All respect to these guys, but the reason Accounting AI might be slow is that they have a fundamental misunderstanding of accounting. "It's all about numbers, unlike law." Very false; the majority of accounting work is classifying transactions or assets based on unclear government guidance paired against unique business operations. "The numbers have to be 100% correct 100% of the time." Also false; they need to be correct within the limit of "materiality", which might actually work in favor of LLM's; if they hallucinate every ten-thousandth transaction, that would still be better than human systems, and within materiality limits.
"The majority of accounting work is classifying transactions or assets based on unclear government guidance paired against unique business operations"-Thanks Jared for this enlightening words!
I also disagree with the claim that accounting is about numbers. It is about exercising professional judgements to determine how to classify those numbers. Usually it is about the classification of the transactions under both accounting standards and tax law.
I wouldn't agree that Reagan Era software is still being used. I've being doing accounting since 1988. Back then spreadsheets didn't even have tabs. You couldn't just download your transactions from the bank and have it automatically reconcile. If there were ETL systems back then they weren't on every user's desktop like Power Query. LLM's are still of great value. I can translate a German Tax return to English, I use it to record transactions from paper bank statements. It was pretty hard to do with OCR software. I can easily generate code if I need to. I can cook up new functions and behaviors in Excel.
What is your typical workflow to convert paper bank statements or even PDF bank statements to correct structures data entries? I am building something in this space thus wondering how users are currently doing this. Thanks a lot in advance.
I mean here in Asia lot of us are using CSV from bank, we don't get OFX files or they aren't yet common, but yeah there are converters for CSV to OFX then the accounting programs can import them and then need to match based on reference field or description or enter it as new transaction
@@travelinggreek true. To say in theory and even practice Block Chain is a public ledger, reconciliations will be public data, so there will be privacy issues which now people are not realizing transacting on Blockchain. They already have built De-anomyzing tech for reconciling Bitcoin tranfers they use it in battling financial crime, it's just hidden from general public to give a facade of privacy.
Great vid. Have a feedback though,,,, the way captions show up is way too distracting ! For a non native speaker, I end up reading them a lot, even though the content is very clearly put across
We're working with structured data, and the fun challenge is creating AI that can bridge accounts payable and accounts receivable, along with a few other areas. Plus, we want it to truly understand the context! Nothing out there is capable of that as of today
Why is A16Z so out of touch, these days? Accountancy is first and foremost about accuracy. AI hallucinations and errors are well-documented. How many CFOs want to sit in front of a Senate Subcommittee because the AI doing their accounts made a grievous error?
Not to mention the accounting standards board is run by people who have negative infinity interest in making themselves and the industry at large financially worse off. Do these people not understand basic economics?
Yep that really depends on the technology as well but anyway it's not technology going to replace rather it's another person who uses that technology. But with AI models keep safety of data privacy like everything someone asks an AI can be recorded and information is passed on to outside while using those services. One way to resolve is to use self or private hosted AI
But yeah like lot of changes learn more on how to be review and reporting will help to handle much more transactions, data entry jobs will be affected a lot, lot of encoding work will be removed and more value for people who have much higher and more understanding of books of accounts and who are trustworthy even more trustworthy than an AI model on truthfulness they will win in accounting
@timexpeachtree Not saying not to learn but I know for sure lot of organizations don't trust third party AI as they can't see where data is going and how safe it is to share data , that's where we come in , learn as much as possible but learn application rather then technical aspects that's what I would say, many can and should differ from me, leading AI for one of top 10 US multinational globally that's how I see it and my background is CFA, CPA, CA & ACCA with MBA so you can understand Finance and Accounting is my bread and butter.
@@GameboyMufti yepp exactly so u get it. Finance will become optimized NGL especially for people who do bare minimum data processing only it will be difficult once Ai is entrenched within software lot of data processing will become redundant, only assurance and compliance and validation and of course trust will be needed from a Halal person, rest all the system and ecosystem will take care. Finance will become a powerful field only for those who want to achieve and be controlling and energizing their organizations.
