My initial tests with ActualBudget have been successful. The import from YNAB worked very well, except for recurring and future transactions, which are not automatically carried over. However, on mobile, the account reconciliation feature is unfortunately missing. Since this feature is essential for me, I’ve decided to set ActualBudget aside for now and stick with YNAB until it is supported. While the feature works perfectly on PC, it remains unavailable on mobile, even when accessing the server via a mobile browser… quite disappointing.
1 Like
As the last discussion about budgeting apps has been 1 year ago (and I’m looking for a budgeting option).
Have there been any new budgeting/expense tracking apps that people are wanting to try? I’d also be keen to hear from people who haven’t shared their setup yet!
Using Raiffeisen inbuilt in the Efinance. Pretty good to get a quick picture.
Only issue: Credit cards are not shown as well, could be done with some manual work, but I am too lazy for that.
1 Like
Swiss banks are opening access with standard APIs: Multibanking for Private Individuals | SIX
5 Likes
Just out of curiosity - since I see a good number of the folks on this forum are doing detailed expense tracking - does anyone have a fairly “hands-off” approach to tracking monthly expenses?
I’d like a system where at the end of the month I can just dump csvs / pdfs of my transactions in a folder, and what I get are totals for different spending categories I predefined (I would probably need to review the tagging of transactions with spending categories). I don’t want anything that requires me to upload any of my financial documents to any online service.
I got the first part done fairly quickly (first time properly giving vibe coding a go) - I just dump documents from different sources in a folder and then get a list of monthly transactions in a csv. The difficult part is however the automatic tagging and summing up. I got a basic setup working fairly quickly (relies on a local ML model and an sqlite db) but it seems like I’ll have to spend significantly more time fine tuning this and fixing bugs. So before I go down that rabbit hole, just wanted to check if I’m missing some well known tool that would achieve something similar.
Easiest way may be to buy everything using your bank debit card and then you can download all expenses from your bank report and classify from there.
1 Like
Maybe check the beancount thread see if it’s something that you might like?
(you can also find some simple classifier example for beancount, tho haven’t used them yet)
Another challenge is that you can have multiple categories as part of the same transaction, i.e you can buy not only food in Migros/Coop or have different things as part of the same order from Amazon, etc.
This is most time consuming for me. I think in some cases you can potentially automate this by getting details from different APIs but this will require even more efforts and amount of corner cases is almost indefinite.
2 Likes
I think you have to choose between 2 things,
i.e. making a bigger effort
A) Upfront, i.e. at the “collection” stage
B) At the end, i.e. at the “analytics” stage
I ultimately chose the former,
and add manually every expense into Toshl.
Either immediately, or at the end of the day (I leave bank/Revo notifications active until “processed”),
just knock them into their respective categories.
With time it becomes a habit, and doesn’t take more than 1min daily (which for me is worth it at the…)
…far end is then ultra simple:
- Export the csv
- Append it into the master list in Google Sheets
- Which auto-populates my pivot there
The latter (B) entails parsing/recategorizing/splitting etc. from potentially several sources,
which I did try early on,
but chose against continuing.
Maybe with today’s rise of LLM-based “agents”,
the time will come where A+B get somewhat more automatized into C (but B part will still require our “final checks/validation” here and there).
1 Like
Independent of the tool, it depends what insights you expect from having different categories and be precise about them.
I enter all transactions manually, but very rarely split them by expense category. For example, the weekly supermarket bill goes into groceries, even though it also contains household items (which don’t have an own category) and could include presents, toys or electronics (which standalone would be elsewhere).
Only some pretty big item would be singled out, or the whole bill categorized differently.
Where I do split is recurring transactions like salaries, to not only cover incoming cash, but also gross salary, social security and pension.
My current approach is fairly efficient.
I manually classified transactions for many years, primarily to find out last few years that our spending is super consistent. So last year decided to make it more efficient and catch up on open source “AI” models and packages. Took a few hours to write a classifier on my existing transactions using Python. Did trial and error with a few models, best one at the time (google bert base uncased) gave >90% accuracy over the validation dataset. It’s “just” a text classifier so only (& obviously?) sucks on incidental larger expenses, so I manually review expenses exceeding 200 francs, which are a dozen per month and still make up >80% of all spending. I now do a full year in an hour, good enough.
4 Likes
BERT is good where you need the speed, but for your use case it might be better to use simple logic to easy to identify regular expenses and use a full fat LLM to process the long tail, sending big unknown/unconfident items to a manual review queue.
Good suggestion, also because I already had a few dozen of those rules. I don’t have the hardware to run a semi-decent model locally though, so this incurs some costs. Plus writing and optimizing a half-decent prompt with examples would also have taken time. It would work indeed slightly better on some of those incidental expenses where the supplier name reflects my categorization (e.g. bike xxx or ski zzz). Given that I learned something including from your comment 
I’m happy with the choice.
Even without a GPU, you can run Qwen3-4B quantized which should do the job. Or the 16BA1 Ling Mini(?) (I can’t remember the name). Qwen3 30BA3B is very powerful if you have enough RAM.
Since this anyway just needs to run in the background as a one off task, no problem to run it as an overnight job.