The industry has realised that executing big data in isolation failed to deliver value, and that organisations instead need to think in terms of their data ecosystem.
A data strategy needs to encompass unstructured data, big data, and all the other types of data, he said. And a next-generation data strategy involves "the integration of... all of the useful data."
Where analytics processing commonly took place on premises, it is increasingly being done in the cloud, although government, financial and health data tends to remain on premises.
Few organisations have as yet gone all-in on the cloud, although cloud-first is a common strategy. Data gravity is thus moving towards the cloud, but an analytics strategy requires access to data that is generated and stored in both places.
Completely replicating data across two platforms is an expensive process, so Brobst favours either "surgical replication" of precisely the data required in both places, or the use of Teradata QueryGrid, which integrates multiple file systems and moves data on demand.
More generally, Teradata's Vantage platform separates compute and storage, allowing a single copy of data to be used with multiple analytics engines. "You get a much cleaner implementation" whether the data is stored in a Teradata file system, Hadoop, or a cloud-native object store, he told iTWire.
There has been a "stampede into native object stores," he said, because they are economically attractive and deliver good availability and throughput. But they are not suitable where low latency is required – in that case Teradata's technology (available on premises or in the cloud) is more appropriate.
Vantage's flexibility in this regard means organisations can optimise for specific requirements.
"Getting the architecture right is really important," he warned. Throwing everything into a data lake and leaving it for someone else to worry about later just creates a data swamp or a "data dumping ground."
Brobst pointed to Gartner research showing that 90% of data lakes have delivered no value because they lacked both architectural principles and governance, and other research that found 80% of data science work fails to go into production.
Organisations need strategies to promote data products (ie, models and insights that provide real business value), he suggested.