GenAI evolving, remains dominant data and analytics trend

Generative AI is the single dominant trend in data management and analytics. Nothing else is even close.

It's been that way since OpenAI launched ChatGPT in November 2022. The technology marked a significant improvement in the capabilities of large language models (LLMs) and showed the transformative potential of GenAI in the enterprise.

One key possibility is to be assistive in nature, with GenAI-powered natural language processing enabling virtually any employee -- not just data scientists and other analysts -- to use business intelligence tools to inform decisions. Due to the complex nature of most data management and analytics platforms, computer science skills, statistical expertise and data literacy training were all prerequisites before generative AI reduced those obstacles.

Another key possibility that has made generative AI such a singular data and analytics trend is exponentially improved efficiency. Generative AI applications can be trained to be agents unto themselves that take on time-consuming, repetitive tasks that data engineers and other experts previously needed to do manually.

But while GenAI first became a major trend because of its potential, it is evolving.

ChatGPT's launch was closely followed by the development and release of a spate of competing LLMs. Initially, the technology's transformative capabilities were theoretical. Now, they are becoming a reality, according to Yasmeen Ahmad, Google's managing director of strategy and outbound product management for data, analytics and AI.

Now, vendors including Google are developing generative AI tools to better enable customers to use their platforms to build GenAI models and applications. Enterprises, meanwhile, are taking advantage and creating pilot models while going through the proof-of-concept phase.

Generative AI, however, doesn't exist in a vacuum. As a result, enterprises are emphasizing complementary capabilities such as data quality and data governance, which aim to ensure that the information feeding and training GenAI can be trusted. In addition, real-time data and automation are key to making sure that generative AI isn't a reactive technology.

Yasmeen Ahmad

Ahmad recently took time to discuss why generative AI has been such a pervasive trend in data management and analytics, including its assistive and agentive nature as well as its potential for unlocking unstructured data that has long been difficult to operationalize.

In addition, she spoke about other data management and analytics trends and how they are complementing generative AI to advance what enterprises can do with data.

Editor's note: This Q&A has been lightly edited for clarity and conciseness.

Generative AI has obviously been a major trend over the past couple of years. Is it accurate to say it's been the top trend in data management and analytics?

Yasmeen Ahmad: One hundred percent. Generative AI has been a massive trend across multiple dimensions. Generative AI is a fundamental technology that is truly transforming the way data platforms are being built, the way data platforms are being used -- a lot of enterprise data has been dark -- and then generative AI is changing the way that humans are working. It's transforming their experience. It's a big trend because it's so multifaceted and multilayered in the impact it's having across all these different dimensions.

Generative AI is a fundamental technology that is truly transforming the way data platforms are being built, the way data platforms are being used ... and then generative AI is changing the way that humans are working.

You addressed this in part in your last answer, but to delve a little deeper, what does generative AI enable that makes it such a dominant trend in data and analytics?

Ahmad: Generative AI is a fundamental technology transforming the landscape of data management and analytics in two very important dimensions. First, what we see from organizations is that 80% to 90% of enterprise data today is unstructured. It's PDFs, documents, images, videos. That is data that has not traditionally been analyzed. We didn't have the tool set. Generative AI is the tool set to unlock multimodal data that previously was inaccessible. That, in itself, opens up new insights, new use cases that weren't possible before.

In addition, combining multimodal data with traditional structured data adds to and enhances traditional analytics. Gartner reported that 66% of enterprise data is dark data. Generative AI is eliminating that dark data.

What are some of the use cases you alluded to that weren't previously possible?

Ahmad: We have a customer, Symphony Communications, that is using generative AI with call center transcript data. It's audio data that previously might have been tagged manually to capture some data to do sentiment analysis. Now, with audio transcripts, they can get deep, rich, meaningful insights by analyzing the words. They can generate responses that a call center agent can read to a customer live, in real time. Beyond that, they can get much more nuanced to understand what customers are talking about, what the sentiment is. They have the ability to do translation on the fly. They have all of this rich analysis that wasn't previously possible. That's one example.

Another is HCA Healthcare. They're using Google's BigQuery and BigLake multimodal data foundation to bring together traditional structured patient data with documents and notes from clinicians and physicians and with image data from X-rays and MRIs. Traditionally, they analyzed the structured data to look for trends in their patient population. What they were not able to do was bring together that rich data you get with physician notes and with images to really do analysis around diagnoses and look at patient healthcare. They're using generative AI with traditional models to improve healthcare in a way they just weren't able to before.

Where are enterprises in their generative AI development cycle -- are they still in the idea stage or have they moved to the development and production stages?

Ahmad: We are seeing very fast innovation in the generative AI space with customers accelerating through the exploratory phase of pilot testing and proof-of-concepts to getting into pilot production. With many generative AI use cases, we still see a human in the loop -- they're not fully automating the generative AI technology. But they are putting it into the hands of their businesspeople, business users, to drive outcomes. I've never before seen this pace of innovation with a new technology.

What makes that pace of innovation possible?

