# Introducing the architecture for agentic premium (https://riska.io/blog/architecture-for-agentic-premium)

_2026-10-17 · Riska Underwriting Team · announcement_

Agents can do the high-context work of an underwriter at the speed of a machine, and carry that judgment across every account on the book. Most insurance software was built for a person who opens one file and finishes it. Today Riska is introducing a hybrid architecture engineered for people and agents in equal measure: the Data Mesh connecting every submission, policy, premium posting and claim, Ask Riska speaking SQL over the live book, and reporting and connectors that keep every motion on the same records.

## Key takeaways

- Agentic premium is a hybrid architecture: people and agents share one connected record of the book, one set of permissions, and one path from question to action.
- Ask Riska speaks SQL, so an agent can query millions of policies, claims and relationships and return an exact figure with the rows behind it.
- Reporting and saved views run on the same live records the workbench edits. No overnight copy, no export that goes stale before the broker rings back.
- MCP and the developer platform let other agents and tools work against that same book, under the same access controls.
- Calls, emails and notes sit on the account as first-class activity, so the next person and the next agent start from what was actually said.

## This is the architecture for agentic premium

Riska combines the speed of infrastructure with the context and control of application software, in one architecture built for people and agents.

It works at the speed of an agent, across the whole book, to the standard of your best underwriter. Every task an agent completes adds context for the next, so the file gets richer with every interaction.

Three interconnected layers make it possible:

- **Context.** Every signal on an account (documents, ratings, money, conversations) attaches to the insured, policy and claim it concerns, on one record that stays current.
- **Orchestration.** Agents and workflows use that context to act on every signal across underwriting, billing and service.
- **Ecosystem.** Connect Riska to the tools and agents that already run the business, with consistent access to its data, logic and actions.

Every product we have shipped has laid the groundwork for this. Today we are naming four advances that extend what it can do.

### Three layers, one book

Context feeds orchestration. The ecosystem reaches both.

- **Context** The Data Mesh connects every account signal to its insured, policy and claim
- **Orchestration** Ask Riska, workflows and reporting act on that context
- **Ecosystem** MCP, the API and apps reach the same data and actions

_Each layer compounds the others._

## SQL engine

**Ask Riska** now speaks SQL, the data language of agents.

Agents use it to query millions of records and relationships at once and perform complex calculations directly in Riska. Questions that previously required exported datasets or a custom report can now be answered in a single query.

For example:

- Find which brokers produce the highest bind rates on accounts over $50,000.
- Calculate written premium and loss ratio across regions and owners.
- Identify open submissions associated with accounts that cancelled in the last year.

Riska performs the calculation and returns an exact result. Every answer shows the rows behind the figure, so a person can inspect before they act.

The same query surface powers [Reports & Metrics](/platform/reporting): filter, join and total any object you have modelled, then save the answer as a view the whole team reads. Identify accounts that cross a concentration threshold and route them, or calculate values across related records to prep a renewal.

## Data connectors

Riska exposes the live book to the agents and tools you already run, without a second copy of the data.

Connect assistants through [MCP](/apps/mcp), call the typed API from the [developer platform](/platform/developers), and keep dashboards on the same records the underwriter edits. Queries run against Riska, so analytical tools and agents work with live book data rather than an overnight sync.

- Build portfolio forecasts in the BI tool you already trust.
- Track performance against custom targets on live premium and loss.
- Analyze deals across segments, teams and regions without an export.

When a policy changes, the updated data is immediately available to connected tools. Access follows each person's Riska permissions. Workspace admins control which credentials stay live.

For teams with an existing data stack, Riska becomes a live, directly queryable source. For teams building one, advanced analytics no longer starts as a six-month integration project.

## Agent scale

A person might review a few dozen accounts in a day. An agent can work across thousands in the same morning.

Riska is engineered so agents and people share the same object model and the same access controls, whether the book is hundreds of policies or millions of records.

- Research every account coming up for renewal.
- Continuously requalify the pipeline against today's appetite.
- Keep millions of records current as new documents and calls land.

Thousands of agent turns can work across the customer data, with the same permissions that govern the rest of Riska. As the book grows, agents take on more of the file work without giving up the reliability the team depends on.

## Activities on the file

Not every important customer interaction happens in an email or a meeting note someone remembered to file.

Riska extends the data model so the interactions that actually move a risk sit as structured, first-class data on the account: support threads, broker DMs that never left Slack, product events, and recorded calls.

[Call Intelligence](/platform/call-intelligence) is the clearest example. A broker meeting is recorded, summarised, and landed on the submission or policy it concerned, with next steps already lined up as tasks.

- A support escalation is tagged before the renewal conversation starts.
- Product events and mid-term asks sit beside the exposures they change.
- Whoever picks up the file, person or agent, starts from what was said.

Give the team a complete picture of every relationship, and give agents the context they need to act on it.

## One architecture, from question to premium

Revenue teams already have a name for this shape of software. Insurance needs it for premium: one connected record of the book, agents that can ask it a precise question, and reporting that never drifts from the file.

[The Data Mesh](/platform/data-mesh) is the connected record. [Ask Riska](/platform/ask-riska) is how a person or an agent questions it in SQL and in plain language. [Reports & Metrics](/platform/reporting) is how the answer becomes a view the whole team keeps. [Workflows](/platform/workflows) carries the result into the next action, under the same permissions.

Open Ask Riska and ask a question about your pipeline, loss ratio or renewal book to see the architecture at work. MCP and API access are available for teams that want other agents and tools on the same spine.

## Questions we get asked

### What is the architecture for agentic premium?

It is a hybrid architecture where people and agents share one connected record of the book, one permission model, and one path from a question about the book to an action on a record. Context, orchestration and ecosystem stay connected rather than reconciled.

### Does Ask Riska really speak SQL?

Yes. Ask Riska can express questions as SQL over your live book, run the calculation in Riska, and return an exact result with the rows behind the figure. The same surface powers saved views and portfolio reporting.

### How is this different from exporting to a warehouse?

Exports go stale the moment a policy changes. Riska answers against the live records the workbench edits, so the figure you quote the broker is the figure on the file.

### Can other agents connect, or only Ask Riska?

Other assistants can connect through MCP, and systems can call the typed API. Every request runs as the signed-in user or machine credential, inside Riska permissions.

### Where do calls and emails fit?

They land on the submission or policy they concern as first-class activity. Call Intelligence summarises meetings and opens tasks from what was agreed, so the next person and the next agent start from the same file.

### Is this only for MGAs?

No. Carriers, MGAs and brokers that keep underwriting, billing and service on one book get the same architecture. The point is one connected record for people and agents, not a single distribution model.