Outdated business apps can cloud your AI vision

Outdated software applications are creating roadblocks to AI adoption at many organizations, with limited data retention capabilities a central culprit, IT experts say.

Decades-old apps designed to retain a limited amount of data due to storage costs at the time are also unlikely to integrate easily with AI tools, says Brian Klingbeil, chief strategy officer at managed services provider Ensono.

Moreover, the cost of maintaining outdated software, with a shrinking number of software engineers familiar with the apps, can be expensive, he says. With legacy apps tying up a significant portion of an organization’s IT budget, less money is available for new initiatives, further slowing down AI adoption. According to IDC’s 2023 CIO Sentiment Survey, organizations were spending an average of 12.8% of their IT budgets on tech debt at that time.

The data retention issue is a big challenge because internally collected data drives many AI initiatives, Klingbeil says. With updated data collection capabilities, companies could find a “treasure trove” of data that their AI projects could feed on.

“If the application itself is legacy, with bad code stored in different places, all sorts of problems can come from the app itself,” he adds. “What legacy apps have in common is they tend to have been written when storage cost a lot of money, and now storage is basically free.”

Customer concerns about old apps

At Ensono, Klingbeil runs a customer advisory board, with CIOs from the banking and insurance industries well represented. The problems that legacy apps create for AI projects have been a recent topic of conversation with those CIOs, he says.  

Banking and insurance are two industries still steeped in the use of mainframes, and Ensono manages mainframes for several customers. While big iron and its software remain important to these organizations, and could have a future with AI, mainframe-dependent companies are losing their internal expertise as older IT workers retire.

“These older apps that have been built on mainframe shouldn’t necessarily be replaced,” Klingbeil says. “It can either be that the mainframe could work perfectly fine, or it’s expensive to move, or it’s risky to move, or not worth the money. But they can be modernized.”

Klingbeil and Ensono have seen the challenges that legacy apps present for AI firsthand. When building a machine-learning-powered tool to predict the maintenance needs of its customers, Ensono found that its customers used multiple old apps to collect incident tickets, but those apps stored incident data in very different formats, with inconsistent types of data collected, he says.

Other IT leaders see the same challenges that legacy apps create for AI. The head of data and analytics at a large enterprise recently told Jeremiah Stone, CTO of integration-platform-as-a-service (iPaaS) provider SnapLogic, that its data was in no condition to be useful to AI because of poor management of its applications in past years.

“In many cases, outdated apps are completely blocking AI adoption,” Stone says. “The open secret among CIOs is that a huge chunk of investment going into AI is being spent with service partners building modernization strategies or upgrading outdated systems.”

Stone called outdated apps a “multi-trillion-dollar problem,” even after organizations have spent the past decade focused on modernizing their infrastructure to deal with big data.

“We are in mid-transition,” Stone says. “We haven’t realistically begun updating and normalizing the wider semi- and unstructured data applications that pervade business processes and that are exactly the data and business flows that can most benefit from the latest waves of AI innovation.”

Modernize in stages

To fix the problem, CIOs should first take inventory of their existing IT infrastructure and identify the areas with the greatest need for modernization, Stone recommends.

“Ultimately, a mixture of old and new systems will remain — a situation that requires robust integration strategies to avoid data chaos and siloed solutions,” he says. “The aim is to create integration pipelines that seamlessly connect different systems and data sources.”

CIOs should focus on software modernization projects that matter the most, adds Justice Erolin, CTO of software development firm BairesDev. CIOs should identify applications that directly affect their AI initiatives and work on those first.

In some cases, companies can modernize their business applications by adopting middleware and APIs to connect legacy systems with newer technologies, instead of a wholesale rewrite of the code, he adds.

“This allows for the extraction and integration of data into AI models without overhauling entire platforms,” Erolin says.

CIOs should also use data lakes to aggregate information from multiple sources, he adds. AI models can then access the data they need without direct reliance on outdated apps.

Data engineering to bridge the legacy-AI gap

Some IT leaders, however, don’t believe outdated apps are a huge roadblock to AI projects.

While it can be tricky to extract data from legacy software, a far bigger problem is in the next step, says Robert Cloutier, lead data and AI engineering manager at Nexapp, a provider of networking, IoT, and edge computing solutions. After the data is extracted, IT teams need to interpret the extracted data and align it with the specific requirements of AI-based apps.

“The journey toward effective AI utilization is not solely about overcoming technical integration obstacles,” he says. “It’s about bridging the gap between raw data and actionable insights.”

Some older business apps collect and save a limited amount of data, but others have all kinds of information that can be valuable to the organization, Cloutier adds. In some cases, IT leaders hesitate to tap that data because they don’t know how to extract it, but the correct data engineering expertise can figure it out.

“Those old systems have been running for decades, so there is really a ton of valuable information you leverage,” he says. “There are treasures in there that they don’t even want to tackle because they want to change the system, or all the legacy data, and then you have to wait for years.”