# Data Mesh (https://riska.io/platform/data-mesh)

_Data Mesh_

## Every decision, connected to your whole book.

Never act outside your guidelines. All the information your agents and users need to get work done.

## One figure, read by every decision.

A value lands on the record once. It is then what prices the risk, what the referral branches on, and what written premium and loss ratio roll up from.

- **Submissions land on the mesh.**: Extraction knows which insured, location or coverage each field belongs to. What comes back is a proposed change to those records, cited to the page it was read from, not a summary of the file.
- **Price comes from the record.**: A rate table names the object it rates, and a rating run pulls its inputs from that record. The premium it computes is written straight back onto it, so the quote and the book never disagree.
- **Referrals fire on the change.**: A trigger names the attribute it watches and the value that should fire it. Change a limit or a loss count and the run opens with the changed record already in scope.
- **Loss ratio is a query, not an export.**: A portfolio metric is a saved query over your records. Group written premium or loss ratio by any attribute you can reach from the object it counts, then split it by product and by lifecycle state.

> "“We stopped arguing about which system was right. There is one record now, and the rating, the referral and the month-end report all move with it.”" — Chief underwriting officer, Program carrier · 12 lines

## Ask Riska answers from the mesh.

Ask Riska works from your model and your records, so an answer about your book is computed from your book, not from a general reading of insurance documents.

### It knows what you call things.

The assistant is handed your model in your own words, with every object and relationship carrying the label your team gave it, whether that is a program, a layer or a class code. Rename an attribute and its picture of the model updates on the change, not on a schedule.

### It keeps hold of the account in hand.

Every record the conversation touches is kept with its real id, its object and its label. Ask about the second renewal on the list and it resolves to that policy, then opens where you would have clicked.

### It cannot invent a figure.

Earned premium, incurred losses and counts are computed over the whole filtered set and returned with the population they cover. When it draws you a table, it points at the query result rather than retyping the rows.

## Built once, true on every record.

Every record is reachable the same way, so the things most policy systems rebuild per module are built once here and work on submissions, policies, claims and brokers alike.

- **One search box**: Insureds, brokers, policies, claims, files, notes and tasks, from a single field.
- **One history**: A record’s activity tab is its change log: who moved which fields, when, and what the record read before. The audit trail is already written.
- **Notes on anything**: Notes, threaded comments and mentions attach to any record type, including the objects you added for your own programs.
- **Favorites and recents**: Pin or return to any account, policy or referral, in folders you arrange yourself.
- **References that resolve**: A pointer to a record carries its display label, so the same insured reads correctly on a task, a comment and an audit row.
- **One set of permissions**: Agents and automations call the same interfaces your underwriters do, and see exactly what the caller is allowed to see.

## Every change, accounted for.

An underwriter, a release or an agent can move the same record, and the record keeps one account of all three.

### An underwriter edits the record

The edit is written once and logged with the actor, the time and the columns that moved. Open the entry to see the row before and after.

### You change the model

Objects, attributes, views, document types and rate books are versioned together as one release. Publish it and every quote, workflow and report reads the new version at the same moment.

### An agent changes it

An agent write is stamped with the agent and the conversation it came from, and a change to a record waits for a person to approve it before it lands.

## Every surface reads the same mesh.

Triage, appetite, rating, workflows, reporting and Ask Riska all read and write the same records. Start with the desk that needs it most.

## See one account, connected.

Bring an account you keep in three systems today. We will put it on the mesh once, then show you the same record in the workbench, in the rating, in the loss ratio report and in Ask Riska.