Falabella TARS — Google Cloud conversational AI for IT-internal incident management across 7 LAC countries

South-America (Chile origin; deployed across 7 LAC countries)

Content

Grupo Falabella, the Chilean-origin retail conglomerate with brands including Falabella Retail, Sodimac, Tottus, Banco Falabella, Mallplaza, and the IKEA franchise, deployed a generative-AI conversational agent called TARS on Google Cloud’s Conversational Agents platform with Gemini models, to manage internal IT incident tickets across 7 countries (Chile, Peru, Colombia, Argentina, Brazil, Mexico, and Uruguay — the Falabella operating geography).

The deployment is Google Cloud customer-storied and named in detail, with the following published metrics (from the Google Cloud Customer story in Spanish, dated 2025):

• +22,000 tickets created via TARS, the AI agent
• 33% reduction in tickets created by human agents in the helpdesk during phase 1 of deployment
• Ticket creation time: 16 minutes → 4 minutes
• Ticket resolution time: 2 hours → 30 minutes
• +80 increase in ticket-categorisation precision
• Multi-channel ingestion: voice, web, Microsoft Teams, WhatsApp

The deployment architecture is detailed in the Google Cloud case study, naming the following Google Cloud services: Pub/Sub (for ticketing-event orchestration), Cloud Run Functions (for API integration), BigQuery (analytical engine), Vertex AI Search (RAG corpus of internal-process documentation), Contact Center AI Platform (multi-channel convergence), Conversational Agents / Dialogflow (primary conversational platform), Firestore (rapid-access data store), Cloud Storage, Dataform (data-pipeline orchestration), Looker (ticket visualisation), Cloud Logging, and Gemini models for natural-language interactions.

Falabella today has >36 million customers, 90,000+ collaborators across 7 countries. Named engineers in the case study: Antonelli Torriani (Director de Operaciones TI Omnichannel Retail); Sergio del Pino (Tech Manager Omnichannel Retail); Richard Andrade (Cloud Engineer); Yeny Villavicencio (IT Manager Tottus Corporativo); Franco Peña (DevOps Engineer); José Luis Calderón Ríos (CIO Sodimac México).

STRATEGICALLY IMPORTANT DISTINCTION — this is NOT customer-facing retail AI. TARS handles internal IT incident-management tickets (employees reporting tech issues, store-network problems, etc.). It is functionally distinct from the corpus’s consumer-facing generative-AI retail pattern:

This distinction matters for the corpus’s regional comparison: NA has well-documented consumer-facing retail AI (Walmart Sparky, Instacart Cart Assistant, Kroger Gemini, Loblaw PCxpress). LAC has documented internal-operations generative AI (Falabella TARS) — but the consumer-facing pattern in LAC, while rumoured in strategy decks, has no deployment scope verified to the TARS level. Cencosud’s CenCoDay 2025 deck names “AI-Driven Hyper-Personalization” as a strategy framing but does not name a deployment with scope numbers. Walmart Chile AI deployments are not verified in public sources.

Why this deployment matters for the corpus despite the IT-internal framing:

1. It documents that LAC retail conglomerates can deploy generative-AI agents at scale. TARS being 22,000 tickets is non-trivial operational scale; the platform is multi-country (7), multi-brand (5+ Falabella brands), multi-channel (4+ input surfaces).
2. It demonstrates the Google Cloud Conversational Agents + Gemini pattern as the corpus’s first Latin American deployment. The corpus now has:
   • OpenAI-mediated consumer retail: Loblaw PCxpress, Walmart Sparky
   • Anthropic-mediated consumer retail (rumoured): Kroger
   • Google Cloud Conversational Agents + Gemini internal-operations: Falabella TARS (this unit)
   • Microsoft Azure-mediated agronomic AI: Kilimo (Microsoft × Kilimo partnership)
3. It positions Falabella as a regional AI-deployed retail player, distinct from Loblaw (Canada-only) or Walmart (US-anchored). Falabella’s cross-Andean operating geography (Chile, Peru, Colombia, Argentina, Brazil, Mexico, Uruguay) gives it reach the NA-anchored retailers do not have, even though its consumer-facing generative-AI product is not (yet) the equivalent of NA counterparts.
4. The “Solving for IT internal first, then customer-facing” pattern is structurally distinct from NA where consumer-facing retail AI comes first. Falabella’s stated future direction in the case study is to extend TARS to: “automatically document solutions, participate in incident review meetings with resolver teams, generate post-mortem tickets, and help identify root causes for reported problems. The goal: create a solution that can resolve repetitive cases without human intervention, so support teams can focus on the most complex incidents.” — i.e. automating tier-1 / tier-2 IT support before pivoting to customer-facing surfaces.

