Ping An
AI conversational agent for service and sales support on the customer base
In 2025, Ping An's AI agents handled about 1,702 million service interactions (80% of total volume), helped realize RMB 133,179 million in sales, and made it possible to reactivate 30% more policies via an AI plus human setup.
Key points
- Fleet of AI service and sales-support agents on the customer base.
- In-house agent platform, proprietary LLMs, and an in-house voice layer.
- 80% of customer service, RMB 133,179 million in assisted sales, +30% policies reactivated.
- Evidence A, confirmed status via the group's 2025 results.
Objective
Absorb a massive service volume without inflating headcount, and make AI a sales and retention channel in its own right by assisting agents and re-engaging customers whose policy has lapsed.
The deployment
Ping An runs a fleet of AI agents on its customer base. On the service side, AI representatives take the front line of incoming requests (policy questions, claims, routine operations) in natural language, in writing and by voice. On the distribution side, the same components assist sales agents and drive sales and follow-up campaigns. An 'AI + human' reactivation system splits the re-contact tasks on lapsed policies, with the AI prioritizing the cases and preparing the action, and the human closing. The group quantifies the impact in its 2025 results communications: about 1,702 million service interactions handled by the AI representatives, or 80% of the total customer service volume, RMB 133,179 million in sales realized with the help of the AI agents, and 30% more policies reactivated thanks to the AI + human setup. Internally, more than 230,000 employees use the agent platform and more than 70,000 agent applications have been built.
Results Proof A
Figures published by Ping An in its 2025 annual results communications (the group's report and official release), reproduced verbatim by the PR Newswire press distribution. Primary results source, several consistent operational metrics.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- plateforme Plateforme d'agents IA Ping An The group's internal platform on which more than 230,000 employees work and more than 70,000 agent applications have been built.
- llm Grands modeles proprietaires Ping An LLMs and NLP components developed in-house by the group for service, sales, and reactivation; exact models not published in the sources.
- outil Briques voix (reconnaissance et synthese vocale) In-house voice layer that lets the AI representatives handle requests by voice as well as in writing.
- infra Base clients unifiee du groupe Data foundation (policies, interactions, payments) that feeds the prioritization of follow-ups and the relevance of the answers.
How it runs, concretely
For ops teams-
1Handling inbound service AI
The AI representative answers routine requests in natural language, in writing or by voice, and resolves what can be resolved without a human.
-
2Sales assistance AI
The AI agent prepares and prioritizes opportunities, assists the sales agent or runs the direct follow-up depending on the channel.
-
3Reactivation of lapsed policies AI and sales network
The system splits the re-contact tasks; the AI targets and prepares, the human contacts and finalizes the reactivation.
-
4Escalation and human closing service team and sales reps
Complex or high-stakes cases go to a human advisor.
Unified customer data (policy history, interactions, payment status). If the customer base is not up to date, the prioritization of follow-ups and the relevance of recommendations degrade.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- Unified customer base (policy, interactions, payments)
- History of sales and reactivation outcomes
- Logging of service conversations
Org prerequisites
- Sales network ready to work in tandem with the AI
- Compliance function for assisted sales and profiling
- Governance of the conversational models
Possible stack
- Custom agent platform (Ping An route)
- LLM + rules engine + voice layer
- market conversational-agent solutions for a narrower scope
The plan, step by step
- Step 1Target a high-volume, low-stakes inbound service flow, and frame compliance: escalation rules, informing the individual, human oversight (GDPR art. 22, AI Act for insurance).Deliverable: Validated scope + documented escalation rules
- Step 2Deploy a conversational agent on this flow with systematic human escalation on complex cases.Deliverable: Agent in production on one channel
- Step 3Extend to voice and track the share of volume handled by the AI, first-contact resolution, and satisfaction.Deliverable: Automated share tracked monthly
- Step 4Add sales support: the AI prioritizes and prepares the follow-ups (lapsed policies, opportunities), the human contacts and finalizes.Deliverable: AI + human reactivation setup with a measured rate
- Step 5Industrialize an internal agent platform so teams can build their own use cases under governance.Deliverable: Agent catalog and model governance
First step: Target a high-volume, low-stakes inbound service flow, connect a conversational agent to it before adding sales support.
The same pattern, proven in other industries
Sources
- S1 Ping An Reports 2025 Performance on Positive Trends, Operating Profit Up 10.3% YoY Primary archive pending
- S2 Ping An Reports 2025 Performance on Positive Trends, Operating Profit Up 10.3% YoY (PR Newswire APAC) Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.