Rendered at 21:37:07 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
alexpotato 10 hours ago [-]
Having worked at some giant firms that heavily depend on SAP, I would think their position as "system of record for procurement/production" would get stronger with AI.
e.g. AI + a good database of what your company does seems like an excellent combination.
snarfy 8 hours ago [-]
I've been impressed so far with Snowflake's offerings of "AI + a good database"
johannes1234321 8 hours ago [-]
Or AI can be a threat: If AI can spit out good enough special purpose tools a lot of the SAP adaption business goes away. If AI then also assists in data migration the moat goes away.
johnbarron 8 hours ago [-]
EAI has existed for twenty years...Tibco, webMethods, IBM AppConnect...SAP is still what it is. How is AI going to make a dent on this?
elias_t 9 hours ago [-]
The same argument can be made about snowflakes and other providers that handle the data layer. If they can leverage AI in a safe contained environment they would become the clear winner
happyPersonR 7 hours ago [-]
LOL
Although they may internally use LLM, I wanna say said they were not going to allow their core offering to be integrated with LLM
Although maybe they reversed course?
gruturo 10 hours ago [-]
Would a technology famous for hallucinating data often enough, and in many context unable to guarantee it won't assume and hallucinate when it feels "confident enough" despite a robust system prompt, be really considered for a "system of record for procurement/production" usage?
(yes yes RAG exists)
notme43 9 hours ago [-]
In the ERP space, AI is being used mostly for repetitive tasks, documentation and forecasting-type activities or where a non-deterministic answer might be useful. Scanning in documents and using OCR to create an order, asking what page I use to do X, historical demand modelling and constraint scheduling. Also have seen it used to do order flow type stuff like suggesting products to sell if they buy ABC. I'm not really seeing it being used to do heavy analysis and they've generally been careful not to put it in places where a hallucination could result in a bad business decision.
victor106 8 hours ago [-]
> Scanning in documents and using OCR to create an order,
I remember this being done more than 10 years ago.
The other uses cases you describe also were being done. Of course with AI maybe it’s become easier and cheaper?
notme43 7 hours ago [-]
Yes but it generally expected documents to be in a certain layout and the user had to fill in a lot of the gaps. It's more able to identify elements on an incoming PO, match a supplier invoice, or create a receipt based on a packing list with inference.
EDIT: The other activities, yes I think it made it cheaper and easier. They both required human intervention to make an accurate forecast or execution plan. Forecasting requires humans to do things like apply smoothing factors, identify abormal demand over huge datasets, or identify seasonal demand cycles, etc.
Constraint scheduling requires a human to intervene in dispatching and resource allocation as well. AI can do 95% of the work with a proposal, or even control the process. Supposedly much better than historical algorithms. I haven't implemented this part of it, but I've seen it done.
The documentation part is much more robust. You can ask it complex questions, like "How do I setup a phantom blow-through part in a MS level 1 BoM?" and it walks you through the entire thing.
With the sales suggestions, it can identify things like "The customer is buying spaghetti and pancetta" AI: "They are probably making carbonara, how about some garlic bread?"
spwa4 7 hours ago [-]
Take your favorite OCR software.
Then take a photo of a document and throw it into ChatGPT, or Gemini, along with the word "Transcribe".
ChatGPT wins, by a landslide. BUT AI's advantage doesn't stop there.
"Take this picture of some idiot filling in form 49, I've also attached the PDF, fill in the PDF fields, provide a database record according to the schema attached and flag if there are any obvious problems with the entries".
That works too. "Produce a latex document of this kid's math homework and flag any problems" - works. "Produce a MS word document of this letter" - works. "Read this bill and produce a JSON version following the schema from this example" ... and so on and so forth.
More than that, it is starting to work pretty well with Gemma 31B local model (will still do 30 document analyses on cpu only, at Q4 on a DDR4 or higher machine. Yes it's mostly memory speed that matters) at this point. I mean, that's GPT 5.0 or so quality (95% correct with the occasional problem), but hey, Qwen 3.8 27B may be coming out next week ...
