A synthetic data software co. to boost SAS’s Gen AI portfolio.
Does this acquisition push back SAS’s IPO/sale timeline?
A sign of a changing strategy to hang in and compete?
https://www.prnewswire.com/news-releases/sas-acquires-hazy-synthetic-data-software-to-boost-generative-ai-portfolio-302300944.html
“ History does not depend on who writes it — if they’re honest. A disaster is a disaster.”
History ALWAYS depends on who writes it. A success is a success.
History does not depend on who writes it — if they’re honest. A disaster is a disaster.
But neither Viya nor Hazy will matter to a private buyer. A private buyer will be motivated by the legacy $3B ARR, and won’t care about small new revenue streams.
As BH has said, Hazy does help with an IPO, because AI represents potential growth. An IPO would work best while the market is strong, and before the disaster is widely recognized.
@ezjl+1vskrCuD Depends on who writes the history books. Historians or whiny malcontents.
well guess he "wins" then, paid a lot to sc--w it up for everyone. who cares. not the winner in the history books.
He made more money at SAS than you ever will, implemented a failed company-wide strategy, then scrambled back to the comfort of the ivory towers and left everyone else holding the bag!
Now it’s too late and too expensive to pivot and current executive management has no incentive to. They’re staring at huge golden parachutes once SAS is sold.
What are you staring at when SAS is sold? One week of severance for every year, if you’re lucky.
"The Viya architect evidently had no similar experience."
Yep. And we see where his arrogance got him. Right back to where he started. That was a major comeuppance.
Customer is not always right no matter how much we have been brainwashed to think it.
Should you always treat customer with deference and respect? Of course.
Should you make the think every idea is theirs? Maybe.
But to just shut off your brain and only do what customers say is idiocy.
They are clearly not always right. Nor are you. Nor am I.
If we only ever built what customers said they wanted we would be in the stone ages. You have to evolve things whether the masses see it or not.
Some will succeed many will fail. But just going with what has always worked is simplistic.
There is nothing wrong with academics. SAS was founded by academics.
But one of them worked in his Father's hardware store, where he learned that the customer is always right.
The Viya architect evidently had no similar experience.
@4bok+1vskrCuD Pretty nice theory you have there…
"In theory, theory and practice are the same. In practice, they are not.”
Right! Theory is the mindset of an academic person. Practice is the mindset of a business person.
Theory is great but only when the application of it meets an unfulfilled market need. The Big German was so convinced his theories were right that he wrongly refused to listen to the voices of the business folks.
That is where Viya went wrong and the company remains unsold.
In theory, sure.
“In theory, theory and practice are the same. In practice, they are not.”
Synthetic data can indeed help improve the accuracy of fraud detection models in several ways:
Fraud is often a rare event, so the real-world dataset may not contain enough fraud cases for a model to learn from effectively. Synthetic data allows for the generation of a larger number of fraud examples, helping models recognize patterns in fraudulent behavior that they might otherwise miss.
Synthetic data can help balance the dataset, reducing class imbalance issues that occur when there are far fewer fraud cases than legitimate ones. Balanced data can improve model training and reduce bias toward non-fraudulent outcomes, helping models catch more nuanced fraud cases.
Synthetic data generation methods can simulate a wide variety of fraud patterns, some of which may not yet appear in real-world data but are likely to emerge. By training on this variety, models can become more resilient and better prepared for novel fraud tactics.
With synthetic data, sensitive customer information can be protected, as the data is artificially generated and does not correspond to real individuals. This approach ensures compliance with data privacy regulations and avoids the risk of exposing private information while still enabling effective fraud detection research and development.
Training solely on limited real data can lead to overfitting, where the model learns the specific details of the training data too well and fails to generalize to new cases. Synthetic data introduces new scenarios, helping the model generalize better to new fraud cases.
Incorporating synthetic data into fraud models can be a powerful strategy to increase their accuracy and robustness, but it's also essential to ensure that the synthetic data realistically represents the complexities of fraud behaviors.
@2ror+1vskrCuD well go on then, please explain why it isn't nonsense? I'm genuinely interested and willing to learn.
I think @2ynk+1vskrCuD is right; simply creating more fraud events doesn’t make better data. In fact, it’s a worse match to the real data, which should make the model worse.
There are good uses for synthetic data, but it has to match the real data.
@2ynk+1vskrCuD Just because it sounds like bullsh-t to you doesn’t make it bullsh-t. I could explain to you why it isn’t nonsense but…
"One scenario is fraud. A credit card company may be servicing billions of transactions a day. Very few of those transactions are fraud. But by using synthetic data, you’re creating more of those fraud events and better training the AI to know what to look for when it comes to fraudulent transactions."
This sounds like absolute bullsh-t to me. The only way you could create realistic fraud events would be if you already knew what the predictors of fraud are, and if you already know what those predictors are, how is a larger model going to help? Synthetic data is not going to somehow magically identify more patterns in the data that'll help predict fraud...it's just nonsense.
My understanding, and correct me if I am wrong, is that you must run statistical tests to validate that your synthetic data has the same characteristics as the real data. If not, your synthetic data will generate a synthetic model that does not match the real world.
In some cases like computer vision the validation is relatively simple; there are plenty of PDFs and street scenes publicly available.
In other cases like finance and healthcare, SAS doesn't have the real data, but could perform validation at customer sites under a consulting contract.
It's a reasonable business. Although it might not attract a private buyer, it would help with an IPO.
Will it be as successful as the CyberSecurity product?
