Posted by Apple Podcasts ¦ Amazon Music ¦ Spotify
Hi, and welcome back to Try AI for Growth, a podcast where I share short, sometimes surprising stories about how I use AI to tackle very real, very human problems — at work, in organisations, and in everyday life.
I’m Sara Vicente Barreto, and today I want to talk about automation. Which business owner does not think about it?
But not the shiny, futuristic kind. This episode starts with something far less exciting: a monthly accounting Excel file that already worked. It wasn’t messy. It wasn’t broken.
It was structured, reliable, and deeply embedded in how the organisation functioned.
And that’s exactly why touching it felt risky.
I must have been feeling brave at the beginning of the year, so I gave it a go and made some structural changes to the file.
Starting with Something that is Not Broken
This Excel file has been used for years. It supported multiple critical activities at once – monthly bank reconciliation, accounting classification, monthly and annual management reporting, exporting into other systems, including to our accountant to comply with our fiscal obligations. And the list goes on.
I built it over 10 years ago in Excel, as that continues to be my preferred method of operating. Whilst I have made small improvements over the years, now I know I am ready to start embedding some automation into it. On the one hand, I was personally doing a lot of manual reconciliation over the last 2 years, so I knew the pain points and the repeat points by heart. On the other hand, I now had AI to help me navigate this.
I was VERY careful in my approach. If you change something without thinking it through, the downstream impact is immense.
This matters because there’s a common perception that automation starts with chaos. That it is about fixing bad systems. But, for me, automation was meant to improve, but the system was already working. I really just wanted to save time and reduce dependencies.
A Very Small Problem
I decided to go small and presented ChatGPT with a really small, specific issue. For the purposes of receipt issuance down the line, I needed to convert numeric values into text — in Portuguese — inside an English version of Excel. I still only use my Excel in English, which, in this case, made life a bit more complex.
So my ask was, if you see the numbers 180€, can you spell out one hundred and eighty euros for me?
It’s a small thing. And an obvious one that had to work. It would save at least 30-60 minutes a month cumulatively. Sounds small, but as I said, I wanted to start small. I was a bit scared.
So I asked ChatGPT to build me a custom VBA macro. I don’t consider myself a VBA expert. Far from it. So I did not even know what it needed to build this for me. But that was not an issue.
With the help of AI, you don’t need deep technical expertise to start automating. You need a good amount of curiosity, you need to understand the problem you are trying to solve, as well as the downstream impact of mistakes. And you need time and patience.
The Real Learning Was in the Errors
ChatGPT did the macro for me in a blink. And of course, it did not work the first time perfectly. Or the second. There were technical issues:
- function names
- rounding problems
- logic errors
And then there was my favourite one — a linguistic bug. The macro was producing numbers like “onehundreds” or “threehundreds”. Which is not really how it is supposed to work. Fixing that required more than code. It required understanding how language actually works. Where conjunctions matter. What sounds right to a human reader.
And again, I did not know how to translate any of this into VBA, but ChatGPT did that for me, once I explained where the problem was coming from.
We need to remember that automation isn’t just about making things run or writing the right code. It needs to be right for the context.
The back-and-forth with ChatGPT — testing, failing, correcting — was where most of the value came from. Not the final macro itself.
You ask me – did I become a macro expert? Not at all. In the early part of my career, I did a few basic macros as an analyst in banking. I struggled as I did not understand the code behind it, and there was no accessible way (or time) to learn it. Now, I only need to know it to the extent I can understand errors and corrections.
Seeing the System Differently
Once the macro was done, something unexpected happened. I smiled in delight and emailed the whole team. Even those who don’t usually work with the file thought it was pretty cool.
As we closed our first month on a cleaned-up file and this first automation, I felt brave to venture further. So I started looking at the Excel file differently. I had always noticed patterns that showed up monthly – monthly payments, recurring donors that we recognised by memory to speed up the process, recurring costs that took too long to classify.
Again, nothing is technically wrong at the moment. But the file still requires a lot of the human brain, and truth be said, it is still easy to make mistakes, so I spend a lot of my mental energy checking it monthly and introducing new checks.
The problem wasn’t the Excel. The problem was cognitive load.
For my next round, that is what I will be addressing. For now, I am just happy the process was closed faster and more smoothly.
Innovation can come in small increments
I am not telling you a story of massive change. But sometimes we have trouble visualising the small things that can be done now to improve our workflows. A lot of small businesses depend on Excel to do their monthly reconciliations, sales inventories, cost tracking, you name it. And they work. But they take a lot of time that could otherwise be invested in the business.
This is where automation can help. Good automation that respects what already works. It doesn’t start by rebuilding everything. It doesn’t try to replace human judgment. Instead, it asks:
- How can we add memory?
- How can we support decisions without taking control away?
And you do this by addressing small steps in the process at each time. Those exposed to manufacturing know that the best way to optimise a system is to break it slowly in steps and make improvements along the way in small, low-risk increments. At the end, the benefits will add up.
What This Looks Like in Practice
This is where things become very practical — and very accessible.
Without advanced AI, without new systems, and with only simple VBA and an AI to help you put it together, Excel can still remain your tool of choice. It can have memory of past months, flag recurring transactions, suggest classifications, pre-fill decisions based on history and highlight exceptions.
Simple macros can support monthly account reconciliations, recurring cost identification, client recognition, and consistency checks over time. If a decision has been made three times, it shouldn’t need to be made a fourth time manually. That’s not automation hype. That’s operational maturity.
And it is the kind of automation that adds a layer; it does not replace the part of the system that already works. It does not replace the human; it merely assists us to be faster and get our time freed up.
What you can try
Now you may wonder how this relates to you. So pick a task you do every month that is repetitive, already works, and annoys you just enough that you notice it. Don’t start with something massive.
Then ask yourself
- If this decision was made last month, why am I making it again?
- What part of this relies purely on my memory?
- What would “good enough” automation mean here — not perfect automation?
Start there.
Lessons Learnt
What I learned through this process is that automation is not about sophistication — it’s about intention. I learned that:
- starting small lowers risk and builds confidence,
- the real value of AI is not just speed, but iteration,
- errors are not a failure of automation — they’re how understanding is built,
- and that respecting an existing system matters more than improving it quickly.
Most importantly, I learned that automation is as much about reducing cognitive load as it is about saving time.
When your system remembers, your brain doesn’t have to.
This episode started with a very small problem. But it led to a much bigger realisation. I didn’t need more control. I needed less repetition.
I didn’t need AI to make decisions for me. I needed it to remember the decisions I had already made. I didn’t need to change everything. I needed to respect what already worked.
So next time, I will explore something that naturally follows from this: pattern recognition — and how far you can take it without jumping straight into complex AI systems. But that’s a separate story; perhaps I will bring it to this space too.
For now, the takeaway is simple. Good automation doesn’t replace our work. It gives us back the time and attention to focus on what matters. And it does so by respecting what works.
Thanks for listening to Try AI for Growth. Until next time — keep experimenting and keep having fun.