What advice do you have for others considering SAS Viya Platform?

Our company is very large, so we use multiple tools. We are using Tableau and Power BI in parallel, and based on the use cases, we switch to SAS Viya Platform. Wherever we feel it is really good and where we can build and deliver to users in an efficient, easy, and fast way, we use SAS Viya Platform. However, in our company, which is very large, we use many tools. SAS Viya Platform is good overall. Two main improvements would transform this into a great tool: improving the UI and calculation capabilities, and merging AI capabilities. I would rate this product seven out of ten.

In my daily work, I use SAS Visual Investigator to create an alert for fraud claims. Based on that data, if the alert has been raised, it will go to SAS Visual Investigator, and the business people will investigate that and send it. The shift to real-time detection through SAS ESP has had a noticeable impact on response time. Previously, the entire pipeline was batch-based; the source data came in via the pipeline, landed in SQL servers, and then the VI process happened, meaning frequently a claim could have been flagged after the batch cycle completed. With SAS ESP consuming Kafka streams, suspicious claims are now scored as the data arrives, so alerts reach SAS Visual Investigator within seconds or minutes of the event occurring. The practical impacts are that investigators can act on a claim before it is processed or paid out rather than after, and the high-risk real-time alerts are prioritized separately. It has reduced the time window in which fraudulent claims could slip through undetected. My implementation uses traditionally the library-based access rather than CAS in-memory processing. The data volumes and architecture of my project were already handled through SAS 9.4, so pipelining and feeding into Viya meant that the in-memory process was leveraged only for SAS Viya Platform users, not the data processing team. SAS Viya Platform provides built-in AI and machine learning capabilities through the Model Studio for building and training models and Model Manager for deployment, supporting automated machine learning. However, in my project, the models were largely pre-built, and my exposure is more on the pipeline and deployment side rather than model building itself. Regarding governance in SAS Viya Platform, I believe Model Manager provides a central repository to track model versions. Models go through defined promotion workflows and then audit. In my project context, this is important because any change in the scoring model directly impacts the fraud alert generation. From a security and data privacy perspective, I believe Viya uses a role-based access control system where users only see data and the relevant reports based on their role. Integration with LDAP and Active Directory for centralized authentication also helps. The data in Viya can be governed through column-level and even row-level security, which is critical for sensitive data. SAS also supports data masking to protect sensitive fields from being exposed. In my project, the AI or scoring capabilities of SAS Viya Platform largely meet expectations, but with some important context. It meets expectations concerning batch scoring, where the data for claim data produces consistent and reliably repeatable results. The real-time alerts also meet expectations, as the business rule based scoring in real-time for high-confidence, rule-driven fraud patterns. However, where expectations are only partially met is the rule-based versus true AI; my project is largely based on business rules rather than ML-driven, meaning it only catches known patterns. Evolving or emerging fraud behaviors that a trained model would catch are not captured, and there is no feedback loop. I rate myself a six out of ten on SAS Viya Platform. What justifies this rating is the hands-on experience, VI, real work on SAS ESP, and the SAS Viya Platform upgrade. I chose six because I believe in rating based on what I can actually deliver, not just exposure. The reason for the six is what I am confident in; I have production-level hands-on experience. I held back for a higher rating because seven or above means I can independently architect, administer, and troubleshoot any component of Viya, and I am not there yet. I have not worked with CAS in-memory or Viya version 4; my environment is version 3.5, so my knowledge has boundaries around the older release, including areas such as administration and platform security. I would rather give an honest six that I can back up with real examples than claim an eight and struggle to answer the follow-up.

We have integrated SAS Viya Platform with open source tools; we have built controls to integrate with our handling systems. The integration depends a bit on the version; I was not very happy with how it was done in 3.5 with the SWAT package, but I think it is better now in 4.0, especially concerning integration for R and Python. Assessing the usability of SAS Viya Platform's data visualization tools for non-technical stakeholders, our group has not managed to train others effectively, as there is a gap; they must understand the data they are viewing. Data literacy is too low in our organization; they may know their part, but they need to filter and integrate the data. We keep it mainly in the analytics field to build reports. It is not self-service, and the users must know how to filter data and understand what they are doing. We usually use classical metrics to evaluate the success of machine learning models on SAS Viya Platform, including logistic regression and tree models, focusing on recall, sensitivity, and specificity because relying solely on accuracy can be misleading with unbalanced groups. It is better to look at the groups individually. I give this review an overall rating of 7.

I advise others looking into using SAS Viya Platform to leverage the CAS engine effectively to get the best performance for larger data sets. It is important to have well-defined use cases before adopting SAS Viya Platform, whether it is reporting, advanced analytics, or machine learning. You should also invest time in learning the platform as there are initial learning curves involved. Proper planning of data architecture and integration with existing systems is crucial. Monitoring cost and usage is also important, especially with SAS Viya Platform. SAS Viya Platform is a very easy-to-use platform. I would rate this product an 8 overall.

I would advise others looking into using SAS Viya Platform to go ahead if their budget allows. I have rated this product an eight out of ten.