Duolingo
GenAI customer agent
In 2024, Duolingo deployed Decagon's AI support agent on its chat and deflected 80 percent of incoming volume, versus 30 percent with the previous vendor, with go-live in about one month.
Key points
- AI support agent deployed on in-app chat, then extended to email.
- The Decagon platform built on OpenAI models, with the FAQ synced hourly.
- 80% chat deflection versus 30% before, go-live in about one month.
- Evidence B, confirmed status.
Objective
Bring down the ticket volume handled by humans and the time the team spends maintaining the support tool, on a base of several hundred million users.
The deployment
Duolingo deployed Decagon's AI support agent on its chat in 2024. The agent deflects 80 percent of incoming chat volume, versus 30 percent with the previous vendor. Setup took about one month from the first discussions to go-live. Email was planned for early 2025. The switch also lightened the team's load: automatic hourly sync of the FAQ, less manual maintenance, and human agents refocused on complex cases. Duolingo claims a base of 500 million users, which gives a sense of the contact volume absorbed.
Results Proof B
A quantified case study from the Decagon platform with a named, quoted person (Ian Riggins); cross-checked by the OpenAI customer story on Decagon. A clear deflection figure but from an interested source, hence level B.
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 Decagon AI support agent platform deployed on in-app chat; 80 percent deflection versus 30 with the previous vendor.
- llm Modeles OpenAI (via Decagon) Decagon builds its agents on OpenAI models, according to the OpenAI customer story on Decagon.
- infra Base de connaissances / FAQ Duolingo Source of the agent's answers, synced automatically every hour.
- infra Chat support in-app Initial deployment channel; extension to email planned for early 2025.
How it runs, concretely
For ops teams-
1Connecting the knowledge base operations team
Wire the FAQ and procedures so the agent answers on up-to-date content, with automatic hourly sync.
-
2Chat go-live agency
Deploy the agent on incoming chat; the switch took about one month.
-
3Handling and deflection AI
The agent resolves the majority of requests without a human (80 percent), and surfaces complex cases.
-
4Refocusing the agents operations team
Human agents focus on complex cases; the tool's maintenance load collapses.
The deflection rate and the accuracy of the answers. The system relies on an FAQ synced hourly; if this base drifts, answer quality drops and volume returns to humans.
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
- structured, up-to-date FAQ and procedures
- ticket history
- read access to the user account
Org prerequisites
- a support team ready to move into supervision
- a process to update the FAQ
- escalation rules
Possible stack
- Decagon
- Sierra
- Intercom Fin
- Zendesk AI
The plan, step by step
- Step 1Audit and clean the FAQ and procedures: this base drives the agent's quality.Deliverable: An up-to-date, structured knowledge base
- Step 2Connect the base to the agent platform and configure automatic sync.Deliverable: Agent wired to up-to-date content
- Step 3Test answer accuracy internally and define the escalation rules to humans.Deliverable: Test suite passed, escalations defined
- Step 4Go live on chat and track the deflection rate.Deliverable: Agent live on chat with a deflection dashboard
- Step 5Extend to email and refocus the support team on complex cases.Deliverable: A second channel covered, maintenance load reduced
First step: Audit the existing FAQ and clean it up: it drives the agent's quality, even before choosing the tool.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Duolingo Customer Success Story Interested party archive pending
- S2 Delivering high-performance customer support (Decagon) 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.