Ahmad: The key is that generative AI isn't a new technology where you have to start an entirely new data platform or ecosystem. The way it's being built is that generative AI models are being integrated into existing data platforms, the ability to run a large language model over existing data. There's an ability to tap very quickly into low-hanging fruit. Then, as customers are maturing, we're seeing 600% year-over-year growth in using multimodal data.

There's a gravitational pull toward bringing more multimodal data sets to expand the use cases they were doing initially. There's a lot of exploration happening to understand its potential. We're seeing massive exploration as customers understand how this technology fits into their landscape, how it's going to transform their business. And today, there's lots of human-driven generative AI, but we're already seeing that in the future it's going to be generative AI assisted by humans.

There's been a huge amount of buzz around generative AI that's made it such a big trend in data management and analytics -- is it living up to the hype?

Ahmad: Technologies are typically overestimated in the near term and underestimated in the long term. That analogy absolutely applies to generative AI. There's massive amounts of hype and energy around what it can do that's now being broken down to figure out how to get to business results. But the long-term implications of this technology are transformational. I lean into the idea that it is as big as the internet or mobile phones in terms of the transformational impact it can have on so many parts of everyday life for us as humans and consumers and patients, and also for businesses in the way that they operate and meet the needs of those humans, consumers and patients.

As generative AI moves beyond hype and more enterprises develop pilot models, what are they discovering about the reality of generative AI development?

Ahmad: For enterprises, doing initial use cases and getting to insights in pilots has been great. But generative AI is shining a light on the data platform. The challenge is no longer on having an AI technology -- generative AI has made that easy. The challenge is how to make sure there's a trusted AI-ready data foundation. With generative AI, the efficacy of models is linked to data, the quality of high-volume data needed for training, for tuning, for RAG [retrieval-augmented generation].

The No. 1 conversation we're now having with customers is about making sure their enterprise data is ready, it's trusted, it's governed, especially as they bring together multimodal data foundations. They want to govern all that data the same way they governed traditional structured data, so they need a single access control and governance pane across diverse types of data, and they need to easily use that data for LLM training, tuning, RAG and prompting.

We've spoken extensively about generative AI, so what are some other major trends in data management and analytics?

Ahmad: Still related to generative AI are the ideas of assistive and agentic experiences. Data governance and data quality are top concerns. Often, the No. 1 thing C-suite executives talk to me about is data governance and how to trust their data. What we're seeing is that generative AI can support with that. It can be an assistive technology to understand data drift, finding data anomalies and even building and generating metadata and semantics. And we know semantics are important when training generative AI models because semantics give models context about a business, the language of a business, and help generative AI give more accurate and precise answers.

Traditionally, a human had to build all those semantics and curate all that data and manage the quality. We're actually applying generative AI to manage that problem because it has the ability to generate semantics by looking at the data, looking at relationships. We see generative AI as a massive accelerator for data engineering teams that had a lot of human toil.

If that's the assistive nature of generative AI, what is the agentic?

Ahmad: It's the notion of data agents that operate on the data analytics lifecycle and transform the experience. Rather than a human coming to data and asking for an insight or for data quality to be improved, a data agent is monitoring data, looking for anomalies, surfacing insights, suggesting semantic modeling metrics to monitor. We're moving from a reactive world to one where generative AI is proactive in supporting the data analytics lifecycle. The agentic world is supported by the evolution we're seeing with LLMs.

As we evolve LLMs, it's not just about the size of the model increasing, the parameter size increasing. Of course that's improving quality, but at the LLM scale, we're starting to see emergent capabilities where they're able to reason much better, understand causality. That leads to LLMs being able to reason and understand if the answers it's giving are 100% accurate and whether there are nuances to the answers. They're getting better at evaluating their own answers. That's what will power a more agentic future where an organization will have data agents that essentially power the enterprise.

What about other data management and analytics trends -- what else are you seeing enterprises emphasize?

Ahmad: Two others we're seeing are real time and automation. When you bring generative AI together with real time and automation, now you can truly deliver the transformation businesses are looking for. With digitalization, businesses were able to get much more data about themselves. Now that we're in a world where more data is being captured, the next evolution is to use generative AI for intelligence with real time to be able to generate outputs and action them in real time. So, we're seeing an uptick in streaming. Historically, we were feeding real-time pipelines with aged insights. Now, [enterprises] can run machine learning and LLMs over real-time streams of data and pipeline out real-time actions. There's a flywheel of getting real-time data in, running generative AI and automating it all. It's that true transformation that businesses have been waiting for.

Great innovation comes from bringing together diverse pieces of technology that together create more innovation, and this feels like a moment when pieces of technology are coming together that can drive transformation.

What's the timeline for converging those technologies to transform business?

Ahmad: We have the entire stack from the foundational technology layer to the LLMs to the data platforms with real-time streaming data. The integrated stack exists today. Over the last two years, we've placed a heavy emphasis on unification and simplification because to power this transformation, you need unified platforms and simplified, integrated technology. Google has an open ecosystem, so there are integrations between Google technologies plus integrations with partners. Integration and unification is a key pillar. That foundation is needed to build a transformative world.

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.