@2jwl+1vskrCuD
"...Anyway, if I have real data, why don't I work with it directly?"
But that is just it - you probably do not.
My understanding is that it goes something like this. Suppose that you want to build a model to predict the likelihood of purchasing particular items that you sell. You can use your data to build a model that fits your current customers - you have that data. But you want to expand your sales to people who are not your customers - for them you have no data. Synthetic data (if it is accurate with respect to the population that you want to reach) enables you to model those customers as well.
If I completely misunderstand, please chime in.
Does the acquisition of Hazy mean one more disposal item that does not fit the mindset of the buyer?
It seems that JG is trying to set a post sale trajectory for the company by rebranding SAS as a significant AI player versus the reality that the biggest asset is the declining V9 revenue stream.
SAS has two major problems:
Ah! A new hacked-together product that will have slow performance, that helps to push the IPO myth! What's not to love?
Another desperate jump on a bandwagon without knowing where it's going.
Ki-ler app of synthetic data is computer vision. You generate fake PDF documents to train AI to process, you generate fake street scenes for automatic driving AI's to navigate... Does SAS know computer vision?
Fake "tabular data that reflect the statistical properties of real data" is no different than "god exists". Can you prove it? can you disprove it? Anyway, if I have real data, why don't I work with it directly?
An ignorant SVP in my firm wanted to buy a synthetic data provider and above argument quieted him immediately.
I think, as Mr. Harris suggests, that this acquisition accelerates an IPO.
It doesn't seem to help attract a private buyer. A private buyer's target would be the SAS recurring revenue stream. Buyers of large declining revenue streams include Broadcom and private equity.
A buyer laying out $5-10B for $3B ARR won't care about this $35M investment. It's not big enough to be on their radar, and it doesn't fit their business model.
But if SAS plans to IPO, in this market, sprinkling a little AI on the product line gives a definite advantage.
TRIANGLE BUSINESS JOURNAL
SAS targets 'multibillion-dollar market' with acquisition
By Lauren Ohnesorge – Senior Staff Writer, Triangle Business Journal Nov 12, 2024
As it prepares to go public, SAS Institute is again investing in artificial intelligence by building its synthetic data generation capabilities through a rare technology asset acquisition.
The Cary company announced it is acquiring Hazy synthetic data technology, a move that comes with 13 employees. The plan, according to SAS, is to integrate Hazy’s design into the SAS Data Maker solution, which it debuted in May.
Integration of the technology should be complete by May 2025.
In an interview, SAS Chief Technology Officer Bryan Harris said the synthetic data technology enables an extension of what the company is already doing, including announcing the creation of its AI modeling group earlier this year. He said it all bodes well for the company’s IPO readiness plan, a priority for SAS since CEO Jim Goodnight went public with plans to pursue a public market debut in 2021.
In April, Harris said the company was about a year away from being ready to go public.
On Monday, he said the acquisition is another driver toward the company’s ambitions of someday becoming a public AI company.
“We think this is a multibillion-dollar market opportunity,” he said of the deal, calling it a “massive opportunity revenue-wise, which will obviously drive positive outcomes for our eventual IPO.”
The deal for Hazy's technology
SAS has been in talks with Hazy for eight months to a year, having evaluated multiple options before deciding Hazy’s “technology and their expertise really stood out in the market.”
Harris explains the technology as helping to train advanced AI models. Typically, a company starts with its own data.
“The problem is … the current data, the actual data doesn’t always reflect the complexity of the real world. It only reflects the complexity of the data you are seeing,” he said.
By adding synthetic data, you’re increasing the diversity of scenarios and customer behaviors you’re using to train those models without having to compromise privacy information.
One scenario is fraud. A credit card company may be servicing billions of transactions a day. Very few of those transactions are fraud. But by using synthetic data, you’re creating more of those fraud events and better training the AI to know what to look for when it comes to fraudulent transactions.
Hazy, a British company, was founded in 2017 and raised $9 million last year from investors such as Conviction VC, UCL Technologies, M12 (Microsoft’s venture capital fund) and Wells Fargo (NYSE: WFC).
The company has raised a total of $14.8 million, according to Crunchbase.
In a statement, Goodnight said the deal “represents a pivotal step in our commitment to innovation in the next generation of data management and AI.”
“By integrating their technology, we can offer our customers unparalleled opportunities to harness data safely and effectively, enabling them to experiment and model scenarios that were previously out of reach and gain a competitive advantage,” he stated.
Gartner predicts that by 2026, 75 percent of businesses will use generative AI to create synthetic customer data, up from less than 5 percent last year.
SAS, long known for its analytics, has been making major investments in the AI space. SAS announced a second $1 billion investment into AI last year.
The first $1 billion commitment in 2019 was about the platform. The second, announced in May of last year, was about industry solutions, according to the company.
It’ll all just have to be rewritten in C as TK extensions.
You make synthetic data sound like a bad thing. There are numerous good uses of it that do not compromise the integrity of models or other use cases.
Of course it could build it in house…. And work has gone into it. But why spend time and money building it if can buy something that has already done the hard parts. More expensive? Maybe but often worth the time difference if what you are buying is quality.
Ha. I remember when data warehousing was all about cleaning up the data you had. When did it become fashionable to stir in a good dollop of fake sh1t? Surely the models built thereon have some component of fakeness?
SAS fired enough engineers that they can no longer "build it in house"?
Time to take the keys away from them sweet lasses in HR. They making a mess!
Google search says Hazy is valued at $34.7M as per the latest available filings. Pocket change for the man in charge.