Enabling an agent to query financial data really helps on the analysis side. How are you tackling the data ingestion side? The challenge I’ve seen again and again is logging financial data from different sources in a consistent way such that it is able to be aggregated and queried. I’ve been curious if AI can help there.
We’re tackling ingestion primarily through direct bank connections. Users connect their bank and financial accounts, and transactions flow into our system automatically. From there, we store the data in a structured database and normalize it into a consistent internal format so it can be aggregated and queried reliably.
Right now, the ingestion layer handles most of the heavy lifting—parsing the raw feeds, mapping fields into a standard schema, and ensuring consistency across institutions. Our next version will include layering AI to help with classification and enrichment (e.g. categorizing ambiguous transactions, detecting anomalies, and filling in context where the raw data is thin).
So it’s a mix: the ingestion pipeline makes the data uniform, while AI helps make it more useful and accurate for analysis. As we move toward our “agentic” roadmap, we see AI playing a bigger role in automating the messy parts of ingestion as well.
> “Traditional bookkeeping software assumes its users are trained accountants.”
That’s not the way QuickBooks community talks about their software. I’ve not been in their forums for a while, but a common refrain was to stop users from _trying_ to think like accountants. They would say it’s not the way of QB.
Fair, QuickBooks does market their software towards every small business owner not just accountants. But even so, small business owners still spend hours on entry and categorization. It's enough that most community colleges teach QuickBooks courses. We want to go further: let owners ask questions in plain language and get answers instantly, no menus or reports needed. Our upcoming versions will include agentic capabilities to send invoices, and do much more all from one AI UI.
The story of a bootstrapped AI-first bookkeeping app that lets small business owners talk to their financial data instead of wrestling with spreadsheets. Beta launching this September. Curious if HN thinks this is the future of accounting or just another shiny tool.
This and other data analysis front ends could be a fantastic application for LLMs + tool use.
It’s also a market where getting the wrong answer could result in huge liability, so at this point you’re really rolling the dice that you’re a great LLM whisperer. (There’s no such thing as an LLM engineer, at least not yet.)
I totally agree; the liability is real, which is why we don’t let the LLM “invent” numbers. We use the model as the interface, but all financial data comes from a structured database. In practice, it works like RAG: the LLM interprets the user’s question, retrieves the right data, and explains the result in plain English. That way the math is deterministic, the answers are grounded, and the AI layer just makes it accessible.
I can see that this is potentially a good sweet spot for the current state of AI. More complex and custom enterprise BI queries can get totally bollixed up in interpretation — even humans can’t agree on definitions so there’s no way to know if the query is “correct”. Perhaps in small business accounting SaaS you have the luxury of saying “this is the model, no substitutions please” and produce clearly interpretable answers.
Yeah, I’m biased since my startup is a very non-AI payroll app, but trusting my finances to an LLM sounds frightening and the money saved is not much since just hiring an accountant whose neck is on the line to get it right just isn’t that expensive.
Fair point—though to be clear, the LLM isn’t doing the math, just the interface. The numbers come from structured data, so accuracy isn’t left to chance. Where this really helps is for small business owners who are overwhelmed by QuickBooks data entry and classification. Our goal is to continually improve the experience, making bookkeeping as simple as possible.
We sidestep the “silicon psychologist” issue: the LLM simply interprets questions, while all numbers come from structured data. AI explains results, but it can’t rewrite the books.
That’s exactly the question we’re testing. Will this feel like the future of bookkeeping, or just another tool? We’re launching the beta in a week and are eager to see how real users respond. Curious what you think would make this genuinely useful vs. gimmicky?
Enabling an agent to query financial data really helps on the analysis side. How are you tackling the data ingestion side? The challenge I’ve seen again and again is logging financial data from different sources in a consistent way such that it is able to be aggregated and queried. I’ve been curious if AI can help there.
We’re tackling ingestion primarily through direct bank connections. Users connect their bank and financial accounts, and transactions flow into our system automatically. From there, we store the data in a structured database and normalize it into a consistent internal format so it can be aggregated and queried reliably.
Right now, the ingestion layer handles most of the heavy lifting—parsing the raw feeds, mapping fields into a standard schema, and ensuring consistency across institutions. Our next version will include layering AI to help with classification and enrichment (e.g. categorizing ambiguous transactions, detecting anomalies, and filling in context where the raw data is thin).
So it’s a mix: the ingestion pipeline makes the data uniform, while AI helps make it more useful and accurate for analysis. As we move toward our “agentic” roadmap, we see AI playing a bigger role in automating the messy parts of ingestion as well.
Well. Big ad.
> “Traditional bookkeeping software assumes its users are trained accountants.”
That’s not the way QuickBooks community talks about their software. I’ve not been in their forums for a while, but a common refrain was to stop users from _trying_ to think like accountants. They would say it’s not the way of QB.
Fair, QuickBooks does market their software towards every small business owner not just accountants. But even so, small business owners still spend hours on entry and categorization. It's enough that most community colleges teach QuickBooks courses. We want to go further: let owners ask questions in plain language and get answers instantly, no menus or reports needed. Our upcoming versions will include agentic capabilities to send invoices, and do much more all from one AI UI.
The story of a bootstrapped AI-first bookkeeping app that lets small business owners talk to their financial data instead of wrestling with spreadsheets. Beta launching this September. Curious if HN thinks this is the future of accounting or just another shiny tool.
This and other data analysis front ends could be a fantastic application for LLMs + tool use.
It’s also a market where getting the wrong answer could result in huge liability, so at this point you’re really rolling the dice that you’re a great LLM whisperer. (There’s no such thing as an LLM engineer, at least not yet.)
I totally agree; the liability is real, which is why we don’t let the LLM “invent” numbers. We use the model as the interface, but all financial data comes from a structured database. In practice, it works like RAG: the LLM interprets the user’s question, retrieves the right data, and explains the result in plain English. That way the math is deterministic, the answers are grounded, and the AI layer just makes it accessible.
I can see that this is potentially a good sweet spot for the current state of AI. More complex and custom enterprise BI queries can get totally bollixed up in interpretation — even humans can’t agree on definitions so there’s no way to know if the query is “correct”. Perhaps in small business accounting SaaS you have the luxury of saying “this is the model, no substitutions please” and produce clearly interpretable answers.
Yeah, I’m biased since my startup is a very non-AI payroll app, but trusting my finances to an LLM sounds frightening and the money saved is not much since just hiring an accountant whose neck is on the line to get it right just isn’t that expensive.
Fair point—though to be clear, the LLM isn’t doing the math, just the interface. The numbers come from structured data, so accuracy isn’t left to chance. Where this really helps is for small business owners who are overwhelmed by QuickBooks data entry and classification. Our goal is to continually improve the experience, making bookkeeping as simple as possible.
LLM engineer -> silicon psychologist who can sometimes sell the beast into making the year-end postings pass all tests?
We sidestep the “silicon psychologist” issue: the LLM simply interprets questions, while all numbers come from structured data. AI explains results, but it can’t rewrite the books.
Sounds like a powerful model if you can get it right
That’s exactly the question we’re testing. Will this feel like the future of bookkeeping, or just another tool? We’re launching the beta in a week and are eager to see how real users respond. Curious what you think would make this genuinely useful vs. gimmicky?
Solid AI use case. As someone that is trying to tack on AI/automation to legacy accounting software, this might be the better way to do it.
We thought about just building an AI plugin for QuickBooks, but decided to build our own platform instead. Harder path, but bigger upside.
What could possibly go wrong with letting an LLM decide how to record transactions? Someone who sold a new car for a dollar might have an idea...