How to avoid data paralysis with a practical AI data strategy

Waiting can be a 'very expensive mistake'

Based in Germany, Jochberger is CGI's vice president and global head of manufacturing and strategic business consulting while U.S.-based Martin is director of national AI strategy and technical lead.

Waiting for data to be perfect is a "very expensive mistake" some organizations make, Martin said.

"Chances are there's some data in your organization that has good data quality -- or maybe it just needs a minimal lift -- that could be a good place to start to see what AI use cases could be built on top," she said.

In the meantime, AI can be put to work on data management tasks, such as monitoring quality and suggesting fixes. It can even generate synthetic data for AI apps to use during early development.

Martin recalled how a client needed to test the accuracy of a large language model but lacked a ground truth set -- data that had been verified as accurate. The development team decided to use AI to generate synthetic data and asked business stakeholders to review and validate it, rather than starting from scratch.

"That way we didn't have to press pause on the whole project," she said. "Instead, we found a way to use AI to enhance our data and move forward."

However, such quick wins should occur in parallel with long-term data improvements.

"Early on, you can deliver value on small data domains that you have governed with high quality while modernizing the rest of your estate," Martin said.

Once the AI roadmap delves deeper into agentic AI, the discussion becomes almost philosophical and addresses questions such as what the organization is prepared to delegate to AI, according to Jochberger.

Helena Jochberger

"An agent can potentially pursue an objective across multiple steps and interact with different sources of innovation, use tools and different systems, and then take action and adapt based on the additional information it has acquired along the way," she said. "That's a very different proposition for an enterprise, and why integration of systems becomes much more significant. The agent does not just need access to the data; it needs the context -- more precisely, the industry context. It needs to know which system it can access and also what its authorizations are."

Other topics discussed in the podcast include the following:

• Talent issues, such as the role of data scientists and outsourcing.
• A telecommunications provider that successfully scaled AI across data domains.  
• Lessons learned along the way.

David Essex is an industry editor who creates in-depth content on enterprise applications, emerging technology and market trends for several Informa TechTarget websites.