What this unit is doing in the taxonomy

Anchors the LAC conglomerate-retail generative-AI internal-operations pattern at deployed scale. Distinguishes:

• From NA consumer-facing generative AI retail (Walmart Sparky, Instacart Cart, Kroger Gemini, Loblaw PCxpress) by being IT-internal rather than customer-facing
• From NA internal-operations enterprise AI (Microsoft 365 Copilot deployment patterns) by being a custom conversational agent on Google Cloud Conversational Agents + Gemini, not a vertical SaaS product
• From consumer-facing generative AI delivery platforms (DoorDash voice ordering discontinued; Uber Eats AI Cart Assistant) by being IT-internal not in the delivery flow

Functionally pairs with:

• Agrosmart / Kilimo (LAC smallholder-side agronomic AI)
• Marfrig × Agrorobótica (LAC beef-processing AI)
• PineSORT / AinnovaTech (Costa Rica pineapple computer vision)
• Microsoft × Kilimo (LAC foundation-model-vendor × agronomic AI collaboration)

Why it matters for talks

• Substantively refines the LAC retail AI picture away from “LAC is 2 cycles behind NA” framing. The internal-operations generative-AI pattern exists in LAC at scale; the consumer-facing pattern is where the question mark sits.
• Useful counterweight to vendor figures in Cencosud / Walmart Chile framing-as-strategy conversations. Cencosud and Walmart Chile claim “AI-driven hyper-personalization” but Falabella is the only LAC retail conglomerate to publish verifiable deployment scope numbers (Google Cloud customer story, July 2025 reference).
• Useful case study for the operational AI as precursor to consumer AI narrative. TARS prioritised IT-internal first (“automate tier-1 tier-2 IT support”); consumer-facing AI surfaces are described as a future extension. This is a different strategic ordering from NA’s consumer-first deployments.
• Useful for the Google Cloud Conversational Agents + Gemini pattern as a deployment archetype distinct from the corpus’s existing OpenAI-mediated or Anthropic-mediated patterns.

Critical context

• Categorically distinct from consumer-facing retail AI. This unit should not be cited as evidence of LAC retail AI consumer-deployment; the corpus’s structural observation is that internal-operations generative AI in LAC retail exists at scope; consumer-facing generative AI in LAC retail does not (yet) exist at the same scope. The corpus needs both patterns kept distinct.
• The 22,000-tickets figure is from Falabella’s own framing via Google Cloud customer story. It is tier-2 (named customer + named services + named scope), but the scope figure is still effectively vendor-reported from the deploying organisation — same V0 / V1 tiering discipline as the in-service AI vendor-figure hygiene applied to other units.
• The case study is in Spanish (Google Cloud Customer es-419 page); the corpus’s English-language secondary press has not picked it up prominently. Worth flagging as a Spanish-language primary-source-in-undercovered-LAC signal: the LAC regional scan notes are Spanish-language-light, this unit partly restores the balance.
• The deployment architecture detailed in the Google Cloud case study is unusually thorough (Pub/Sub, Cloud Run Functions, BigQuery, Vertex AI Search, Contact Center AI, Conversational Agents, Firestore, Cloud Storage, Dataform, Looker, Cloud Logging, Gemini). The narrative thrust: a multi-layer generative-AI stack, not just a chatbot — distinguishing TARS from a basic FAQ-bot deployment.
• Falabella’s cross-LAC operating geography (7 countries, 90,000+ collaborators) gives TARS the corpus’s largest-scale named internal-operations generative AI deployment, but this scale reflects the conglomerate’s size, not the AI’s specific reach.
• Microsoft Teams and WhatsApp as incident-reporting channels reflect LAC-specific patterns (WhatsApp dominance in LAC consumer; Microsoft Teams in enterprise IT). Distinct from NA patterns where retail-employee channels tend to be Slack / Teams / custom-helpdesk-portal.
• The future-direction intent documented in the case study — automate tier-1/2 support, then extend to other surfaces — means TARS as-deployed in 2025 is not the final product. The corpus should treat this as a moving target, re-verifying at cycle time what TARS surfaces it operates in.