RaftPeople 3 hours ago [-]
> forecasting-type activities
AI/ML yes, LLM's no (unless someone is naively using LLM's for forecasting)
mikert89 9 hours ago [-]
The big risk for these platforms is that AI reduces the cost to switch to another ERP. IF AI can map all of the data between two different competitors, the whole business model is finished
johnbarron 8 hours ago [-]
That is not what ERPs are used for.
Would love to hear how AI is going to help here. This is like a young engineer thinking the CDC machine they just bought, is going to make Airbus and Boeing obsolete...
mikert89 7 hours ago [-]
It’s well known that the moat of system of record is how difficult it is to extract your data into another system.
RaftPeople 3 hours ago [-]
> It’s well known that the moat of system of record is how difficult it is to extract your data into another system.
I think that is one aspect but not the biggest issue.
The specific shape of data and shape of functionality gets built around for decades due to incompleteness of any system, expanding dependencies and changing business needs.
Decades of building around the system results in a mountain of work when trying to replace the system, typically a larger body of work than implementing the base system itself.
mikert89 31 minutes ago [-]
most of this complexity is because 1000 different types of businesses are jammed into the same database schema and ERP. Its all non sense
zbentley 4 hours ago [-]
ERPs are often used as a lot more than a system of record: they’re directly engaged with as the LOB software point-of-contact by many expert employees, and function as a sort of encoding of organizational behavior.
They’re a lot more than a data store or system whose primary value is ingesting info and producing reports.
For a lot of business, replacing the ERP is on par with replacing Excel. It’s not about getting the data into another system; it’s about all of the complexity and expertise required to use it.
m4rtink 12 hours ago [-]
So will AI be what finally kills SAP, making world a little bit better? ;-)
Does it? Those are big numbers, but it looks like in 2024 they sold €4b of assets to juice income for 2025.
d1sxeyes 8 hours ago [-]
They didn’t need to? They wouldn’t have a particularly bad year.
g42gregory 3 hours ago [-]
Wait, first they stopped hiring because AI is going to replace everybody. Now they stopped hiring because AI, which was going to replace everybody, costs too much.
You can’t make this stuff up.
therealdrag0 2 hours ago [-]
Sometimes buying a tool to make 10 people 15% more effective is a better choice than hiring 1 person.
gruturo 12 hours ago [-]
So they are using AI to run the company, and AI allocated all the money to AI?
happycube 7 hours ago [-]
Sounds like a paperclip maximizer to me.
Seriously, why the duck aren't they running their own AI systems in house???
andrekandre 8 hours ago [-]
it may just be actually incredible what they are doing. on the other hand it could also be absolutely insane...
qw2187 10 hours ago [-]
Telling that SAP now exfiltrates all your company data to the U.S. AI cartel:
“Under the policy, only Al-related hiring, customer-facing travel and trips directly linked to the company's Al initiatives or mission-critical employee Al training are exempt, reflecting SAP's focus on "be disciplined in how we spend" while expanding its Al capabilities.”
I would pay to understand what mission critical employee Ai training is that can't be provided in other ways
j-bos 8 hours ago [-]
Executive meet and greet
zbentley 4 hours ago [-]
Or sales/FDE deployment
jwsteigerwalt 8 hours ago [-]
This is less of an issue than the headline suggests. I worked for a publicly traded US company remotely. The company was based between Ft. Lauderdale and West Palm Beach Florida. It was a roll of the dice each spring if the CFO would pull the trigger on a travel block for about 6 weeks around spring break when flights and hotels were most expensive. Approvals were available for critical projects. It was only ever about not exceeding a quarterly budget number; nothing else.
WesolyKubeczek 10 hours ago [-]
"Daddy, vodka got more expensive now, does it mean you will be drinking less?"
"No, kids, it means you will be eating less."
leoh 2 hours ago [-]
I sure am glad my employer isn’t a drunk daddy
bucket2015 9 hours ago [-]
Anecdotally, I know several companies that have given the directive of "Use more AI instead of hiring more people". I wonder whether this is part of the story here too?
mosdl 5 hours ago [-]
SAP also made some big acquisitions that cost a lot of cash..
jqpabc123 12 hours ago [-]
In other words, AI is now driving their budget decisions.
cineticdaffodil 11 hours ago [-]
[dead]
simianwords 11 hours ago [-]
If you believe that AI does add to productivity gains, say equivalent to adding 10-20% employees, would you do anything differently than SAP?
throwaway87543 9 hours ago [-]
If AI was actually making you more profitable, you wouldn't need to be cost cutting to avoid bankruptcy.
simianwords 9 hours ago [-]
I'm not sure who is getting bankrupt?
