covers the bulk of the work and it is still where I would start.

The second category is newer. AI now takes the steps nobody could write rules for, like reading a vendor file where the columns are named differently every month, or drafting the variance commentary you used to type at 9pm the night before a review. Three years ago a person had to do those, because there were no rules to encode.

The two fail in opposite ways, and that matters more than the capability does. A scheduled flow breaks loudly the moment a source file moves, so you find out immediately. An AI step fails quietly by being confident and wrong, so you find out in the meeting. Every workflow on this site that involves a model also involves a check, for that reason.

Wanting this and running it are still a long way apart. AICPA and CIMA surveyed 1,446 senior finance and accounting leaders in December 2025 and found 88% expect AI to be the biggest technology shift in the field over the next 12 to 24 months, while 8% called their organization very well prepared for it.

Most of that gap is an ordering problem. People know roughly what the tools do and cannot say which process to point them at on Monday.

The AUTO framework, and why I built it

AUTO stands for Automate Out Time-Wasting Operations. You score each process you own from 1 to 3 on four criteria, add them up, and work the highest totals first. Ten or more out of 12 means build it this month.

How many people are involved

Fewer people, higher score. A process you run alone from start to finish is a 3. One where you wait on two approvals and somebody in another timezone is a 1, because most of the elapsed time there is queueing rather than working, and automation does nothing to a queue.

How repeatable it is

Same steps every time is a 3. Month-end reporting, journal entries, the reconciliation that always pulls the same four tabs. A process that changes shape depending on what the business did that month scores a 1, and it will stay a 1 until somebody standardizes it.

How often it runs

Daily and weekly work scores 3. Monthly is a 2. Anything you touch once a quarter gets a 1 and waits. This is the criterion I ignored on that Saturday.

How many tools it crosses

If it lives inside tools you already have, that is a 3. Every extra platform, login or export step drops it. Crossing three systems scores a 1, and at that point I would rather you fix the systems than automate across them.

Step 1, score every process you own

Write down every process that lands on you, including the ones that appear in nobody’s job description. Most people in FP&A get to somewhere between 15 and 30 before they run out, and the last five are always the ones nobody remembers you do.

Then split each one into sub-tasks. Month-end close is eleven processes wearing a single name, and they score very differently once you separate them.

Score each sub-task 1 to 3 on the four criteria and total it. Here is what mine looked like the first time I ran it.

The scoring sheet I use lives in the AI Library if you want to skip the setup. Four columns and a total, nothing clever in it, and it arrives with the rest of the prompts and templates.

Step 2, add up the scores

What the totals mean

Ten to 12 means build it this month, because the payback lands inside one cycle. Seven to 9 goes on the list for the quarter. Six or below stays manual, and being honest about that group is most of the value here. A good share of stalled automation projects started life as a 5.

A real score off my own list

Budget versus actual analysis scored 11 out of 12. One person, identical every month, monthly frequency, all of it inside Excel and Power BI. It went first, and it is still the automation that gives me back the most hours in a year.

Step 3, build the highest score first

Use the tool that is already on your machine

The best automation tool is the one you do not have to ask IT for. In practice that means Power Query for reshaping data, Office Scripts or a macro for the button-clicking, and Power Automate for anything that has to happen on a schedule without you sitting there. Microsoft’s walkthrough of a scheduled cloud flow takes about ten minutes end to end.

Write the steps down before you automate them

Describe the process out loud as if you are handing it to an analyst who started on Monday. Where the data comes from, what has to be true before you begin, what done looks like. Half the time this kills two steps on its own, usually the ones that exist because of a workaround somebody built in 2021 and nobody revisited.

Test it on a month you already know

Run it against last month’s file, where you know the answer, and compare line by line. I skipped this once on a reporting process that spanned three platforms, and spent longer unpicking what it had done than the manual version would have cost me for the whole year.

Step 4, keep it running

Automation rots. Source files move, somebody renames a column, a system gets upgraded over a weekend, and your flow starts failing on the third Tuesday without telling anyone.

The ten minute monthly check

Open the three automations you lean on hardest and confirm the last run produced what it should have. Ten minutes. Then rescore anything you built more than a year ago, because a process that scored a 7 in 2024 can be a 10 now that a model can read the file for you.

What to automate, by where it lives in your month

This is what the scoring points at. Everything below is something I have built and written up, grouped by when in the month it happens, so you can jump straight to whichever one your highest score maps to.

Month-end close

The close is where the scores cluster, because everything inside it is repeatable and monthly by definition. Start with automating month-end close, then the month-end close checklist if your process has never been written down. Automating a month-end report covers the reporting half, and account reconciliation covers the part everyone likes least. For the AI-assisted version of the whole thing, I ran a full Claude Cowork month-end close start to finish and wrote up where it held and where it did not.

Forecasting and cash

A rolling cash flow forecast is the highest-value build in this group, and the 13-week cash flow forecast with AI is the faster route to the same place. Driver-based forecasting is what makes the output explainable when somebody pushes back, and a waterfall forecast is how you show the movement to a room that will not read a table.

Variance and reporting

Start with variance analysis using rate and volume, because that split sits underneath every piece of commentary you will ever write. Budget versus actual variance analysis is the wider frame around it. Once the numbers settle, automating financial reports handles delivery, and an automated budget dashboard handles the audience that would rather click than read.

Accounts payable and expenses

Automating accounts payable has the clearest payback of anything on this page if you process more than a hundred invoices a month. Automating expense management is its smaller sibling, and automating accounting processes covers the ledger work sitting around both.

Data prep and cleanup

Most automations die here rather than at the clever end. Power Query data cleaning is the workhorse, automating data cleaning for financial analysis covers the AI-assisted version, and Power Query itself is the place to begin if you have never opened the editor. For the button-clicking rather than the reshaping, Excel automation with Office Scripts is what I reach for. One catch worth knowing: Microsoft has script scheduling temporarily disabled, so build the recurrence as a Power Automate flow instead.

The calculators and the templates

Some of this needs a spreadsheet that already works rather than a flow. That is what the break-even analysis template, the net present value calculator, the internal rate of return calculation and the 3 statement model are there for.

Four mistakes that cost me time

I automated a process I had never standardized. When the steps change shape every month you end up encoding the change itself, and then you maintain that forever.

The second one spanned three systems, because the demo made it look easy. Every extra platform is another thing that can change over a weekend without telling you, and I found out about that one on a Monday.

Then I trusted a first run. Everything now gets tested against a month where I already know the answer, which is the cheapest habit on this page.

The last one I still catch myself doing, which is automating something because it is interesting. A process I run twice a year scores a 4 out of 12 and stays manual. Being willing to leave things manual is what keeps everything else maintainable.

Where to go next

If you know which process you are attacking, pick it out of the list above and go. If you are choosing a tool first, AI tools for finance has the comparisons with a verdict on each. AI in Excel covers the spreadsheet layer, Power BI covers the reporting layer, and AI agents for finance covers the part that runs while you are in a meeting.