Your team is excited about AI and automation. They've attended the webinars, read the articles, and now they come to you with ideas:
Every idea sounds good (in theory). But you have limited time, and you don't want this to be another failed project with the board asking you this: "Where's the ROI?" So how do you separate automation that produces results, from automation that goes nowhere? Today I'm going to show you a simple framework that will help you categorize any automation idea. As a result, you'll know exactly what to do to see the best ROI from the time that you invest.
|
|
|
The problem you have with automation is that everything feels 'automatable'.
But there's a math problem that stops most AI & automation projects.
Task A: Monthly bank reconciliation review. Time spent is 20 minutes per month, which is 4 hours per year. Automation effort is maybe 40 hours to build plus 5 hours each quarter to maintain. Break-even? Never.
Task B: Rolling forecast consolidation. Time spent is 3 hours per month, which is 36 hours per year. Automation effort is 5 hours to build plus 30 minutes per quarter to maintain. Break-even is 2 months (plus the satisfaction to let your team work on more valuable tasks).
Without a way to distinguish Task A from Task B, your team wastes weeks automating the wrong things.
And I know this because I've done it myself.
I found myself over-automating things just because I liked building the automation more than having to do the repetitive work itself!
I'd start a project thinking it would be quick, and then halfway through I'd realize it was more complicated than I thought.
In reality, the task I was automating only cost me 20 minutes per month. But automating it would take me 3 days.
Plus, don't forget, processes change over time. Which causes a big problem if your automation breaks before you've broken even on your investment!
The 3-Category Framework
So instead of asking "Can we automate this?", start asking "Which category does this belong in?"
Because there's an important difference between automation, AI workflows, and AI agents.
|
|
And more importantly, each one has very different odds of paying off for you.
By the way, if you want a deep dive on what each category is – Automation vs AI Workflow vs AI Agent, read my previous newsletter.
Today – instead of looking at what they are – we're focused on when and where you should use them.
Category 1: Pure Automation (No AI Needed)
This is for tasks that follow the exact same steps every single time, with no judgment or interpretation required.
Examples:
- Downloading your trial balance from NetSuite or QuickBooks at month-end.
- Moving invoice attachments from your inbox into the right folder based on the sender.
- Consolidating all rolling forecast inputs into a master spreadsheet and mapping it to the financials.
- Consolidating five department budget files into one master spreadsheet.
- Sending a reminder email to your team three days before close.
When to use it vs. when to skip it:
The key question to ask yourself is whether you could write down the exact steps and have someone follow them blindly without making any decisions. If so, this is Category 1.
Category 2: AI Workflows (Human-Triggered)
This is for tasks that require some judgment, but that judgment follows a consistent structure. You trigger it, AI does one specific task, and then you review the output.
Examples:
- Generating a first draft of your variance commentary based on the budget vs. actuals numbers.
- Categorizing hundreds of expense transactions based on messy free-text descriptions.
- Summarizing last month's board meeting notes into a list of action items.
- Drafting the narrative sections of your board deck from the financial data.
- Flagging unusual transactions that fall outside your normal patterns.
When to use it vs. when to skip it:
The key question to ask yourself is whether you'd be comfortable having a junior team member draft this, knowing you'll review it before it goes anywhere. If so, this is Category 2.
Category 3: AI Agents (Skip This For Now)
This is when you give AI a goal and it figures out everything on its own, choosing its own tools, doing its own research, and even writing its own code.
Examples of what agents promise:
- "Handle the full monthly close and flag anything unusual."
- "Research competitor pricing and update our model accordingly."
- "Prepare the board deck end-to-end, including pulling the data."
When to use it vs. when to skip it:
The honest answer is that almost no finance teams are running real AI agents successfully in production yet. Most things being called "agents" are just Category 2 workflows with better marketing.
The only thing I've found that comes close to being an AI Agent, (connected to your data to help with month-end close and reporting) is Concourse. You can see it here.
For now, focus on Categories 1 and 2. They're ready today, they deliver predictable results, and they'll give you the skills to evaluate agents when they mature.
The Tools You Need
|
|
Category 1 (pure automation)
You don't need anything fancy. If you're in Google Sheets, use Google Apps Script. If you're in Excel, use VBA or Office Scripts (for Excel Online). For automations that connect multiple apps together (like pulling data from your email into a spreadsheet), tools like Make, Zapier, or Power Automate will do the job.
Category 2 (AI workflows)
Custom GPTs are the easiest starting point because you can build one in minutes with no code. If you want something that runs automatically on a schedule or connects to multiple systems, n8n is a good option (but there is a bit of a learning curve) so Zapier’s AI Agents (not really agents) might be an easier options. And for one-off tasks where you need AI to help you think through messy data, Claude or ChatGPT work well.
The best part? You already have access to most of these. They're easy to use, and AI will help you figure out how to set them up if you get stuck.
The Current State of Agentic Tools
I told you to skip agents for now. But that doesn't mean nothing is happening.
The honest picture as of January 2026 is that agents are arriving, but they're still not ready to work across all of your activities and apps.
They work inside one app, one folder, or one enterprise system.
Plus, they're really still not yet “Autonomous teammates who can run your monthly close while you sleep.”
But as I said earlier, I was part of the launch of a really cool Agentic Finance tool for month-end called Concourse yesterday. You can watch the launch video here, and join me for our masterclass here.
What else is worth watching?
Microsoft has rolled out Agent Mode in Excel which I cover in my YouTube video here, which lets you describe what you want and have Excel build it for you.
It's super useful, but it only works in the currently open file and can't access other apps or other workbooks. So it's "agentic inside Excel," not "agentic across your finance tools".
Anthropic also just released Cowork, a desktop tool where you give Claude access to a folder and a goal, and it works through the task whilst checking in with you.
It's closer to being able to fully delegate junior analyst type work, but it's still in research preview and comes with some risks (always check the usage info).
What still stands from what I’ve previously said:
If something calls itself an "agent" but can't take actions and run multi-step processes across apps by itself, it's probably still just a workflow with better marketing.
Experiment, but don't bet your month-end close on them yet.
The One Thing To Remember
Not everything your team requests requires automation.
So, next quarter when someone comes to you with an idea about something to automate, don’t say "let me think about it" or "sure, give it a try."
Ask: “What’s the payback?”
They'll either have the answers (which means it's probably a good project) or realize they haven't thought it through yet.
So take your next 5 automation candidates and compare them to the framework.
You might find that 3 of them aren't worth doing at all. And that's a good thing, because now you know where to focus your time.
Don't take on projects that break.
Break even (fast) and brag about it instead 😉
Best,
Your AI Finance Expert,
Nicolas
P.S – Did you enjoy this? Reply and let me know (I read all replies)
P.P.S – If you want to see more tricks and tips on using AI for Finance, check out one of my latest videos where I covered 100 SECRET tips on AI for FINANCE / AI for CFO. In it you’ll not only learn about automation but also prompting and the importance of data confidentiality.