PolygonSheep 2 hours ago [-]
No one, because they cut costs to avoid it.
realusername 10 hours ago [-]
These are enough gains that you don't need to believe, they should be measurable
simianwords 9 hours ago [-]
Not really. A company can work and do things without having to run RCTs to prove everything. The market itself is the RCT.
realusername 9 hours ago [-]
If you doubt that you have gained something as high as 20% gains, it probably just means you got nothing.
And also, individual productivity isn't the same as organisation productivity, especially at such a large company.
andrekandre 8 hours ago [-]
> gained something as high as 20% gains
all other things being held constant, i suppose it should be as easy as +20% profit rate? tho im not an executive so what do i know...
drob518 9 hours ago [-]
This tells me we’re getting close to “peak bubble.” We saw these same sort of stupid headlines just before the dot-com bubble popped, too. When companies are distorting regular business to do the new thing, they’ve gone past the point of reasonableness on the new thing and we’re ready for a snap-back.
cyanydeez 11 hours ago [-]
teach everyone they live in a lobster bucket, and suddenly everyones a lobster.
budsniffer952 9 hours ago [-]
This "AI business" has really opened my eyes to how..insane the average person is, and just how bad discourse has gotten.
Look at the conversation: people are writing off this technology (on a tech forum no less) as useless, because it can make mistakes. We criticize the cost, and the business models, and the companies, the purveyors and users. Why? Because we don't quite understand how best to use it...3 years in.
People have lost all concept of time scales, I guess. They expect everything to be polished and mature NOW. How many people realize that could take a decade or more?
So, yeah, a stodgy old company has figured out they don't need to hire and don't need to travel, and AI is expensive. So what?
thelastgallon 9 hours ago [-]
This kind of thinking (we'll figure it out how to use it correctly, it is awesome technology) is not new. There was blockchain psychosis (blockchain/crypto solves everything!) for a decade, which is now replaced with AI will solve everything.
frameset 9 hours ago [-]
A bit of NFT madness to go with the Blockchain stuff too.
rf15 8 hours ago [-]
NFT was surprisingly short-lived considering it rode on the coattails of the blockchain
Zanfa 7 hours ago [-]
And can’t forget Metaverse and the $70B Meta sunk into it either.
derektank 8 hours ago [-]
The difference is that cryptocurrencies never had an obvious benefit beyond subverting the existing banking sector. A more apt comparison would be to cloud computing, which likewise had its host of detractors (“It’s just somebody else’s computer!”) but which brought very obvious benefits to small developers and enterprises which couldn’t host their own hardware.
gtowey 8 hours ago [-]
"This time it's different" is also a refrain we've heard with every new bubble. It shouldn't be surprising that some people want to take a measured, skeptical approach. Even if it means they are slightly behind the bleeding edge. Even if you have to play catch-up, it can be more efficient to let others find all the dead ends first and avoid them.
lijok 7 hours ago [-]
What does crypto have to do with this? DuPont has sunk billions into synthetics that were looking for a business model and ultimately proved useless. Why not compare AI to that?
budsniffer952 9 hours ago [-]
Is blockchain really your analogy here? Something that is also new?
You are proving my point.
You say it's "not new", but then don't seem to have any gauge of technological history. How long before the internet was monetized? 30 years?
johannes1234321 8 hours ago [-]
About 30 years ago we had the sitcom bubble. Tulip mania was in the 1630ies, about 400 years ago.
There always have been attempts to exploit others for personal gain and people following that.
The new thing is the speed in which this happens globally. But there had been other accelerators bringing equal jumps. Introduction of cross ocean telegraph cables, flight, book printing, ...
True, modern technology makes results globally available with very little marginal costs.
