Build the Foundation
Production-ready AI begins with reliable, accessible information, which requires modernizing fragmented data environments, followed by pilots demonstrating material, measurable improvements.
“Most retailers ready to move to production have their data house in order,” says Ananda Chakravarty, vice president of research for IDC retail insights.
Beyond the technology, retailers need backing from executive stakeholders, supported by teams assigned to defined use cases, and employees capable of integrating AI into existing operations.
This also means redesigning workflows, a focus on change management and calculating the economics of token consumption and other usage-based costs.
Chakravarty highlights the importance of ensuring software vendors and systems integrators work seamlessly; otherwise, gaps may arise between a successful pilot and a scalable deployment.
READ MORE: Find out how AI is changing the way we shop.
Keep People in Control
Retailers can reduce deployment risk by beginning with repetitive, narrowly bounded processes.
Chakravarty points to associate knowledge bases and contact center service as practical starting points because the underlying workflows and resolution paths are generally well established.
“The solutions are well developed for various call center systems, with standard agent execution functions — usually, password resets and routine handled tasks that can be almost scripted,” Chakravarty says.
However, they still require supervision, with human approval mandatory for all agentic AI use cases. Retailers should define approval thresholds, track discrepancies and review automated decisions at least as rigorously as they assess human work.
Chakravarty explains this is especially important when a faulty answer or unauthorized action can interrupt a purchase or damage customer trust.
“Monitoring will be a requirement from day one, plus regular oversight; at the very least, the same level that humans on the job would receive,” he says.
