WinterCorp: The Data Platform and Architecture Experts for the Age of Generative AI

Recent advances in artificial intelligence (AI) have opened vast new opportunities for companies to benefit from data and analytics. In this age of generative AI and machine learning (ML), we can produce new insights and take extraordinarily valuable business actions that were never possible before.

But there are challenges on the path to those great business advances. Foremost among them are the enormous resources that AI can consume. Typical estimates are that training a large language model will cost $4 million or more — and most companies will need many such models. Further, most models need to be tuned up on a regular basis to keep them in production — and these tunings can also be very expensive.

No company can afford to ignore the breakthrough business results available from AI. But, no company can ignore the cost implications, either.

It turns out that the key to managing the costs and the time to value — for both classic and generative AI models — is the data platform. At enterprise scale, the data platform is where the data used by the AI is stored, harmonized, prepared, and managed. AI algorithms often consume large volumes of data, interacting intensively with it in complex patterns. This is a significant challenge for the data platform: some platforms will perform well at large scale and others will work well only at small-to-moderate scale when executing AI and ML functions.

Further, customer needs will vary in complex ways. So, a data platform that supports the AI operations of company A may not work well for company B; it’s a matter of finding the best fit. But with a data platform well suited to your needs, both your AI models and your other analytic workloads will run at very high efficiency, even at large scale. Then you will be positioned to compete in the new AI economy.

Data Platform Assessment/Selection

The standard approach to selecting a data platform emphasizes factors that are of secondary importance when implementing AI at enterprise scale.

As Richard Winter says, the WinterCorp approach to data platform requirements is different because it begins with a quantified estimate of the customer’s unique requirements. This places the appropriate level of importance on cost, performance, and scale — factors that are essential to success in the enterprise. These estimates are based on the customer’s business interests & plans, AI & analytic use cases, present & future workloads, database structure, and industry & technical trends. WinterCorp takes expected growth and changes in strategic data requirements into account. The resulting evaluation, as well as any needed tests or proofs-of-concept, therefore account for the scale and complexity that the customer actually needs to meet key business objectives.

WinterCorp data platform assessments and evaluations are often enhanced by the WinterCorp Cloud Data Warehouse Lab, in which Dr. Norbert Kremer leads the design and performance of live tests and proofs-of-concepts of data platform performance, scalability, and cost. WinterCorp has in depth experience with advanced analytics, machine learning, and AI as these relate to the data platform.

Many customers make the mistake of choosing a data platform based on brand, standard practice, or feature lists; by contrast, WinterCorp conducts an engineering evaluation so that the customer’s needs for performance, cost control, scale, and data availability are satisfied through a conscious strategy backed by fact, analysis, cost models, and test results.

AI and Data Science Traditions

Most data scientists do their work outboard of the data platform, using AI or data science toolsets that run directly on the file system or on a platform supplied with the toolset. This approach is fine on a small scale but breaks down at the scale needed by larger enterprises.

WinterCorp has seen cases where an outboard AI process was migrated to run inside the data warehouse, resulting in a reduction of time and cost of over 175 times. The original process took 35 hours; when moved inside the database, it took 12 minutes. This example shows why it is so important to have the AI architecture built around a scalable data platform. However, the customer would not realize this large gain with every data platform; the data platform has to be efficient and scalable for the pattern of access generated by the AI process.

Most data scientists and AI specialists do not expect to use this type of architecture. However, companies running a substantial amount of classic machine learning are increasingly moving in this direction. Richard Winter believes that generative AI will intensify this trend because of the extraordinary cost and time involved in training and running genAI models.

Scalability Assessment

One of the scariest things that Winter says he sees is a customer investing heavily in AI without an affirmative plan to address performance, scalability, and cost. Often, the target architecture is created — or the target platform is selected — without concern for these issues. According to Winter, an executive with a stake in such a project needs to ask, “How do we know that our service levels will be met? How do we know that we can fund this initiative in production  or earn a return on our investment? How do we know that we have a feasible solution with this data platform or AI architecture?”

If there is no convincing quantitative answer — if there is no test result, analysis, or model to show that the key hurdles will be cleared — then the executive should insist on an independent review to assess the situation. No company should be in the position of relying on hope or vendor claims when a large investment or a critical business objective is at stake.

It is possible to measure, manage, and control the engineering risks in a large AI or analytic project, and the necessity to do so is not diminished by the adoption of a new generation of technology.

WinterCorp can provide an independent assessment of the performance, scalability, cost, and other engineering factors for a data platform planned, under development, or even in production. The findings and recommendations from such a review have helped many customers avoid severe problems and achieve critical business goals.

About the CEO

A specialist in the technology and implementation of analytic data management at scale, Richard Winter advises clients on data strategy and data architecture, focusing on the AI and advanced analytics in the modern data platform. He has been retained to make architecture and platform recommendations or perform engineering tests for more than 50 leading enterprises, government agencies, and technology vendors.

In 1992, Winter founded WinterCorp, one of the few consulting companies in existence that is focused entirely upon data management at a large scale. He currently serves as the CEO and principal architect of WinterCorp. He has been an expert in the technologies of the analytic data platform, commercial and open-source, relational and non-relational, cloud, on-premise, and hybrid, streaming, and batch. Winter also serves on the faculty of TDWI, The Data Warehouse Institute, where he teaches courses on the architecture and scalability of cloud data platforms and their uses in advanced analytics and AI.

“Unlike other consultants, WinterCorp conducts an engineering evaluation so that the customer’s needs for performance, cost control, scale, and data availability are satisfied through a conscious strategy backed by fact, analysis, cost models and test results.”

“No company can afford to ignore the breakthrough business results available from generative or classic AI.  But, no company can ignore the cost either. Thus the need for WinterCorp’s analysis, modeling and testing for the performance, scalability and cost of the data architecture is more critical than ever in this age of AI.”

“We have architected data lakes and databases for startups to major enterprises. We know both new and established products and technologies and can satisfy the world’s toughest requirements.”