Meta launches Business Agent for in-app shopping, support

Meta launched Business Agent, an AI tool that automates shopping and routine customer service inside Instagram and Messenger. WhatsApp integration is planned for a later date.

The agent can answer product questions, recommend items using a merchant’s catalog, guide buyers through checkout within the chat and complete payments without redirecting users to external sites.

Keeping the purchase flow inside the messaging app aims to reduce cart abandonment that can occur when shoppers are sent to separate checkout pages. A typical interaction begins with a shopper asking about size or color on Instagram, moves into a purchase conversation in Messenger and ends with an in-chat payment.

For customer service, the agent is designed to handle repetitive, low-complexity tasks such as order status checks and basic returns. The company positions the product as a way to automate high-volume first-tier interactions so human agents can focus on escalations and complex cases.

The system uses continuous learning from ongoing conversations and accepts automated syncing of product databases and inventory updates. Retailers with seasonal catalogs or rapidly changing stock can push updates to the agent so recommendations and availability reflect current data.

Because the agent runs natively inside Meta’s apps, it can access a user’s in-app history and connections to personalise recommendations and enable secure, in-chat payments. Meta says replicating the same level of integration through external vendors is difficult, which can shorten deployment timelines for small and medium merchants.

Operational requirements include maintaining clean, machine-readable product data to avoid incorrect responses and building robust customer identity verification before the agent can process returns or reveal order information. Authentication workflows must integrate with existing single sign-on systems.

Engineering teams need to define permitted actions for the agent and establish escalation procedures to hand conversations to humans. The company recommends extensive pre-launch testing, including running large volumes of simulated conversations to find edge cases and prevent customers from getting stuck in automated loops.

Adoption choices involve trade-offs between platform-native convenience and independent architectures. Using the native agent gives immediate access to the platform audience and reduces upfront development, while a custom stack allows selection of language models, specific data residency controls and long-term portability. Many organisations are expected to use the platform agent for high-volume routing and their own systems for large financial transactions and complex account issues.

Deployment time varies by size: smaller retailers can roll out faster with lower costs, while larger enterprises will need time for data clean-up, security integration and quality assurance before full production.