However on the spread of ideas the last item mentioned on the list above is extremely interesting to study: Book printing brought translations of the Bible into more places and took reproduction and interpretation of those texts out of the hands of the few (mostly the church) to broader masses (for as long as people could read) which lead to lots of political conflicts and redistribution of power in Europe with impact to colonies worldwide. By people following one interpretation of a text and derived rules more than another. To a large degree since books with ideas could now be replicated within days instead of years. As now ideas are replicated within minutes rather than days.
j-bos 8 hours ago [-]
This is reasoning from shallow similarity. If you have not been able to find areas where AI actually does solve things, I don't know what to tell you. Go look at the number of vulnerabilities being solved. Go look at the speed of their solves. On the opposite side, go look at the number of vulnerabilities being found and exploited. Go look at the speed of their being exploited. Yes, it's a negative externality, but it's an example of an extreme productivity boost, which is easy to measure. Take a minute dropping on YouTube or X and actually look for people who are making small scale manufacturing. Look at the speed of their development cycles. Not software development cycles, real world atom development cycles. And then compare it to the speed of their development as of five years ago. Notice a difference? Even the complaints of crypto, while valid, are a bit overstated given that there are hundreds of millions of people in third world countries where money transfers are not reliable, not stable, and they depend on crypto. Yes, there's a bunch of criminals as well. I agree. My point is superficial dismissal is head-in-the-sand behavior. And that's your choice. I mean, it's fine by me.
lijok 7 hours ago [-]
This isn’t unreasonable at all. AI has promised to cause an unprecedented, continuous disruption to the working class. “Being replaced by robots” - a topic anyone of working age is very familiar with. Why would you be surprised there exists a proportionate amount of pushback in the form of incoherent rhetoric?
aboardRat4 8 hours ago [-]
It's not useless, it's just dangerous to use. It's like heroin basically. It does make you feel good instantly, and it can be a life-saver in certain situations of extreme pain or chronic pain (cancer), but there is a reason people are prohibiting the original substance, and also trying to develop less dangerous versions, such as methadone and tramadol.
Foobar8568 8 hours ago [-]
What I find the most amusing is ... "AI code is crap".
I think I heard enough time "this codebase is shit, we need a full rewrite" to know that no one agrees what is quality code.
Suddenly, everyone are expert writing the most elegant, clean and without bug code.
ArtTimeInvestor 11 hours ago [-]
It will be interesting to see if SAP can turn into the basis for companies that use AI heavily or if companies that use AI heavily will prefer to replace SAP.
Who are the contenders that start fresh with focus on AI?
One that comes to mind is IONOS. They seem to be in the process of building something like a "SAP in the times of AI" from the ground up. Starting with the demands of SMEs. It looks like an interesting approach to me: Having a "knowledge base" for each customer, initially filled with the data from the website. Then let the customers add more data. Then offer agents that do work for the customers. Answer calls and emails and do marketing for example.
jimnotgym 11 hours ago [-]
I have in mind what I think is a very strong framework to allow companies to replace legacy ERPs quickly with AI help. I guess I should get on with it and publish something!
sofixa 11 hours ago [-]
> It will be interesting to see if SAP can turn into the basis for companies that use AI heavily or if companies that use AI heavily will prefer to replace SAP.
Considering the human and institutional complexity of implementing ERP, CRM, whatever the hell Concur is supposed to be, etc. I'd bet on most organisations implementing AI around and with those systems instead of replacing them with something vibecoded / "AI native" vendors.
e.g. AI + a good database of what your company does seems like an excellent combination.
Although they may internally use LLM, I wanna say said they were not going to allow their core offering to be integrated with LLM
Although maybe they reversed course?
(yes yes RAG exists)
I remember this being done more than 10 years ago.
The other uses cases you describe also were being done. Of course with AI maybe it’s become easier and cheaper?
EDIT: The other activities, yes I think it made it cheaper and easier. They both required human intervention to make an accurate forecast or execution plan. Forecasting requires humans to do things like apply smoothing factors, identify abormal demand over huge datasets, or identify seasonal demand cycles, etc. Constraint scheduling requires a human to intervene in dispatching and resource allocation as well. AI can do 95% of the work with a proposal, or even control the process. Supposedly much better than historical algorithms. I haven't implemented this part of it, but I've seen it done.
The documentation part is much more robust. You can ask it complex questions, like "How do I setup a phantom blow-through part in a MS level 1 BoM?" and it walks you through the entire thing. With the sales suggestions, it can identify things like "The customer is buying spaghetti and pancetta" AI: "They are probably making carbonara, how about some garlic bread?"
