Artificial intelligence (AI) will play a growing role in unlocking the value in enterprise data, according to Google Cloud’s lead executive for data analytics.Gerrit Kazmaier, vice-president and general manager for database, data analytics and Looker at Google Cloud, told Computer Weekly that the cloud and search giant’s customers are already combining AI with more conventional business intelligence tools.This is because AI helps bring together structured and unstructured data, said Kazmaier. AI systems are starting to perform increasingly complex analysis, but they can at much faster speeds and with much greater volumes of information than human experts.Google is supporting its customers in this by drawing on its background in search, as well as its cloud resources and its experience developing the transformer model, one of the foundations of generative AI systems.“We are reimagining, let’s call it, the Google search for enterprise data,” said Kazmaier. Much of this is about combining the potential of AI tools, including generative AI, trained largely on public data, with the domain and business specific information held in businesses’ enterprise applications and data lakes.“So far, Google search is active mostly in the public domain or the public web,” he said. “There was ultimately a big opportunity of bringing this to the enterprise domain, basically giving every data point that exists in companies, which are not part of the worldwide web, a similar interface.“Everyone knows how to use Google. Every CEO on the planet, I’m confident, knows how to use Google to search the public web. I’m equally confident that only a very small number of persons on this planet and certainly a small number of CEOs would be able to use a dashboarding tool for themselves to find information about their own enterprise.“With generative AI [GenAI], we have the opportunity to talk to your enterprise data, as you can talk to public data via Google search.”
Google ‘gets’ data
Google has a “cultural understanding” of the need to make information more accessible, according to Kazmaier. This is at the heart of its mission to bring AI and conventional analytics together.
“From a technologist point of view, it starts with searching the world’s information and making relevant information universally accessible and useful. That is required to build technology, which is heavily used today in generative AI,” he continued.
“There is a reason why Google was the original inventor of the transformer model, which is now the underlying architecture of all of these models be it Gemini [formerly Google’s Bard], or ChatGPT, [Meta’s] Llama and so on.
“There is a deep understanding first of all, when we say that we want to map someone’s question to a meaningful answer, about the technology that we need to build to understand the semantics for processing that efficiently, and to give it back in a form factor a human can work with.”
Google has set out a roadmap to build AI into its analytics tools, integrating BigQuery with Vertex AI, enabling data to AI workflows in BigQuery Studio and allowing users to create machine learning models in BigQuery ML and export them to Vertex AI, as well as adding features to Looker and Looker Studio.
In Google’s view, one of the applications for generative AI in the enterprise with the most promise is helping non-specialists interact with business data.
Rather than learning coding or analytics skills, or to write queries and design dashboards, GenAI should allow business users to interact with a database, data warehouse or data lake application using natural language – and to get a response in natural language too.
This has two key advantages, aside from ease of use.
It removes the need to filter data to match the format and capabilities of a dashboard. This inevitably means some information will be truncated or removed. And only a minority of enterprise users have the skills to drill down into the analytics tools themselves.
An AI-based system has the potential to be more accurate as it can deal with larger data volumes, and a broader range of data sources. Kazmaier referred to this as “wide data”.
The other advantage is that users can interact with AI-driven systems in a more iterative way. They can fine tune and tweak queries, asking further questions until they find the information they need.
Kazmaier cites the example of Camanchaca, a seafood firm in Chile that is using a suite of standard BI tools, including BigQuery, Vertex AI and Looker. It created an AI agent to give all employees access to the company’s data.
“This unlocks data and analytics for the non-data analysis professional. Everyone has a question to ask. Not everyone has an analyst to answer that question,” he said.
“There are these new use cases emerging for generative AI capabilities, which give us more than dashboarding and traditional data analytics. The consumer is changing, from the data analysts, now to every knowledge worker being given access to meaningful data analysis.”
This allows business intelligence to move from simply displaying data to interpreting information, in the way a human analyst would, according to Kazmaier.
“When you look at data you want to have someone knowledgeable, like a professional analysis, to help you interpret that. What does that represent conceptually, or how does that compare?” he said.
“That’s not a question that is necessarily answerable by the data point itself, but you need to someone really calibrated if you will, who understands how to interpret, ‘Is this a good or a bad margin? Is this a good or a bad, day’s sales outstanding?’.
“This can be trained and encoded and is generated by the agents that we are introducing in our BI offering. So, basically, you are collaborating with an analyst that can help you to understand and to interpret the data that you’ll see. One of the key problems that we have is traditional BI is that we have to compress information to a level that becomes human comprehensible.”
According to Kazmaier, the consumers of data are changing. More users want access to data, and AI – especially generative AI – offers a way to open up that access in a way conventional BI cannot.
But there is more to the integration of AI into business intelligence, and into Google’s roadmap, than simply providing a better interface. AI offers a way for firms to stay ahead of the seemingly endless growth of enterprise data – and hopefully drive some business value from it at the same time.
Kazmaier talks about “wide” rather than big data: not just having more data, but adding more data points to analysis. AI systems are well placed to decide if it is worth taking additional factors into account, he said, and they have the processing power to do this fast enough, as to not hold up decision making.
“One of the biggest changes that we have seen is the use of unstructured data,” he said. “If you think about it unstructured data, roughly, represents 90% of the world’s data. Traditionally, this data has not been used in data analytics. There were specialised applications for documents, or for automating certain processes like paying invoices, but it has not been considered a part of an enterprise data landscape that we actively use, explore and analyse, like you do with structured data.
“With generative AI, working with unstructured data, people understanding it and extracting information from it, becomes enormously flexible and available,” he continued.
And AI tools allow business users to dive deeper into the data and better understand the trends in their organisations: moving from “what, when and where” questions to, ultimately, “why”.
“You have large models being trained on public data, and you can ask them about public domain questions and it’s amazing what it can do,” Kazmaier added.
“But these models are not being trained to use an enterprise’s data, and that’s quite interesting. How do we deploy these large [language] models with enterprise data so you can open up all of the insights that you have to your data, so all of them are of use in the company?”
AI agents, he said, are already providing those answers.
