RESEARCH@DBTA: Enterprises Wrestle With Data Trust Issues Behind AI

Two surveys show that many executives are skeptical they can move forward with AI demands, which include the right data, at the right time. AI is a high priority, and is in production at most sites, but supplying trustworthy data is still a challenge at the back end.

Four in 10 executives don’t trust their data to generate accurate AI outputs, a Teradata survey of 300 executives shows.

While enterprise decision makers trust the potential of AI, at least 40% lack confidence in their company’s strategy to execute, as well as the data readiness to ensure the reliability of AI outputs. Another 61% said they fully trust the reliability and validity of their AI outputs.

“That’s not much better than a coinflip difference between trusting AI outputs and not,” the survey authors point out. “The pressure to beat out competitors can create risk when the trust in your data and AI is lacking.”

Within internal AI projects, 63% of executives surveyed report using a mix of closed and public datasets, while only 29% rely exclusively on closed datasets. The key factors in delivering trustworthy AI, cited by a majority of respondents, include security (61%), transparency (55%), and reliable and validated outcomes (50%).

Beyond quality data, survey participants seek enhanced efficiency in operations (74%), demonstrated successful use cases (74%), and improved decision-making processes (57%) from their AI deployments. AI is already up and running at a majority of the enterprises surveyed. Most respondents (84%) expect to see results from their projects within a year of deployment. More than half (58%) said the results would be quantifiable within 6 months. Another 60% said they have already seen demonstrable ROI with their existing AI solutions.

Still, trust in the fuel behind AI remains problematic. A separate survey of 3,100 global IT leaders, released by Alteryx, finds almost 1 in 4 (24%) have an extremely low, or no, level of trust in their organization’s data, while only 20% have “complete trust”.

Only 10% of respondents state they have a “modern” data stack, digitally enabled and uninhibited by silos, the Alteryx survey finds. Almost half (47%) report they are actively updating their data stack infrastructure to make it more modern. About 1 in 5 (22%) report they are dealing with data bias challenges, and 20% need help with data quality. Overall, data executives in the survey ranked data quality as their top goal for new technology investments.

When asked about the future makeup of the data stack 3 years from now, the survey’s authors saw some shifts from data stacks’ current composition. Respondents say in 3 years, the top elements of their data stacks will be CRM software (currently the top applications), data science and AI platforms, and ERP software. Spreadsheets dropped from third place to seventh (30%) in the current stack.

The diminishment of spreadsheets may be attributable to the rise of data science and AI platforms within data stacks. “IT leaders likely hope these more advanced platforms will allow companies to automate many tedious, manual aspects of managing data—and, perhaps even more significantly, make sophisticated modeling and predictive capabilities more accessible,” the survey’s authors suggest.

Still, it’s significant that spreadsheets are expected to make up 31% of the data stack in three years. “[T]eams will rely less on spreadsheets for sophisticated data integration, manipulation, and analysis while other technologies step in,” the researchers predict.

At least 22% of the organizations in the Alteryx survey state they have chief data officers overseeing their data, although the results varied widely and didn’t show a consensus across respondents regarding data ownership. For 11% of companies, the board of directors is the ultimate owner of the data; 8% indicated senior executives owned the data.

“While there’s no correct answer for who should own data, the lack of consensus may be concerning when data access and management are requirements for a successful generative AI implementation,” the Alteryx survey authors note.