Then take a photo of a document and throw it into ChatGPT, or Gemini, along with the word "Transcribe".
ChatGPT wins, by a landslide. BUT AI's advantage doesn't stop there.
"Take this picture of some idiot filling in form 49, I've also attached the PDF, fill in the PDF fields, provide a database record according to the schema attached and flag if there are any obvious problems with the entries".
That works too. "Produce a latex document of this kid's math homework and flag any problems" - works. "Produce a MS word document of this letter" - works. "Read this bill and produce a JSON version following the schema from this example" ... and so on and so forth.
More than that, it is starting to work pretty well with Gemma 31B local model (will still do 30 document analyses on cpu only, at Q4 on a DDR4 or higher machine. Yes it's mostly memory speed that matters) at this point. I mean, that's GPT 5.0 or so quality (95% correct with the occasional problem), but hey, Qwen 3.8 27B may be coming out next week ...
AI/ML yes, LLM's no (unless someone is naively using LLM's for forecasting)
Would love to hear how AI is going to help here. This is like a young engineer thinking the CDC machine they just bought, is going to make Airbus and Boeing obsolete...
I think that is one aspect but not the biggest issue.
The specific shape of data and shape of functionality gets built around for decades due to incompleteness of any system, expanding dependencies and changing business needs.
Decades of building around the system results in a mountain of work when trying to replace the system, typically a larger body of work than implementing the base system itself.
They’re a lot more than a data store or system whose primary value is ingesting info and producing reports.
For a lot of business, replacing the ERP is on par with replacing Excel. It’s not about getting the data into another system; it’s about all of the complexity and expertise required to use it.
You can’t make this stuff up.
Seriously, why the duck aren't they running their own AI systems in house???
https://www.sap.com/products/ai-platform.html
“Under the policy, only Al-related hiring, customer-facing travel and trips directly linked to the company's Al initiatives or mission-critical employee Al training are exempt, reflecting SAP's focus on "be disciplined in how we spend" while expanding its Al capabilities.”
no-paywall: https://news.ycombinator.com/item?id=49222851
"No, kids, it means you will be eating less."
And also, individual productivity isn't the same as organisation productivity, especially at such a large company.
Look at the conversation: people are writing off this technology (on a tech forum no less) as useless, because it can make mistakes. We criticize the cost, and the business models, and the companies, the purveyors and users. Why? Because we don't quite understand how best to use it...3 years in.
People have lost all concept of time scales, I guess. They expect everything to be polished and mature NOW. How many people realize that could take a decade or more?
So, yeah, a stodgy old company has figured out they don't need to hire and don't need to travel, and AI is expensive. So what?
You are proving my point.
You say it's "not new", but then don't seem to have any gauge of technological history. How long before the internet was monetized? 30 years?
There always have been attempts to exploit others for personal gain and people following that.
The new thing is the speed in which this happens globally. But there had been other accelerators bringing equal jumps. Introduction of cross ocean telegraph cables, flight, book printing, ...
True, modern technology makes results globally available with very little marginal costs.
However on the spread of ideas the last item mentioned on the list above is extremely interesting to study: Book printing brought translations of the Bible into more places and took reproduction and interpretation of those texts out of the hands of the few (mostly the church) to broader masses (for as long as people could read) which lead to lots of political conflicts and redistribution of power in Europe with impact to colonies worldwide. By people following one interpretation of a text and derived rules more than another. To a large degree since books with ideas could now be replicated within days instead of years. As now ideas are replicated within minutes rather than days.
Suddenly, everyone are expert writing the most elegant, clean and without bug code.
Who are the contenders that start fresh with focus on AI?
One that comes to mind is IONOS. They seem to be in the process of building something like a "SAP in the times of AI" from the ground up. Starting with the demands of SMEs. It looks like an interesting approach to me: Having a "knowledge base" for each customer, initially filled with the data from the website. Then let the customers add more data. Then offer agents that do work for the customers. Answer calls and emails and do marketing for example.
Considering the human and institutional complexity of implementing ERP, CRM, whatever the hell Concur is supposed to be, etc. I'd bet on most organisations implementing AI around and with those systems instead of replacing them with something vibecoded / "AI native" vendors.