← Back to all agents
A

Annette

Orchestrator agent for The Body Shop Indonesia Stores

Store ops where the team already is.

100 stores across Indonesia run daily operations through WhatsApp, the chat already open on every store phone. No new app, no training.

Same task, before and after

Member data update

1 hour→20 seconds

Queued behind the customer service workload.

The groups

EXISTING SURFACE · ONE AGENT PER GROUP100storesACROSS INDONESIAWhatsAppTHE SURFACEMMembershipAGENTUpdate customer detailsPPromotionAGENTConfigure POS for promosSStock takeAGENTGuide and monitor end to endCSCustomer serviceAGENTScore responsiveness and sentimentITIT helpdeskAGENTCCTV, videotron, daily follow-upONE AGENT PER WHATSAPP GROUP100storesACROSS INDONESIAWhatsAppTHE SURFACEMMembershipAGENTUpdate customer detailsPPromotionAGENTConfigure POS for promosSStock takeAGENTGuide and monitor end to endCSCustomer serviceAGENTScore responsiveness and sentimentITIT helpdeskAGENTCCTV, videotron, daily follow-up

How it works underneath

LangChain and LangGraph route each WhatsApp group to its own agent, and every agent carries only the tools its group needs. Membership writes customer records. Promotion configures the POS. Stock take reaches server health and employee config. Customer service runs the analytics. IT helpdesk drives ticket ops. The chat stays familiar while the tools reach the systems that actually run each job.

  • Membership: update membership.
  • Promotion: update POS config.
  • Stock take: server health, employee config and stock take config.
  • Customer service: CS analytics.
  • IT helpdesk: ticket ops.

Everything is logged

Every action lands in a database and gets analysed. The helpdesk mirrors to Notion, so the queue is readable from a phone. Annette runs on the same knowledge graph as ICA, so the business definitions carry over without being restated.

Lesson learned

The first version had its own interface and almost nobody logged in. The same model moved into WhatsApp and adoption followed within days, because store teams were already in that app all day. The problem was not the model, it was the distribution channel. Tapping into a flow people use every day beats asking them to learn a new one.

← All agentsGitta →