The interview is thirty minutes on a prospect's schema landscape (optional parts can extend it to forty-five). Within two business days it ends in a document of this shape: the systems that hold structure and meaning, what needs to plug in, where two systems disagree and what that is costing, a directional stage placement, and the remediation options ranked, with the parts the prospect can do alone marked as such. The map is the prospect's whether or not they go further.
1. Landscape map
| System | What structure it holds | Format | Owner (role) | Read by agents? |
|---|---|---|---|---|
| Policy admin (core) | Policy, coverage, renewal, product codes | SQL DDL | Policy Ops lead | Yes |
| CRM (Salesforce) | Customer, household, contact preferences | JSON Schema | CRM product owner | Yes |
| Warehouse (Snowflake) | Claims, premium, actuarial aggregates | SQL DDL | Data platform lead | Yes |
| Content CMS (Contentful) | Product copy, disclosure text, FAQs | JSON Schema | Marketing ops | No |
| Partner exchange | Binding, endorsement payloads | XSD | Integration architect | Yes |
| Internal wiki | Ad-hoc field notes, FIBO-derived ontology snippets | Wiki pages | Claims Ops and Legal both edit the FIBO-derived snippets; last silent overwrite: March. 3 agent answers used the overwritten FIBO sense in April; tickets INC-4412, INC-4489, INC-4510 (fictional). | No |
2. What needs to plug in
- When a renewal quote is requested, the agent needs to resolve the insured party and active coverages across CRM and policy admin, so that the quote uses one identity and one product code set. Source: policy admin. Consumer: quote agent. Direction: pull, through connectors.
- When a claims intake bot classifies a loss, the agent needs validated coverage semantics from the Core Model, so that routing matches the conformance contract. Source: warehouse and policy admin. Consumer: claims agent. Direction: pull, through the MCP endpoint.
- When product copy changes in Contentful, downstream agents need a change log of term and disclosure updates, so that answers stay aligned with filed wording. Source: Contentful. Consumer: advisory agent. Direction: push on change.
3. Disconnect register
| Id | The two systems | The disagreement | Stated cost | Owner |
|---|---|---|---|---|
| D1 | Salesforce and policy admin | 'customer' vs 'policyholder' vs 'insured'; household vs named insured | About 120 hours a quarter of reconciliation; 6 priority incidents; wrong insured identity produces 14 wrong-tool logs a week, counted from the platform incident log | No owner |
| D2 | Policy admin and partner XSD | Renewal date: anniversary vs billed-to vs effective | A partner launch blocked; 3 weeks of slip | Integration architect |
| D3 | Policy admin and Snowflake | Product code grain: plan vs coverage vs rider | About 40 hours a month of model rework | Data platform lead |
| D4 | Wiki ontology and the JSON Schemas | FIBO-derived terms not mirrored in live schemas | Agents invent synonyms; about 15 failed tool calls a week | No owner |
Costs are as the owners stated them in the interview and were not measured, with one exception: the 14 wrong-tool logs a week in D1 are counted from the platform's own incident log.
4. Complaints and opportunities
Complaint: quote agents fail the identity join on about 18 percent of multi-policy households, adding about 25 minutes of average handle time.
Opportunity: one governed Core Model, with connectors and an MCP endpoint, could reduce cross-system lookups on the quote path from five hops to one.
What would have to be true: a conformance contract, published schemas with a change log, and named owners for D1 to D4 before agent operations scale.
5. Stage placement
Sources and Structure are largely in place (SQL DDL, JSON Schema, XSD), but Semantics and Validation are uneven: the FIBO-derived ontology lives on the wiki and does not bind the live schemas. Retrieval and Agent Operations exist in pilots. A Learning Loop from failed tool calls back into the Core Model and its change log is not yet institutionalized.
This placement is directional, from a structured conversation. The Diagnostic Briefing places each stage on the L0 to L5 scale with the reasoning shown; the Core Model Blueprint measures it.
6. Remediation options
| Option | Fixes | Effort | Self-serve? | With help | Deliberately does not fix |
|---|---|---|---|---|---|
| Publish an identity and product-code Core Model slice; wire connectors for the quote path | D1, D3 | Weeks | No | Core Model Blueprint, Implementation | Partner renewal-date XSD; CMS change log |
| Stand up the MCP endpoint and a conformance contract for claims classification | D4 (partial) | Weeks | No | Implementation, Agent Context Cost & Failure Audit | Full FIBO alignment; Contentful sync |
| Self-serve disconnect register and owner RACI, from wiki to schema index | D1 to D4 (visibility) | Days | Yes | Diagnostic Briefing | The semantic merge itself |
| Self-serve change-log template for Contentful, feeding the agent context pack | Enables item 3 of section 2 | Days | Yes | Diagnostic Briefing | Policy and CRM identity model |
| Managed operations: scheduled failed-call review feeding Core Model updates | D4 | A quarter | No | Managed Operations | Legacy XSD partner freeze |
Ranked by impact over effort: the Core Model slice, then the MCP endpoint and conformance contract, then the disconnect register, then the change-log template, then managed operations. Connector availability is as labelled on coremodels.io/connectors on the day of the interview; the catalogue is the source of truth.
7. Not covered
- Regulatory filing systems, actuarial pricing engines, and third-party credit and identity bureaus.
- Runtime authorization, PII redaction policy, and multi-region schema promotion.
8. Reviewer
Reviewed by: name withheld in the sample, Ariesnet. Date: withheld in the sample.