Navigating the genAI era: Opportunities and challenges

Real-time processing drives tangible impacts on business outcomes, enhancing fraud detection, personalized recommendations, and targeted marketing.

Alex Patino
Alex Patino
Marketing Manager
March 20, 2024|8 min read

Back in the day, people did manual backups on tape and then exported them manually to a location in a data center to create a backup copy.

Those were the days.

Data handling in the era of artificial intelligence (AI), machine learning (ML), and generative AI (GenAI) is a massive undertaking. As companies grapple with large volumes of data, experts at the “Real-time database to build your AI future” conference in Singapore, co-hosted by AWS and Aerospike, shared insights into how these technologies help organizations streamline real-time data processing.

Consolidating data silos

“A key strategy in handling large data volumes is the consolidation of data silos into a centralized data lake,” said Gururaj Bayari, senior specialist solutions architect at AWS. “This integration allows for efficient governance and access control.” This approach has eliminated barriers to accessing data, providing the agility needed in data models.

For example, Bayari recalled that each team or department had its own reports, which weren’t shared. “Even when it was shared, it was a CSV file or an export in an Excel file, or it had its own siloed reports.” But that made making change challenging. To make changes, IT had to get agreement from multiple stakeholders before even planning to deploy it into production. “So even if you want to add an additional field into that Excel or CSV export, or if you want to modify that siloed report, that would require almost a month’s lead time,” he said.

Moreover, because of the advent of visualization and automated analytical tools, consolidating data silos into a data lake means a department can get insights from the data without even requiring an IT department, Bayari said. “Your financial team can look at the cost for marketing, what the cost is for sales operations, and then what the returns we are getting are. What is the revenue that is generated?” he said. “Similarly, in real time, the marketing team can look into this data and see my digital advertising campaign – how much in sales opportunities are generated by this? What are the prospects that are generated? That’s the kind of insights we get by integrating all of these silos into one system or getting that single unified view of the data.”

Data virtualization: Access from anywhere

Data virtualization also took center stage, giving companies the flexibility to retrieve and analyze data from diverse sources without the logistical issues of physical data relocation. “With data virtualization, we can now view, interpret, and manage data from various sources without physically copying it to a centralized location,” Bayari said.

“Back in the day, what was relevant for a company was really the four walls of the data within the company,” said Aveekshith Bushan, Vice President for APAC in Japan at Aerospike. “But what we’ve seen over the years is that this has changed. Now, it’s not just the data within the four walls, but data that is all over the place. So agility of your data models becomes very critical.”

For example, Bushan found during COVID that the best predictor of where an outbreak would happen or how it would progress was not through government channels but through Google flu trends. “It had a better insight of where this would happen than any kind of government organization. That’s because search is an unbiased way of looking at it, and it could capture people’s behaviors,” he said.

Different ways of gaining access to data are also a factor, Bushan said, explaining how some telecommunications companies with which he was familiar were leveraging that data. One company used the geospatial location of its customers to upsell and cross-sell services to customers. At the same time, another has tracked the behavior of its users, including likes and dislikes, to customize the preferences of each of its users. “That can help you as an organization, regardless of which industry vertical you are in – what can you sell to the customer, what can you sell more – because you need to expand the larger bucket of offerings that you have,” he said.

Real-time data processing: Impact on business outcomes

The talk then shifted to the tangible benefits of real-time data processing and analytics on business outcomes, particularly in high-demand industries such as financial services, personalized recommendations in the e-commerce sector, and geospatially targeted marketing in the telecommunications industry. Real-time predictions have been instrumental for fraud detection in financial services, said Purushothama Shenoy, CTO at IBM Singapore.

In fact, real-time response isn’t just a benefit but a requirement, Bayari said. “Latency is very, very critical,” he said, particularly in B2C use cases. “I remember the days when people would say, ‘I run a report which is T minus two,’ ‘T minus three,’ sometimes ‘T minus seven’ as well,” he said. “Today, if you’re running a business that is T minus one and over, you’re probably not going to be relevant anywhere.” Bayari noted that when the Unified Payments Interface (UPI) was developed in 2016, large banks in India thought it wouldn’t be successful. However, with the advent of smartphone payment systems such as Google Pay, UPI took the lion’s share of the market. “Now, some of these large banks have no market share whatsoever in UPI,” he said. “Around 75% to 80% of the market is phone paying – no big banks at all.”

GenAI: Challenges and opportunities

While GenAI comes with challenges such as trust deficit, biased data, and intellectual property concerns, it drew attention to its potential to enhance productivity and customer experiences. Training models effectively on unbiased and diverse datasets can mitigate these challenges, Bayari said.

And what organizations are experiencing now with GenAI is just the beginning, Shenoy said. In five years, people aren’t going to a bank website for a transaction but will be doing it through a conversational AI assistant on something like WhatsApp, he predicted.

“Let’s take the banking example,” Shenoy said. “I want to look at the credit card which is best suited for me. So I give details. I’m a travel guy. I want travel benefits and all of those points. It should be able to give me a recommended card.

“Then I say, ‘Yes, okay, I like this card,’ but I don’t want to go to another website and apply for this card,” Shenoy continued. “I will continue my whole journey in the same conversational AI interface. I say, ‘Okay, I like this card. Apply for this card then and there, using the same conversational interface. Then I apply, and then I get my card.”

That’s where companies like Aerospike will play an advantage, Shenoy added. The application process will go to the backend, where a database like Aerospike will have a huge customer base with all the profile information ready and in a user cache for faster response instead of having to hit the backend each time, he explained.

But that’s not all. “We all know that once you get a card, you need to activate it,” Shenoy said. “Today, I might have to log into a bank’s digital service, give my card number, and click activate. But I will use the same conversational interface. I will say, ‘I received my card. I want to activate it.’ So, the whole user experience in a conversational AI interface with GenAI will now be much more experienced from a customer standpoint. That’s the future we’re going to have.”

Privacy considerations

At the same time, panelists warned that organizations need to be concerned about privacy regulations when using personally identifiable information in AI. The European Union passed the General Data Protection Regulation (GDPR) in 2016, and India passed similar legislation, the Data Personal Data Protection (DPDP) Act in 2023. Shenoy reported using metadata to determine how to mask or redact data before providing it to AI users to build a model.

A related issue is data sovereignty restrictions around the data’s geographic location. “We are also looking at federated learning, where we are unable to share the data, but then we still need to leverage the model and then the data built in it,” Shenoy added. “Indonesia has a huge data residency requirement. They don’t want to share the data beyond Indonesia. Still, there are some attributes that I want to leverage across APAC or Asia, so in those cases, we will look at federated learning as an option.”

Emphasizing real-time access and agility

As businesses integrate these technologies, only time will tell what new avenues will open up. “It’s an evolution right now,” said Bushan. “But I think the human element is still missing, which means the creative aspect of GenAI still has a long way to go – which is good news for all of us in terms of jobs and so on. The creative element is going to take time.”