The bloke said it, unlike creative industries, there is zero tolerance for errors or hallucination in accounting. And accounting deals with numbers not words, and each number needs to be considered in context. This all beyond llms at this stage.
LLMs have to take a back seat and give it over to rule based or logic based AI on this subjects. A fundamental problem of AI hype is people assuming LLMs are the complete of AI.😅 There are lot more areas of AI that need to be updated and then it will be within the software they will be able to automate lot of details.
@@De-tw7bywon't be offices of people this time around it will be servers and mainframes in datacenters with lakhs of automation agents running and your traditional chartered accountant overseeing the books of 100x without paying anyone else, making their mind how to extract most value from AI while minimizing the cost of operation and thinking up new algorithms or the next new government subsidy. Bear in mind those who thought to survive my doing minimal or lite work will be automated, no more number punching and clock-punching it will be accuracy and validations, and staying withing bounds of accepted deviation
@@De-tw7by Exactly lol. the last mile of tech will remain in the west, everything else is outsourced. SWEs in the west will be the cause of their own dmeise.
All respect to these guys, but the reason Accounting AI might be slow is that they have a fundamental misunderstanding of accounting. "It's all about numbers, unlike law." Very false; the majority of accounting work is classifying transactions or assets based on unclear government guidance paired against unique business operations. "The numbers have to be 100% correct 100% of the time." Also false; they need to be correct within the limit of "materiality", which might actually work in favor of LLM's; if they hallucinate every ten-thousandth transaction, that would still be better than human systems, and within materiality limits.
You can build system that prevents hallucinations (source: been building stuff since babbage). Hallucination or false positives are not the issue.
Welcome to VC where investors pontificate about areas where they have no domain experience
@@A704T yeah seems like it
I would like to see how this ages 😂
"The majority of accounting work is classifying transactions or assets based on unclear government guidance paired against unique business operations"-Thanks Jared for this enlightening words!
I also disagree with the claim that accounting is about numbers. It is about exercising professional judgements to determine how to classify those numbers. Usually it is about the classification of the transactions under both accounting standards and tax law.
I wouldn't agree that Reagan Era software is still being used. I've being doing accounting since 1988. Back then spreadsheets didn't even have tabs. You couldn't just download your transactions from the bank and have it automatically reconcile. If there were ETL systems back then they weren't on every user's desktop like Power Query.
LLM's are still of great value. I can translate a German Tax return to English, I use it to record transactions from paper bank statements. It was pretty hard to do with OCR software. I can easily generate code if I need to. I can cook up new functions and behaviors in Excel.
What is your typical workflow to convert paper bank statements or even PDF bank statements to correct structures data entries? I am building something in this space thus wondering how users are currently doing this. Thanks a lot in advance.
I mean here in Asia lot of us are using CSV from bank, we don't get OFX files or they aren't yet common, but yeah there are converters for CSV to OFX then the accounting programs can import them and then need to match based on reference field or description or enter it as new transaction
It’s the tech hyperbole
awesome. but please remove these distracting subtitles in the video.
the ones we can activate are sufficient.
Make your own channel, then you can create vids however you want
@@alexatedw so people should never give feedback?
@@HurricaneChise only if asked
AI will definitely ease accountants' pressure by handling repetitive tasks that take up a lot of time.
@@travelinggreek true. To say in theory and even practice Block Chain is a public ledger, reconciliations will be public data, so there will be privacy issues which now people are not realizing transacting on Blockchain. They already have built De-anomyzing tech for reconciling Bitcoin tranfers they use it in battling financial crime, it's just hidden from general public to give a facade of privacy.
Great vid. Have a feedback though,,,, the way captions show up is way too distracting ! For a non native speaker, I end up reading them a lot, even though the content is very clearly put across
We're working with structured data, and the fun challenge is creating AI that can bridge accounts payable and accounts receivable, along with a few other areas. Plus, we want it to truly understand the context! Nothing out there is capable of that as of today
Why is A16Z so out of touch, these days? Accountancy is first and foremost about accuracy. AI hallucinations and errors are well-documented. How many CFOs want to sit in front of a Senate Subcommittee because the AI doing their accounts made a grievous error?
