Retail & E-Commerce Cloud Solutions

Agentic shopping: your next big customer isn’t a person

A growing share of shopping traffic is an AI agent acting for a customer. It reads your catalogue, compares you against four competitors, checks whether the size is actually in stock, and in a rising number of cases completes the purchase without the customer ever loading your product page.

That’s a channel shift, and channel shifts reward whoever is ready early.

Why this is an opportunity rather than a threat

The retailers who panic about agents are the ones whose advantage was friction. If your conversion depended on a shopper being too tired to check three other sites, an agent removes that. Nothing to be done there.

But most retailers lose sales for the opposite reason. A customer can’t tell whether the item is in the store down the road, the size chart is a JPEG, delivery timing appears only at checkout, and the returns policy is four clicks deep. Human shoppers give up quietly. An agent gives up instantly and tells the customer you had nothing suitable.

Fix the machine-readability and you win demand you were already losing. Clean structured data, accurate real-time stock, and honest delivery promises are exactly what an agent rewards, and they happen to be what human customers wanted anyway. Which is the point: this isn’t a separate agentic strategy bolted on, it’s the merchandising hygiene you’ve been deferring, now with a revenue number attached.

What an agent needs from you

Structured product data first. Schema.org product markup, a well-formed Merchant Center feed, and attributes that mean something (actual fabric composition, actual dimensions, actual fit) rather than marketing adjectives. An agent can’t infer that “generous cut” means size up.

Real-time inventory second, and this is where most estates fall over. If your stock position is a nightly batch, an agent will confidently promise a customer something you sold this morning. That failure costs more than a missed sale, because the customer blames you rather than the agent. Streaming inventory through Pub/Sub and Dataflow into a serving layer that answers in milliseconds is the fix, and it’s the same pipeline that makes click-and-collect trustworthy.

Then a programmatic surface. Agents interact through APIs and increasingly through Model Context Protocol servers that expose your catalogue, availability and pricing as callable tools. Emerging payment standards, including Google’s Agent Payments Protocol, aim to let an agent complete checkout with a verifiable mandate from the customer rather than a stored card. This part is still moving, and anyone who tells you the standards have settled is guessing. Building the data layer underneath is the no-regrets move, because every version of this future needs it.

The uncomfortable part

Agents comparison-shop without loyalty or fatigue. If your only differentiator is price, agentic traffic will compress your margin faster than any price-comparison site did. Retailers with genuine differentiation, exclusive range, reliable next-day delivery, decent returns, service that solves problems, do better in an agent-mediated market than in a search-mediated one, because agents can actually read and weigh those attributes. Retailers competing purely on being findable will struggle.

We’d rather say that plainly than sell you an agent strategy that can’t fix a positioning problem.

What we build

• Real-time inventory. Change data capture from POS, ERP (SAP, Oracle) and e-commerce through Datastream and Pub/Sub, so stock positions are current everywhere and fewer orders are cancelled for stock you no longer have.
• One view of sales and customers in BigQuery. Store receipts, online sessions and loyalty data together, for basket analysis and targeted promotions.
• Customer service agents. Gemini Enterprise agents that answer delivery questions, start returns and check local stock at any hour.
• Search and recommendations. Vertex AI Search in place of keyword search that returns nothing, with recommendations based on what shoppers are trying to do.

Related work

• Snowflake cost guarantee: move from Snowflake to BigQuery and cut your bill by 20%, or we pay you $50,000 in cash.
• Data and analytics: BigQuery, Looker and real-time streaming.
• Retail data integration guide: connecting POS, ERP and BigQuery for real-time inventory.
• Gemini Enterprise for CX: customer service agents grounded in your BigQuery data.
• All case studies