Snowflake Doubters Voice Reservations Over Data Warehouse's Attempt To Break Into Financial Services

Published date: .

Cloud-based data warehouse darling Snowflake has launched its latest venture into financial services, while Teradata, something of a stalwart in data warehousing for banks and insurers, is attempting to broaden its appeal with machine learning implementations.

Though the world has been focused on Snowflake's stratospheric rise – it went from $1.5bn value in 2018 to $120bn shortly after its IPO last year – it may struggle to make inroads into the lucrative financial services markets, according to insiders and industry experts.

The so-called cloud-native data warehouse biz launched what it calls the Financial Services Data Cloud this week, accompanied by the claim that 57 per cent of Fortune 500 firms in the sector are on its platform. It is described as an industry-tailored platform that brings together Snowflake technology with "partner-delivered solutions" and "industry-critical datasets."

The idea is that financial services companies use it to help launch new customer-centric products, with the process made easier by having all the data already on the platform, so the company says.

Snowflake says companies will be able to use built-in security and governance capabilities to collaborate on data projects. Features include private connectivity for multiple public clouds, enhanced encryption with bring your own key (BYOK), built-in classification and anonymisation of sensitive data, and integration with third-party token providers, in compliance with SOX standards.

The platform offers the ability to securely share data across multiple public clouds with support for sharing from multi-tenant environments. A data catalogue is set to come via Alation. Meanwhile, there are a whole bunch of features coming via partners, including investment management company BlackRock's Aladdin Data Cloud – a Snowflake-powered system designed to help investment managers make more use of data – SI Cognizant, data integration platform Dataiku, and Deloitte.

Matt Glickman, Snowflake veep, said the company's platform was becoming the "go-to destination for traditional and alternative datasets" in financial services while third party tech outfits were also helping customers build new services with the system.

He also claimed that, working with service companies, customers in financial services were putting "more and more of their key production workloads in a Snowflake."

This might surprise some cloud-wary observers. One insider with years of experience working with data warehousing systems in financial services said banks and insurers were struggling to see reasons for moving their trusted on-prem systems to the cloud because security, performance, and cost were still issues.

While Teradata has built up a reputation in financial services, where it counts HSBC and Lloyds among its customers, Snowflake is new to supporting core workloads in this market.

The insider said Snowflake can handle eight concurrent users per cluster, meaning when the system adds more users, cost increases.

In Teradata, "although the current system is a known finite capacity in terms of throughput, compute and storage, that is a guaranteed known cost. It's the unpredictability of the costs that are a big shock to people [using Snowflake] because not only can they not explain this month's bill, they've no idea what the next month's bill will be," he said.

In addition, technical teams in the financial services industry were sceptical that Snowflake had anything new to offer, based as it is on a relational database. "There's no wiggle room left to move the state of the art forward," the insider said.

Although Snowflake had helped on-boarding and arguably data sharing, it was not solving a real problem. "What is the technology enabler in the in this announcement, what's the new IP that is being brought into play?" he asked.

An analyst at a global technology research firm, who asked not to be named, agreed Snowflake was struggling to make inroads into core enterprise data warehouse systems in financial services.

Although they might do some tactical analytics and data processing in Snowflake, banks and insurers rely on features Teradata has built around its on-prem data warehouse appliances for 40 years.

"Teradata has integrated machine learning [for performance] and integration of multiple data types: things Snowflake haven't engineered yet. If you've got a big bank that's reliant on those features, then migrating to Snowflake on the cloud is kind of difficult, and from a cost-benefit point of view it would be difficult to argue now Teradata also has a cloud system. I don't think they will deal a knock-out blow to Teradata any time soon," he said.

Meanwhile, Teradata has announced new features to deploy machine learning models for platforms such as data lake spinner Databricks directly into its own.

Scott Toborg, Teradata director of product management, told The Register that, for example, a logistic regression model a data scientist had built in Apache Spark could be exported in the Predictive Model Mark-Up Language (PMML) interchange format, with coefficients, biases, and any other parameters that are used to describe that model.

"Then we import that file over to Teradata, extract the data and then create the Java code necessary re-execute that model," he said. The industry analyst said the combination of Databricks (Spark) for building machine learning models and Teradata for deploying them on business data could be sophisticated.

Despite the investor limelight shining on Snowflake for the last couple of years, insiders say it has a long way to go in terms of convincing core enterprise data warehouse users to migrate their main systems, although it is picking up peripheral and tactical workloads.

Incumbent vendors like Teradata have their own cloud story. For example, Unilever has migrated its on-prem Teradata system to Azure over the last 18 months, according to a customer testimonial.

All the while Teradata continues to add features, meaning the battle in enterprise data warehousing is far from a done deal. ®