Not to mention the accounting standards board is run by people who have negative infinity interest in making themselves and the industry at large financially worse off. Do these people not understand basic economics?
Timely video. Basis AI just announced a Series A today, building AI agents for accountants
Timely or timed?
Hmm...didn't realise a16z also has a 'moderator' panel to delete comments that counter this hubris
I know. Don't forget Marc is also throwing hubris & no one moderates him.
AI could save accounting firms billions, but at what cost? Fewer jobs? Cheaper services?
Its amazing how software testing which is soo repetitive is not automated. Google , Microsoft anyone?
Excellent video!
Captions are brutal
IRS expense deduction categorization is the most abundant, thoughtless work in accounting.
We Finance and Accounting people are intelligent enough to create new ways to keep us employed for next 50 years 😁
No
Yep that really depends on the technology as well but anyway it's not technology going to replace rather it's another person who uses that technology. But with AI models keep safety of data privacy like everything someone asks an AI can be recorded and information is passed on to outside while using those services. One way to resolve is to use self or private hosted AI
But yeah like lot of changes learn more on how to be review and reporting will help to handle much more transactions, data entry jobs will be affected a lot, lot of encoding work will be removed and more value for people who have much higher and more understanding of books of accounts and who are trustworthy even more trustworthy than an AI model on truthfulness they will win in accounting
@timexpeachtree Not saying not to learn but I know for sure lot of organizations don't trust third party AI as they can't see where data is going and how safe it is to share data , that's where we come in , learn as much as possible but learn application rather then technical aspects that's what I would say, many can and should differ from me, leading AI for one of top 10 US multinational globally that's how I see it and my background is CFA, CPA, CA & ACCA with MBA so you can understand Finance and Accounting is my bread and butter.
@@GameboyMufti yepp exactly so u get it. Finance will become optimized NGL especially for people who do bare minimum data processing only it will be difficult once Ai is entrenched within software lot of data processing will become redundant, only assurance and compliance and validation and of course trust will be needed from a Halal person, rest all the system and ecosystem will take care. Finance will become a powerful field only for those who want to achieve and be controlling and energizing their organizations.
Interesting
I’m not sure what AI can do that isn’t already being done by existing software. As you recognized, there is no room for error.
Literally Automation Anywhere and UiPath exist and already ran this thesis/playbook down to death.... This sounds good, but is utterly missing reality
Finance functions are primed for disruptive through AI; big and bloated with too much SG&A attributed to them
The bloke said it, unlike creative industries, there is zero tolerance for errors or hallucination in accounting. And accounting deals with numbers not words, and each number needs to be considered in context. This all beyond llms at this stage.
I won't say it's beyond but certainly needs work.
Hallucinating is all but gone now
LLMs have to take a back seat and give it over to rule based or logic based AI on this subjects. A fundamental problem of AI hype is people assuming LLMs are the complete of AI.😅 There are lot more areas of AI that need to be updated and then it will be within the software they will be able to automate lot of details.
@ lololol
You can say Everything needs ai ..I mean common guys, let’s get some serious analysis
Say goodbye to all the back offices in India
The back office will only increase in India.
@@De-tw7bywon't be offices of people this time around it will be servers and mainframes in datacenters with lakhs of automation agents running and your traditional chartered accountant overseeing the books of 100x without paying anyone else, making their mind how to extract most value from AI while minimizing the cost of operation and thinking up new algorithms or the next new government subsidy. Bear in mind those who thought to survive my doing minimal or lite work will be automated, no more number punching and clock-punching it will be accuracy and validations, and staying withing bounds of accepted deviation
@@De-tw7by Exactly lol. the last mile of tech will remain in the west, everything else is outsourced. SWEs in the west will be the cause of their own dmeise.