Google's BigQuery ML needs big changes to compete

On the eve of Google Cloud Next '19, experts assess the impact of Google's BigQuery ML since its release and their hopes for the automated machine learning tool's future.

When Google first released BigQuery ML, it was accompanied by much buzz, but experts said the suite of SQL extensions has since been eclipsed by other automated machine learning platforms. Many are looking to the Google Cloud Next '19 conference in San Francisco to see what updates are in store for BigQuery ML and how Google plans to keep pace in an increasingly competitive field.

At one anticipated session at Google Cloud Next, the tech giant is set to announce a host of new models and features focused on improving and simplifying BigQuery ML's data science and machine learning capabilities. Google will be joined onstage by, a customer that is expected to elaborate about its migration to BigQuery ML and how it's using new models to assess data quality.

"I'm sure [Google] will focus on progress and try to highlight some enterprise-class customers who have migrated from rival deployments rather than developing new data warehouses from scratch," said Doug Henschen, an analyst at Constellation Research. "I'd also expect more ties between BigQuery and Google deep learning and data science capabilities."

Google did not immediately respond to an email requesting comment for this story.

In its online documentation, Google has touted a number of what it says are BigQuery ML's advantages over other approaches.

I feel like Google's BigQuery ML has fallen behind compared to some other players in the market like a DataRobot or an H2O Driverless AI.
Lynne BaerAnalyst, Amalgam Insights

Among them, according to Google, are enabling data analysts to build and run machine learning models using existing BI tools and spreadsheets; eliminating the need to program models in Python or Java; and making model development faster by removing the need to export data from a data warehouse.

Lynne Baer, an analyst at Amalgam Insights, said she is looking forward to hearing from, but still is somewhat skeptical that Google has advanced its platform enough to match those of competitors.

"I feel like Google's BigQuery ML has fallen behind compared to some other players in the market like a DataRobot or an H2O Driverless AI," Baer said.

Baer added that Google needs to collect more customer stories like's, including from users across more verticals. She said Google needs to make better arguments for using BigQuery ML in industries such as healthcare, financial services and government.

When BigQuery ML launched

Introduced in 2011, BigQuery, an enterprise-grade data warehouse, is one of Google's most mature cloud services. In July 2018, Google released its first beta version of BigQuery ML, new software attached to BigQuery. Google touted BigQuery ML as a tool that lets data analysts and data scientists build select machine learning models using standard SQL commands -- instead of advanced languages such as R, Python and Scala -- without having to move data across platforms.

Encouraging data analysts to use their existing SQL skills to start building machine learning models was -- and still is -- an exciting idea, Baer said.

"It seemed like a great way to help data analysts slowly upskill and slowly gain these data science and machine learning skills that were going to be necessary for doing bigger and more complex data science projects," she said.

Abhishek Kashyap, product manager at Google
Cloud, walks through the process of building a
machine learning model using BigQuery ML.

After BigQuery ML's release, Daniel Mintz, chief data evangelist at software vendor Looker Data Sciences Inc., said BigQuery ML likely wouldn't change how data scientists build machine learning models. But he said the tool makes it possible for data analysts who know SQL but are not as familiar with machine learning yet to start developing models -- which is valuable.

Henschen said, over time, BigQuery has impressed many customers with its "ease of administration, scalability and performance."

"The idea was to let customers focus on the business questions rather than the running of a data warehouse and, for many customers, it has delivered," Henschen said. Giving more people access to Google's data science tools with the release of BigQuery ML was a positive step in that direction -- and a selling point for the BigQuery platform as a whole, he added.

Where BigQuery ML stands now

However, since the initial enthusiasm about BigQuery ML, Google has been quiet about the product, Baer said -- with the vendor mainly announcing minor documentation updates in the intervening period.

"Those other products are drag and drop, they are much simpler to use and they're more powerful because they offer more kinds of machine learning model types than Google's BigQuery ML, which -- at least up until now -- has only offered simpler models like the ability to do linear regression and logistic regression," she said.

Another criticism of BigQuery ML is that data analysts are limited to pulling data hosted on BigQuery to build their machine learning models.

"You've got the cloud data warehouse providers like Redshift, Snowflake and Microsoft that are happy to provide data to whatever other machine learning products exist across the spectrum," Baer said.

Henschen said BigQuery has mostly attracted net-new deployments and workloads focusing on or involving data from Google sources -- not migrations, which is something Google is attempting to go after.

"BigQuery has its appeal, but AWS, with Redshift, and Snowflake have more aggressively gone after enterprise-grade replacements of legacy Oracle, Teradata and IBM Db2 and Netezza deployments," Henschen said. "They've done more to support the technical demands of data and workload migration from alternatives."

Whatever comes out of Google Cloud Next, it's not too late for BigQuery ML to overcome its alleged shortcomings and make a bigger market impact.

"It's Google," Baer said. "They have enough money to throw at the problem if they really want to fix it."

Senior news writer Chris Kanaracus contributed to this story.

Next Steps

Google Cloud launches Vertex AI managed ML platform with MLOps

Dig Deeper on Data science and analytics

Data Management
Content Management