Ensuring Your Data Foundation Is Not a House of Cards
AI is witnessing a Cambrian explosion of innovation.
Companies and startups are now vying for new investment as they look to AI models to automate, streamline and solve human challenges. Even Gartner admits the surge in innovation in this article.
COVID-19 provided a booster shot for the interest in AI. The advantages are clear. AI models can unearth vital insights for business leaders looking to navigate uncertain seas while arming operational leaders to streamline critical processes and respond quickly to market changes. It also cuts down our reliance on human intervention — necessary for today’s operational restrictions — while freeing humans to add value in different ways.
AI’s ability to ingest data is already helping research, as seen in today’s vaccine advances and how we fight cybercrime. For these reasons and more, many consider 2020 a tipping point for AI.
Yet, the surge in interest is not creating a similar explosion of production-level AI solutions. Many remain stuck within departments as narrow AI implementation. A lot do not scale, and some become obsolete once the company changes its business processes or the market environment shifts.
So, why the gap between this enthusiasm and real-time delivery? A Cognizant whitepaper, “AI: From Data to ROI” highlighted key reasons and warned that deploying AI does not mean immediate high ROI.
Modernization argument
One major reason is data. “What fundamentally drives AI is data,” says Guruprasad Raghavendran, associate vice president for AI & analytics in ASEAN, Greater China and Japan at Cognizant.
Many would have heard the “garbage in, garbage out" mantra. But that’s only part of the problem. Another issue is the “monolithic data structures,” says Raghavendran. This is why he believes AI enthusiasts need to first consider data modernization.
“For me, data modernization is fundamental for AI. [Companies] that have not modernized their platform but still embarked on the AI journey have not been very successful. To fully realize the benefits of data on AI, you need to have foundational knowledge, and that comes for your data,” Raghavendran explains.
In fact, the whitepaper drew a strong correlation between AI maturity and data modernization. It cited nine out of 10 AI leaders saying that they are maturing or are in advanced data management stages. For those who are just beginning their AI journeys, the hardest lesson for 60% was having the right IT architecture and data modernization processes.
WATCH Guruprasad Raghavendran highlight the role of DataOps in data modernization.
Data inequality
In reality, monolithic data stores grew organically over the years. Ripping and replacing these stores or restarting the data journey from scratch is an expensive proposition.
Raghavendran believes you should not “boil the ocean” when modernizing your data. More important is to understand what types of data your AI needs. And the answer may not be so obvious.
For example, the whitepaper highlighted that the most easily accessible data sets are not enough to make the most intelligent decisions. Many companies rely on their first-party data, IoT streams, and social media to find out more about their customer behaviors. But in the future, geospatial, psychographic, competitive, and real-time data may become more critical.
It is one reason why the whitepaper predicts that companies will be pulling data from broader and more diverse data sets by 2023.
Another issue is data perishability. As companies use a more comprehensive selection of data types, they need to ensure the information they hold is current, accurate, and relevant. This means that some data have significant value at some time but may become irrelevant past a specific date — a potent issue for real-time or IoT data.
For these reasons and more, Raghavendran feels it is more important to have the right data platform. It can help you connect to these various sources and allow for data ingestion while allowing you to keep tabs on your data sets’ relevancy. Essentially, it puts you in control of your data stores and enables you to create a data journey or strategy.
WATCH Guruprasad Raghavendran explain why it may be important to get first-party data right, but it should stop your data modernization journey.
Scaling chasm
One significant benefit of using a data platform is scale — a major challenge for many AI initiatives. “Let’s say a small department piloted an AI program. To industrialize it, you need a platform to take care from data ingestion to [developing the AI model],” Raghavendran explains.
To create an agile and scalable data platform, companies need to move away from a monolithic architecture. “When you are in an [era] where data is the business, you cannot have a platform that consolidates multiple systems into one single, monolithic structure. [This is] because the speed and diversity of data changes a lot,” says Raghavendran.
Instead, companies need to build scalable, elastic, and secure data platforms. When you add AI and machine learning on top of it, the platform becomes a “platform of intelligence.”
However, this is not an easy feat. Companies are still feeling their way in their AI journeys, and it will be some time before we achieve the data platforms we genuinely need. “I feel we have a long way to go yet,” adds Raghavendran.
People power
Like everything in digital, data modernization requires the right people. Raghavendran sees people as an essential pillar for successful data journeys.
“But unfortunately, people skills are also a major stumbling blog for implementing an AI program,” he explains.
One problem is that companies seem to pay more attention to technologies than growing their talent pool. This is changing, says the whitepaper. Companies are now starting to shift their AI budgets beyond getting the latest AI technologies. Raghavendran explains that it includes hiring the best talent, training staff on AI, investing in external partnerships, and building a culture of collaboration between analytics teams and business units.
AI technology is also becoming more commoditized. New approaches to reinforced learning that rely on less data also allow companies to fast-track their machine learning initiatives. Off-the-shelf, pre-trained models for common industry use cases also enable companies to jump on the AI bandwagon.
“But you still need people to make all of this work. This is why I believe that companies need to focus on their AI talent right from the start.”
WATCH Guruprasad Raghavendran further detail why having the right people matters for AI success.
Data ethics matter
You’ve the platform and the right people, so what else is missing?
For Raghavendran, it is data ethics. And this is a hotly debated area that is still evolving at the moment.
The concerns about ethics are genuine. Algorithms can correlate data in a way that can unintentionally blur the boundaries of data privacy. While it may be simple to draw the ethical lines with human analysts, it is much more complex when you depend on machine algorithms to quickly ingest data from various sources. Already regulators are drawing a red line on explainable AI that is causing concerns about the future of deep learning using neural networks.
Another issue is that AI algorithms can also create data that can create personally identifiable information (PII), as highlighted in this article.
What is clear for Raghavendran is that companies shouldn’t shy away from data ethics. He feels that data ethics and data governance should be combined even though they have very different concerns.
“The whole ethics and compliance have to be part of the governance pillar. And that's fundamental for you to scale AI,” he concludes.
Winston Thomas is the editor-in-chief of CDOTrends and HR&DigitalTrends. He is always curious about all things digital, including new digital business models, the widening impact of AI/ML, unproven singularity theories, proven data science success stories, lurking cybersecurity dangers, and reimagining the digital experience. You can reach him at [email protected].
Image credit: iStockphoto/forgiss
