AI trends: customer experience
C · Weak signal Google Cloud's customer-experience edition of its AI agent trends report; same five trends and the same vendor survey as the general edition, with CX examples and product links added.
Summary based on the full text·Original: about 20 minutes to read
In short
A January 2026 Google Cloud report that recasts its five agent trends for 2026 (agents for every employee, every workflow, customers, security and scale) for customer experience leaders, with emphasis on financial services and retail. It describes service representatives working with background agents for knowledge lookup, quality review, coaching and recommendations, "concierge" customer agents grounded in CRM and order data, and agent-initiated payments under Google's Agent Payments Protocol (AP2).
Key numbers
- 49%executives at agent-using organisations who have deployed agents for customer service and experience (n=1,814)
- 36%of those customer-service deployers, share already reporting ROI on those agents
- 40% vs 37%financial services vs retail and CPG executives whose organisation has launched more than 10 agents
- 1-3%share of calls human supervisors typically review for quality, which an AI quality agent could cover in full (Qualtrics)
As reported by the source.
English·PDF·opens services.google.com
What it reports
- Survey figures reused from Google's ROI of AI 2025 series: 52% of gen AI users have agents in production; 49% of agent users deploy them in customer service, and 36% of those report ROI.
- 40% of financial services executives (n=556) and 37% of retail and CPG executives (n=585) say their organisation has launched more than 10 agents.
- Quality agents could review all interactions rather than the 1-3% of calls supervisors typically sample.
- Customer examples: Citi says over 182,000 staff in 84 jurisdictions have access to internal AI tools, with about 21 million interactions through Q3 2025; Home Depot's "Magic Apron" guidance agent; PayPal and Salesforce work on AP2 and A2A protocols.
- Agentic commerce raises open questions: proving a customer authorised a purchase, verifying an agent's request is accurate, and assigning liability for fraud or errors.
How they got it
Internal Google Cloud interviews, customer case studies and the vendor's ROI of AI surveys (3,466 decision makers). No new data collection or method beyond that.
Limits
- Numbers are recycled from Google-commissioned surveys and self-reported; ROI is perceived, not measured.
- Case material is supplied by customers or Google and is promotional.
- Product calls to action throughout.
Why it matters for companies
The concrete ideas are the agent roles around a service representative (knowledge, quality, simulation training, trend detection) and the move from sampled to full quality review, which changes what supervisors do. The payment-authority questions for agent-initiated purchases are real and unresolved, so legal and risk teams should be involved before any agent can transact. Treat the adoption and ROI figures as vendor marketing.
Summary based on the full text. Read the original for the full argument and data.Open PDF on services.google.com
- Tags
- AI agents