The concept of data-driven leadership isn’t new. Executives and other business leaders have been advised for many years on how to interpret data and balance intuition and experience with data to make the best decisions for their organization and its employees.
Now, with artificial intelligence (AI) and machine learning (ML) in the mix, there are different considerations for leaders using data analysis to help inform decisions and strategies. In concert with AI, the kind of data that leaders will be relying upon is largely unstructured data. Rather than bar charts and spreadsheets, they will be analyzing documents, images, audio, video, sensor data, social media posts, text and chat data, and more—spread across the enterprise. That increases the complexity considerably for leaders, and their methods will need to adapt accordingly.
As AI tools become more available, easy to use, and embedded within everyday applications, it’s getting easier to incorporate analysis into daily decision-making—to a point. AI has unearthed widespread risks with false, erroneous, and libelous outputs that can have harmful consequences. AI can create havoc if inputs and outputs are not managed and carefully evaluated by the human eye as valid. Zillow had to cut 25% of its workforce after an ML algorithm for home pricing made an error that led to a $304 million inventory write-down in Q3 2021, according to CIO.com.
Business and IT leaders need new tools and processes to ensure the safe and ethical use of AI for decision-making.
5 STEPS TO GET READY FOR AI-DRIVEN DECISION-MAKING
How do leaders navigate this new AI-driven landscape for decision-making, and what do they need to know about both AI and unstructured data to be successful and avoid unnecessary risks?
1. Data Analysis
Before you can leverage corporate data for AI, you need to understand what you have. This begins with the ability to view and analyze data across a hybrid cloud infrastructure—from your data center to the cloud. You’ll want to gather basic metrics such as common file types, data volumes and growth rates, top owners, and duplicate and orphaned data.
Before you do anything, clean up this data. Get rid of what you don’t need because it is consuming expensive storage and increasing risks for a ransomware attack. Indexing all the unstructured data is central to enabling efficient search and use of the data for AI.
2. Data Classification
Once you’ve cleaned up your unstructured data estate, it’s time to classify it. This way, data scientists, analysts, and other stakeholders can more easily find what they need rather than asking IT to hunt through directories or manually inspect folders.
By using AI tools to analyze files and enrich the metadata (the data about your data) with tags describing the file contents, you can help departmental users curate discreet data sets for their AI projects. Having a global file index where you can store the output of this AI preprocessing is critical because otherwise you will have to repeatedly run AI on the same data for different users, which can get very expensive.
3. Data Governance
Data security and governance for AI are top concerns, as found in my company’s 2024 State of Unstructured Data Management and other industry surveys. You need the ability to restrict certain data sets from AI—highly sensitive data such as customer and employee records, financial data, R&D data, and more.
Increasingly, given the petabyte scale data environments that organizations retain, you will need automated data management tools to segment and classify data and orchestrate data workflows to AI, including auditing results for accuracy and potential liability.
4. AI Overload
With so many AI tools and so much data to ingest and analyze, it would be easy to quickly overwhelm business leaders with data points. Even worse, bias could be unwittingly propagated when the wrong data is used. IT and business leaders will need to come together to agree upon the organizational goals for AI, which use cases are most valuable today, and which subset of tools can meet those objectives.
First, executives may need training on how to use AI safely: Generative AI outputs, for example, must be reviewed and assessed for accuracy, bias, and completeness. Ultimately, IT should be responsible for understanding how AI is integrated into existing applications that leaders use, and what risks might be lurking underneath the covers.
5. Assessing AI Outputs For Validity
Organizations need tools and practices to weed out false and libelous outcomes. Some human oversight of AI will be necessary for the foreseeable future; false positives are bound to occur. Establishing AI governance processes which integrate human review and input will be part of the oversight. If business leaders get burned by using AI because the results make them less credible or even culpable, AI won’t achieve its potential and may be outlawed altogether—hurting potential competitive advantage.
It’s still early days for AI data ingestion and decision-making, but the pace of AI infiltration into society is faster than any other technological revolution to date. IT and business leaders will need to come together quickly to manage data risk and make AI safe and easy to use. Without these guardrails, AI won’t deliver upon its many potential benefits.