ADT
genAI customer agent
ADT, a major US home security player, deployed an AI agent built on Sierra in late 2024 to handle troubleshooting and account questions, across a volume of around 2 million care requests per month, with tight guardrails and human review.
Key points
- AI customer agent on the help center (troubleshooting, password, EasyPay).
- Built on the Sierra platform, with guardrails and human escalation.
- Context of around 2 million care requests per month.
- Evidence level B, live status confirmed.
Objective
Absorb a share of the 2 million service requests per month with a reliable AI agent, without degrading quality in a sector where a misclassification has real consequences.
The deployment
ADT, one of the largest US home security players, deployed an AI agent in late 2024 built on the Sierra platform (Bret Taylor's company). The agent handles help center questions: troubleshooting (for example why a panel is beeping), password reset, EasyPay direct debit enrollment. ADT handles around 2 million care requests per month. The capabilities announced as extensions cover payment, appointment rescheduling, and ordering yard signs or batteries. The agent runs with tight guardrails and human review, because a mistake in intent classification for a security customer carries a real cost.
Results Proof B
Case study from the Sierra platform (official interested source) cross-checked by trade press that covered the partnership in November 2024; no deviation metric published, hence a B level rather than higher.
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 Sierra AI agent platform for customer service on which the ADT agent is built: workflows, guardrails, and escalation rules.
- llm LLM orchestres par Sierra Language models managed by the platform; neither ADT nor Sierra publishes the exact models used.
- outil Classification d'intention et escalade Component that distinguishes handleable requests (troubleshooting, password, EasyPay) from sensitive cases to route to a human agent.
- infra Connexion aux systemes compte et facturation ADT Access to the customer account (EasyPay, status, billing) to handle transactional requests.
How it runs, concretely
For ops teams-
1Definition of intents and guardrails customer experience team
Map the handleable requests (troubleshooting, password, EasyPay) and set strict limits on what the agent may not do on its own.
-
2Agent build agency
Wire the workflows, the connections to the customer account, and the escalation rules on the Sierra platform.
-
3Request handling AI
The agent answers troubleshooting and account questions, routing sensitive cases to a human.
-
4Review and extension customer experience team
Monitor conversations, fix errors, gradually open up new capabilities (payment, appointments).
The accuracy of intent classification. For a security customer, confusing an emergency with a billing question carries a real cost, so human escalation is wired at the slightest doubt.
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
- help center knowledge base
- read access to account and billing
- ticket history by intent
Org prerequisites
- support team supervising the agent
- escalation rules on sensitive cases
- conversation review process
Possible stack
- Sierra
- Decagon
- Intercom Fin
- LLM + in-house orchestrator
The plan, step by step
- Step 1Map the support intents, pick a low-risk batch (password, account status), and write the red line of cases to escalate automatically.Deliverable: Intent map + validated escalation rules
- Step 2Build the agent on the platform: workflows, read connection to the customer account, guardrails.Deliverable: Working agent in a test environment
- Step 3Open a pilot on a share of the traffic with systematic review of the conversations.Deliverable: Pilot report: resolution rate, intent classification errors
- Step 4Gradually broaden the scope (payment, appointments) and set up weekly conversation review.Deliverable: Agent in production on the intent batch, volume deviation dashboard
First step: Pick the low-risk intents (password, account status) for a pilot, and define the red line of intents to escalate automatically.
The same pattern, proven in other industries
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
Bank of America
conversational financial assistant at scale
Sources
- S1 How ADT deploys a Sierra AI agent to make every second count Interested party archive pending
- S2 ADT Partners with Sierra for Better AI Customer Support Secondary archive pending
- S3 Inside OpenAI Chairman's $10 Billion AI Customer Service Startup Sierra Established press 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.