I’m not necessarily interested in the traditional full budgeting and planning type stuff, but more like “AI take all these statements and tell me how to save money” purpose built tools. Anyone used anything they’d suggest?
(And to hopefully head off any unhelpful answers like I got on Reddit, I am not trying to have an AI manage my money, nor am I talking about just a wrapper for ChatGPT. AI in the broad sense of the term that can be intelligently used as part of larger programmatic workflows.)
Edit: For anyone actually trying to understand the possible applications, I found this: https://midday.ai/updates/automatic-reconciliation-engine/. The product is overkill for the sort of personal use I was asking about, but this article does a good job of showing the why and how.
Saving money doesn’t really need an AI, there really aren’t many ways to cut budgets down short of paying attention to the most obvious spending problems (Too much housing, too much vehicle, too much food (especially eating out), and too much entertainment(too many subscription platforms))
It’s usually a far better investment of time to improve your earning power (upskill, change jobs, add hours, etc.)
Cutting $40 per month from your budget by not going to the movies every week just isn’t going to have the impact you need in your life.
Where I could see an LLM being useful is categorizing entries and maybe proposing sanitization (for example when the payment provider uppercases or abbreviates stuff)
I found this which is overkill for personal use but does a good job of laying out this sort of application: https://midday.ai/updates/automatic-reconciliation-engine/
“Instead of just comparing text strings, we use 768-dimensional vector embeddings to capture the semantic meaning of transactions and receipts.
// Generate embeddings for transaction data const transactionText = prepareTransactionText({ name: transaction.name, counterpartyName: transaction.counterpartyName, merchantName: transaction.merchantName, description: transaction.description }); const embedding = await generateEmbeddings([transactionText]);These embeddings allow our system to understand that “AMZN MKTP” and “Amazon Marketplace Purchase” refer to the same thing, even though the text strings are completely different. The system learns patterns like:
- “SQ *COFFEE SHOP” → “Square Coffee Shop Receipt”
- “PAYPAL *DIGITALOCEAN” → “DigitalOcean Invoice via PayPal”
- “APL*APPLE.COM” → “Apple App Store Purchase””
Yep, that’s exactly the sort of thing I’m thinking about here. And it doesn’t even need to be full on chat style LLM, just some decent NLP that can recognize WALMART, WAL-MART, or WMART are all the same thing and label it.
But for some reason this question brings out all the assumption people who want to give financial advice or talk about the AI image the saw last year with 6 fingers.
Honestly, you don’t even need NLP for this. Excel supports regex now so you could just do a call like
=REGEXTEST(A1, "(?!)^w.*mart$"). Then just mark by type and graph out to see where your main spending is coming from.You’re missing the point, that would require sitting down and manually doing that for every conceivable payee. Walmart is just an example. The value of any sort of “intelligent” component would be for this to happen automatically and seamlessly for the user. Hell, the AI layer could just be “write regex for al the possible similar payees across these documents”.
that would require sitting down and manually doing that for every conceivable payee
That’s just called
VLOOKUP(). I think you’re over-complicating this process. If you sit down and look at your finances, you’ll notice that the number of payees you have isn’t some absurd unmanageable amount. As others have mentioned, there’s no real use case for involving AI this way. There’s no scale, no real benefit to financial tracking, etc. I get this is just to use AI for the sake of using AI, but that’s not really a goal when writing financial software.
You aren’t going to find a useful AI system for personal use in Finance, you simply don’t have the scale needed to benefit from it. You don’t make enough transactions per month that looking over it yourself is going to be any slower than reading the AI summary.
The question as with most process optimization and data analysis is, what’s the actual result you’re hoping for? If you want it to be able to summarize WALMART, WAL-MART and WMART so you can see those numbers added together, you already know you spent a lot at Walmart. Whare are you going to actually do with that information?
You need to live not as hard
Why?
I am not aware of any self-hostable AI-powered financial/budgeting software. I am using Actual budget and finding it pretty awesome tho. It has an API you might be able to do something with
All I want is software that I can put DD’s and known regulars in and it can tell me how much I’ll have in 3 or 6 months time.
I’ve tried firefly, actual budget and most of the docker based ones but still stuck using the forecast app on the ipad.
They’re all just way to over complicated for what anyone I know wants and given how AI can’t get the simplest things right, I’d rather someone just work on simplicity that works, like microsoft money didn’t before they killed it. 🙄😕
It’s not self hostable, but you mean something like this? https://calendarbudget.com/
This is an interesting and smart use case for locally run (and properly siloed) AI. Kind of an “identify any broad trends when it comes to spending habits based on this statement data, with particular focus on discretionary spending and areas where I can save money”.
Have you tinkered with any LLMs for this? If so, which ones do better than others? I could see SLMs being really good for this eventually.
I played a bit with the basic concept of identifying and categorizing merchants by importing a transaction csv into google sheets and writing a custom function that called the OpenAI API, basically just passing the raw merchant string along with “What category of business is this?”. It did well, the next step would have been to add a step that compared to a predefined list of possible categories. I didn’t compare any models or other platforms though. This was last year so I might play with it again.
