The Future Retail Customer Won't Shop Like a Human

The retail industry has spent years building the shopping journey around one kind of customer: a person who searches, reads reviews, and puts something in a cart. Before long, that person may send an artificial intelligence agent instead. Imagine telling an AI assistant, "I need everything for a backyard barbecue this weekend. Stay within my budget and make sure it arrives by Friday." The agent researches products, compares options, checks availability, and places the order. The bigger question is whether retailers are ready when it needs to complete the transaction.

Over the next six months to 12 months, most of what agents do will look ordinary, such as adding items to a cart, checking an order, or updating a payment method, and retail will see it before regulated industries do. How far this goes depends on whether retailers are ready to let agents act and consumers are comfortable letting them complete transactions.

Getting ready requires more than accurate content. The shopper’s request succeeds only if the agent can trust the product data, inventory, delivery dates, and return policies it finds. If the knowledge base carries a three-year-old policy or the inventory count is wrong, the agent may confidently choose an item that cannot arrive on time. That’s how a routine purchase becomes a service problem before the customer ever reaches checkout.

Even with the right products in the cart, the purchase can still fail when the agent tries to act. Large retailers run thousands of overlapping services with little documentation of which belongs where. The shopper may have an account in one regional platform, a payment method in another, and an order routed through a system that can’t see either. An employee might find the colleague who knows the workaround. An AI agent can’t, and the friction hidden in a demo appears the moment it tries to complete the transaction.

A successful purchase carries through checkout, delivery, and any return that follows. When product, order, and service systems work together, the order arrives on time, updates stay accurate, and the experience feels like one continuous transaction. AI orchestration matters because the agent must move through every handoff without losing the order, context, or customer trust.

Even when that experience works today, retailers can’t assume it will work the same way tomorrow. Model updates and new contexts can change how consumer agents behave. Retailers need to keep testing how they interpret product information, make decisions, and move through the purchase.

Most retailers won’t be ready by the holidays to support AI-led purchases at scale. Next year is more realistic, and getting there comes down to how the work is organized:

1. Start with one use case rather than a transformation. Reordering and order status work well because they're narrow enough to finish and easy to measure. Nothing slows a program faster than believing everything must be fixed first.
2. Be honest about readiness before committing. Examine the data, knowledge, systems, and business workflows the use case will touch. Size the gaps, determine whether the technology can predict demand spikes and scale before performance suffers, and decide if the work earns its cost.
3. Match each task to the right kind of agent. An agent can be a person, a rules engine, an application, system, or an AI, and assuming it should be AI gets expensive. Deterministic rules are cheaper and steadier, leaving AI for work that takes judgment.

What looks like a simple request depends on the retailer’s data, systems, policies, and capacity holding up without a person guiding the transaction. Retailers that spend the next year connecting discovery, commerce, and post-purchase will be ready when AI starts placing orders at scale. The rest will find out during peak season.

Chris Bartosik is senior director of software engineering at Concentrix, a global technology and services leader.