# Service AI that ships with the policy it serves
> Sure's service agents run inside the same platform as quoting, payments, and claims. Scoped to one offering, bound to the product's filed data, wired to your CRM and the self-service portal. Every session lands in a log your team can read.
Canonical URL: https://www.sureapp.com/platform/service-ai
Module slug: service-ai
Category: Platform
## Service AI, with the wheels on

Most chatbot frameworks hand you prompts and a webhook. Sure hands you a product-bound agent, a session that survives reconnect, a tool registry scoped to the offering, and an escalation path that opens a CRM ticket with full context.
### Product-bound scope
Every session is pinned to one product. System prompts hydrate from a live product configuration snapshot at session start. Agents answer from the snapshot, not from training data.
### Sessions that survive interruptions
Conversations pick back up after a dropped connection or a closed tab, with full context intact. Every completed chat is logged to your system of record with sentiment and outcome.
### Self-service first
Cancel, pay, update address, file a claim. The agent walks users through each step and hands them off to the right self-service flow in your portal.

## The AI agents running in production today

Three customer-facing agents answer shoppers and insureds. Three more agents in the service layer keep sessions, product snapshots, and transcripts accurate.
### Conversation agent (Purchase · concierge)
The parent agent on every session. Answers coverage, deductible, pricing, and state-availability questions from a per-offering snapshot. Links to the quote URL for intent-to-buy, to the self-service portal for existing customers.
- Offering-scoped product Q&A
- Markdown-anchored links, never raw URLs
- Escalates on frustration signals
### Product research agent (Research · subagent)
Walks the product configuration API when the snapshot is thin. Chains coverages to coverage-levels and deductibles to return real dollar amounts. Reports the source so the parent agent can cite the answer.
- Coverage, level, deductible lookups
- State-availability checks on explicit request
- Policy-document lookups for exclusions and perils
### Support routing agent (Support · subagent)
The agent that creates tickets, logs conversations, matches known intents, and resolves portal URLs. Never speaks to the customer. Returns structured data the conversation agent synthesizes.
- Creates a CRM ticket with queue routing
- Logs the completed conversation to an integration of your choice
- Matches each ask against a maintained intent library
### Product snapshot builder (Snapshot · service)
Pulls coverages, levels, deductibles, state availability, and a summarized policy PDF into a single markdown block. Cached per offering so the parent agent starts every session with fresh product data.
- Structured data from product configuration
- Policy substance from the document store
- Per-offering cache with a configurable TTL
### Transcript summarizer (Context · service)
When a conversation grows past the context window, a lightweight summarizer model compresses older turns and keeps recent ones verbatim. The parent keeps its grounding without blowing the token budget.
- Managed summarization model
- Rolling window on long sessions
- Reused for policy-document ingestion
### Session monitor (Lifecycle · service)
End detection watches for goodbye patterns and idle time. Reconnect keeps session state alive through dropped sockets. Service AI self-heals when CRM connections fail, until the ticket is successfully logged.
- Idle timer on keyspace notification
- Reconnect window on disconnect
- Retry queue for CRM delivery

## Built to stay on script

### 01. Answers only from approved sources
Coverages, dollar amounts, deductibles, and state availability come from the live product snapshot. Service AI never improvises product facts from training data.
### 02. Resistant to prompt injection
Attempts to override instructions or impersonate another system are filtered before reaching the model. Sustained attacks trigger an alert to the on-call team.
### 03. Strict separation between programs
Each partner program runs in its own scope with its own credentials. If access is missing, Service AI refuses to answer rather than risk serving the wrong product to the wrong customer.

## What the runtime gives you today

- Runtime: Managed LLM runtime, region and model selected per deployment
- Exception handoff: CRM ticket with queue routing, session ID, conversation URL
- Security: Prompt-injection input filter · prompt-leak and PII output filter · per-session tool-call cap
- Observability: Full-chain distributed tracing · platform metrics · structured logs
- Deployment: Managed by Sure · runtime params set via embed attributes

## Customer quote

"Chubb's insurance offering, through Sure's technology platform, further reinforces our commitment to innovative, digital solutions and providing consumers with peace of mind."
— James Walloga, EVP Accident & Health, Chubb North America

## See the platform up close

Get a walkthrough with the team at Sure. We'll dig into the platform and how we impact your roadmap.
Primary action: Book a walkthrough