\\Decoding the Data Dilemma: the Key to Success in the AI Revolution
By Jim StrattonSVP, Chief Technology Officer, Product and Technology, Workday
We’ve long talked about the necessity of getting your data house in order to drive better and more effective decision-making, organizational agility, and employee engagement. But with AI quickly becoming a key driver of competitive advantage, the need for quality, timely data is more important than ever. That’s because good AI outcomes—whether that’s a recommended next-best action, some sort of anomaly or threat detection, or even a customer service response using generative AI—depend on plentiful high-quality data being used to train the underlying models.
Many of the generative AI use cases that have wowed the public in recent months—if delivered with sufficient safety and guardrails—will be truly value-additive to our ability to get work done.
But they will also quickly become commoditized. To differentiate yourself from your competitors, you’ll need to leverage your organization’s proprietary data to deliver the best-possible outcomes for your business.
Although data is the lifeblood of AI, reliable data for your own organization is all-too-often hard to find. In fact, among the respondents in “AI IQ: Insights on Artificial Intelligence in the Enterprise,” a study sponsored by Workday, 77% are concerned that their organization’s data is neither timely nor reliable enough to use with AI and machine learning (ML). Similarly, insufficient data volume or quality was the top reason (29%) for their AI and ML deployments falling short of expectations.
Just stop for a moment and let that sink in: the vast majority of organizations don’t fully trust their own data to provide the best-possible AI outcomes.
A big driver of this is the sheer number of applications that the average organization uses. All of these applications use and produce data. A recent Accenture report found that the average company uses more than 500 applications from multiple vendors, and 80% of the survey respondents say they will buy more applications from additional vendors within the next 2 years. It’s a dilemma. You have vast amounts of something very valuable, but have a hard time getting all of it into a timely and reliable form. It’s safe to say that this data dilemma is not going to be solved by most companies anytime soon.
Overly complex technology portfolios stifle value when it comes to data collection and curation, but complexity can also make it hard to get all of your data into one spot to feed into AI algorithms. Each one of these applications is a data silo that must be integrated, curated, governed, and secured if you want this data to fuel the best-possible outcomes and insights.
The integration of disjointed systems often comes with a heavy development and maintenance cost, and by achieving consistency across these systems you are by definition sacrificing the timeliness of the unified dataset. This sacrifice leaves an outdated view that only shows how your company was running, not what is happening right now. If siloed data fosters risk and unreliability, simplifying the data domain via modern platforms offers hope. Key anchor systems that support the enterprise—core platforms such as CRM, human capital management (HCM), financial management, inventory management, and more—can help to reduce security risk and data silos.
To solve the AI data dilemma, start with the business outcomes and insights you want to deliver. Then, and only then, should you start to identify what data will drive those business outcomes and insights.
Many companies very often start with the data and then try to use that data to drive insights. That’s a backwards way of doing things, wastes time, and doesn’t deliver business value.
Once you have the business outcome in mind, the solution to getting the most out of the data you do have—ideally in a unified platform versus a mosaic of systems—is to treat that data like a product in your own company. Explicitly define an owner for that data and a service-level agreement (SLA)—how reliable it is, how timely, and so on. Once you have done the engineering work to make that data available, it’s important that it’s available for the whole enterprise—not just the team that created it, owns it, or was the first to ask for it.
And remember, it’s a myth that quantity is the be-all-end-all of robust AI. It’s our contention, and our experience, that data quality scales AI better than data quantity. It’s also true that data quality, at the scale needed for AI, is harder to get to than sheer quantity, but is fundamental to reliable, responsible, and useful AI in the workplace.
It’s important to note that the transformative nature of AI is likely both over-hyped in the short-term, and under-hyped in the long-term. There are use cases that we can’t imagine yet that will create new industries, business models, and ways of working. The one thing that won’t change, though, is the need for reliable data. Those who do the hard work now of getting their data house in order are poised to reap the benefits of whatever the future holds.
Very confident—we have quality and reliable data.
Somewhat confident—we still face challenges.
Not confident—data quality and reliability are issues.
The Tech Stack Battle: Embracing Data-Driven Insights for Victory
How Workday Is Leading Enterprise Generative AI
Accenture Report: Enabling Technology Interoperability for the Future of Work
Closing the Software Development Gap with Low-Code Applications
IDC Report: What Digital Business Models Require
Harnessing the Public Cloud for Ongoing Innovation
