While AI is fraught with challenges, the opportunities for businesses are enormous. SAS gives them the tools to deploy the technology in a meaningful, secure and productive way. “There has never been a better time to be a developer,” the company states during SAS Innovate on Tour Benelux in Rotterdam.
That statement comes from Véronique van Vlasselaer, Analytics & AI Lead South West & East Europe at SAS. Perhaps that statement comes across as a bit odd, especially now that the workload on developers is greater than ever. However, SAS’ safe, fast tools should get them out of the doldrums. These not only solve common problems, but also work preventively. Let’s explore that further.
Lack of AI in practice inhibits innovation
Bob Messier, SVP Global Technology Services at SAS, highlights Viya Copilot. We wrote about it back in April when it was launched as a private preview. Like other Copilots, it aims to boost productivity of users, in this case those working on the SAS Viya platform. It should ensure that data analytics is democratized with AI functionality. Specialized SAS Viya Copilots then get to work within domain-specific workflows, ensuring there’s tailored AI assistance within each work area.
For AI/ML development, the SAS Viya platform is said to be 30 times faster than competing products. Specifically, this includes alternatives such as SparkML and H2O. In some benchmarks, the speed improvement over commercial competitors is 49x, with more complex models as much as 326x. As a result, to call SAS “fast” is a bit of an understatement, based on tests in collaboration with The Futurum Group.
Finding the right data and preparing it for AI deployment, often a hellish task for developers, is also easier than ever. Thanks to the SAS Data Maker, synthetic data can be generated, which is particularly suitable for government and healthcare use cases. After all, personal data is often not allowed to be processed for AI deployment within these highly regulated domains, even though high-quality, specific data is normally required for AI models that are up to scratch. Van Vlasselaer says that adding synthetic data to training data produces just as good a model as with just authentic data.
The performance of AI models in practice is often unknown. This is because models are often simply not deployed, Messier explains. “As a result, you don’t learn from the real-world application. Without those lessons, you don’t get the chance to refine or retrain the model.” Those who do get said chance can compare AI models within SAS Viya. The platform even provides a full “nutrition score” for these models, if you will, ncluding ratings on bias, explainability and compliance.
Real gains
Those who do deploy AI in practice must, above all, ask themselves for what purpose they are doing so. “LLMs alone do not solve business problems,” Van Vlasselaer asserts. It is also necessary to integrate them into processes, orchestrate them and maintain them. Generative AI can be made useful multiple times within these processes.
An example SAS provides is that an LLM summarizes a customer conversation, SAS Viya then adds the relevant customer data and already suggests a template for an e-mail message to the customer. A bank which SAS provides solutions for, this way of working has already yielded significant improvements. It has now been able to handle 20 percent more complaints, speed up customer responses by 30 to 40 percent and reduce customer handling costs by 8 to 15 percent.
Behind the scenes
Developers can only achieve this AI deployment with the right tools. SAS has the Viya Workbench to this end, which addresses many problems among developers. Setting up a development environment is typically slow and expensive when done in the cloud. Workbench also runs in the cloud, but is intended to make deployment a relative breeze and cheap, too. Developers can code in SAS, as well as Python. Soon, the R language will also be supported. In addition, Viya Workbench supports the three well-known clients used in the field: Jupyter Notebook, Visual Studio Code and SAS Enterprise Guide.
Also, SAS offers App Factory, a React-based interface to build apps with AI models, both for internal use and external customers. CTO Bryan Harris indicated last year that the SAS platform is constantly being adapted to the needs of data scientists and developers, with an view toward greater productivity and faster innovation.
Read more: SAS Viya Workbench and App Factory promise rapid AI development
Stumbling blocks
SAS will also deliver AI outside of SAS Viya through “Models-as-a-Product.” These are models suitable and vetted for specific purposes. Casper Pedersen, Head of Data Ecosystems & Strategy EMEA at SAS, says Models-as-a-Product does not yet have a concrete launch date, but it should appear later this year. The benefit to customers? “Cost, cost, cost,” Pedersen said. These are lightweight, affordable models for certain industries, such as computer vision for factory halls, GenAI for customer service bots, models for fraud detection and more.
Pedersen mentions one application that he admits sounds a bit trite by now: AI assistants. Still, these can just be very useful. An example includes help for drivers who need to find the most sensible delivery location, with AI consulting them in natural language. Without question, this requires a cultural shift beyond just implementing new technology: employees will have to get used to their AI aid. That will naturally happen if the end results are good, but that’s a bit of a chicken-and-egg story. It is up to AI technology to be undeniably, inescapably good. For now, it’s already excelling, for example, in various factories, banks and more.
There are still some general stumbling blocks for this kind of implementation. One of them is unstructured data: how do you ensure that all the data is deployable for AI? Using SAS models, organizations can turn a treasure trove of poorly scanned PDFs with hard-to-read handwriting into useful tabular data, with everything filled in that is even slightly discernible via computer vision. Humans won’t have to do this kind of repetitive work anymore.
Mature
We’ve said it before: SAS is taking a mature approach to AI. At 48 years old, the company is a seasoned player and has had to continually reinvent itself. In the AI era, this experience is proving to be an advantage. It knows better than anyone that new developments must fit within existing processes and practices. The elimination of repetitive work and the likelihood of broad adoption in a variety of work fields are key targets. Compliance and trust are also of great importance. Unlike many other IT companies, it does not have to present AI as a major reinvention, but rather explain it with patience and attention to detail.
The idea that it has never been a better time to be a developer legitimizes itself in this way. There is an acceleration taking place in the data/AI lifecycle, even though only 6 percent of companies in the Benelux are using GenAI and have also fully integrated it into internal work processes. This is not surprising, Van Vlasselaer argues. Gaining confidence takes time. Still, she indicates that there is no denying the interest. “Most conversations with customers are about GenAI,” she says. That will continue to be the case for some time, with SAS able to put forward an increasingly comprehensive package to answer all questions with solutions.
Also read: SAS brings tool to Microsoft Fabric: translating AI into business outcomes