Imagine you buy a new self-driving car, the autopilot fails, and you slam head-on into the vehicle ahead. That’s what happened in several incidents with Tesla’s Autopilot system. Due to insufficient data, its sensors failed to detect white vehicles on sunny days. It mistook the color for a lighting condition.
You can’t always blame AI. Sometimes it’s the data.
AI doesn’t think. Instead, it predicts by using the data that it is given. When AI generates false information, it can be because it’s filling in the data gaps.
Poor quality data costs organizations an average of $12.9 million annually, according to Gartner. IBM estimates that bad data costs the US economy around $3.1 trillion annually. Let’s take a closer look.
Garbage In, Garbage Out (GIGO)
Tech professionals have used this phrase since the early years of computing. Yet in the age of AI, Garbage In, Garbage Out (GIGO) can have more far-reaching, compounded effects.
A SoftServe survey of business leaders found that 65% of respondents believe no one in their organization understands the data collected or how to use it. A Harvard Business Review report states that only 3% of organizations’ data meets basic quality standards. Organizations rely on AI to automate processes, improve customer experience, and drive decision-making. Data quality is more crucial than ever.
Unraveling ROT Data
ROT data is the most common source of GIGO. The acronym stands for data that is Redundant, Obsolete, and Trivial. In other words, this data remains stored but adds no value to the organization.
Improving Data Quality
What are the basics for improving data quality? First, capture all valuable data. It may exist in more places than you realize. Think about the information generated through social media channels, team collaboration channels, or instant messaging services. Look beyond traditional data sources.
Then, remove unnecessary information, or ROT data. My technology team tells me this involves structuring the data, adding relevant context, applying clear labels and metadata, and maintaining version control. These steps make the data suitable for AI use.
Strong data governance is essential. According to Collibra’s Data Intelligence Index, organizations with mature data governance programs generate 70% more than their peers. Establish clear data ownership and accountability. Create transparent processes for collecting, managing, accessing, updating, and protecting information.
Organization-wide data quality checks further strengthen this foundation. Data cleanup is not a one-time project, but an ongoing one.
Internal Consequences of Poor Data
Let’s look at the internal consequences of poor data within any organization. As a leader, my first thought is time and productivity losses. It takes time to clean data issues, which, if managed well initially, would never become a problem in the first place. Statistics show data scientists spend 80% of their time finding, cleaning, and reorganizing data.
Next, when a team is relying on data to make strategic decisions and the data is incorrect, the entire plan will be flawed. Obviously, this can undermine confidence in decision-making. Forecasting becomes uncertain. What’s more, organizations may allocate capital, technological investment, and employee resources based on inaccurate assumptions.
Stibo Systems research found that 32% of organizations said they’ve launched marketing campaigns targeted at the wrong audiences, all based on poor data.
Customer Experience Impact
Brand perception matters to every organization. A PwC report states that 32% of customers would give up on a brand after just one bad experience.
Contact center agents rely on data to assist customers quickly and resolve issues on the first call. If customers need to talk to multiple agents, repeat requests, or call back more than once, the customer effort becomes too high. Likewise, if your customer-facing AI starts acting unreliable or provides false information, your customers will not trust your systems or your brand.
Poor data can create ethical problems within organizations such as inequality or discrimination. In finance, this could affect credit scores and make it hard for customers to open credit card accounts, get personal loans, or mortgage a home.
In Conclusion
Your technology is only as good as your data. When AI systems malfunction, the results can be far-reaching.
