Worked Example⬇ Download .md

Grace Hill — Assumed Internal Estate

A modeled internal technology estate for a mature mid-market property-management compliance & training SaaS company — every entry ASSUMED, none confirmed.

Confidential · June 2026. This document MODELS a plausible internal estate inferred from Grace Hill's public business profile (property-management compliance & training SaaS that ships "Gracie," a generative-AI assistant; CTO Chris Fontan; AI-forward leadership). It is ASSUMED / illustrative — every technology, vendor, region, and figure below is a reasoned guess about what a company of this shape typically runs, NOT a confirmed fact about Grace Hill's actual environment. The only confirmed inputs are public OSINT. All assumptions are replaced with measured fact the moment Aperture connectors run live; until then this is a discovery hypothesis, not a record.

How to read this

The govrn assessment never starts from a questionnaire. It starts from an estate hypothesis — what a company like this probably runs — which Aperture then confirms or corrects against actual metadata. This document is that hypothesis for Grace Hill, written so the delivery team can pre-stage connectors, anticipate the three lenses (Technology, Security, AI), and predict where shadow AI hides before the first connector authenticates.

Read every "Vendor (ASSUMED)" entry as "we expect to find one of these; we will know which when we measure."

Cloud & infrastructure (ASSUMED)

Component	Assumed vendor / shape	Lens	Why we assume it
Primary cloud	Azure or AWS, single primary + DR region (e.g. `East US 2` + `Central US`, or `us-east-1` + `us-west-2`)	Technology	Mid-market B2B SaaS norm; Azure tilt if Microsoft-centric (see identity)
Data residency	US-only; SOC 2 / tenant-data isolation expected	Security	Fair-housing + tenant PII raises residency stakes
IaC	Terraform or Bicep/ARM; partial coverage, drift likely	Technology	Mature-but-not-perfect IaC is the mid-market default
Edge / CDN	Cloudflare or Azure Front Door	Technology	Standard public-app fronting

Product & development stack (ASSUMED)

Component	Assumed vendor / shape	Lens
Web application	Multi-tenant SaaS (React/TypeScript front end; .NET or Node services)	Technology
Services / API	REST + some async workers; relational core (SQL Server or PostgreSQL)	Technology
Source control + CI/CD	GitHub or Azure DevOps (Azure tilt likely); pipelines with uneven gating	Technology + Security
Artifact / package	GitHub Packages / Azure Artifacts; container registry	Security
Observability	Datadog, App Insights, or Grafana stack	Technology

The CI/CD lens question is not "do they deploy?" — it's what gate, if any, sits between a model-prompt change and production. For a company shipping Gracie, that gate is the AI-lens crux.

AI footprint (ASSUMED) — where the engagement earns its fee

Component	Assumed vendor / shape	Lens	Shadow-AI exposure
Gracie LLM pipeline	Retrieval + generation over compliance/training content; prompt-orchestration layer	AI (product)	The shipped product — primary attestation target
Foundation-model APIs	Anthropic, OpenAI, and/or Azure OpenAI; possibly more than one, undocumented	AI	Multiple provider keys = multiple egress paths to inventory
AI gateway / proxy	Possibly none — direct SDK calls	AI + Security	No gateway = no central log, no cost attribution
Employee coding AI	GitHub Copilot, Cursor	AI	Code + snippets to third-party models, often un-reviewed
Employee chat AI	ChatGPT (personal and Enterprise unclear), Claude, Gemini	AI	Highest shadow-AI risk — tenant/fair-housing text pasted into consumer tools
Embedded vendor AI	AI features inside CRM, support, M365 Copilot	AI	"Accidental" AI no one inventoried

Where shadow AI hides (our prior): unmanaged personal ChatGPT/Claude logins on corporate identities; a second model provider used by one team and unknown to security; Copilot/Cursor on repos touching tenant data; M365 Copilot quietly enabled; direct provider SDK calls with no gateway, so no one can answer "how much are we spending on inference, and on what." Aperture's job is to surface each with a dollar figure and a control gap.

Business SaaS (ASSUMED)

Component	Assumed vendor / shape	Lens
Productivity / email	Microsoft 365 (Azure tilt) or Google Workspace	Technology + Security
CRM	Salesforce or HubSpot	Technology
Support / ticketing	Zendesk or Intercom	Technology
HR / HRIS	Workday, Bambo.HR, or Rippling	Security
Data warehouse	Snowflake (assumed) or Azure Synapse	Technology + AI
BI	Power BI or Tableau	Technology

Identity (ASSUMED)

Component	Assumed vendor / shape	Lens
IdP / SSO	Microsoft Entra ID (most likely) or Okta	Security
MFA / conditional access	Enforced for admins; coverage gaps for contractors likely	Security
Provisioning	SCIM partial; orphaned accounts a standing risk	Security

Identity is the spine of the AI lens: every shadow-AI tool is reached through some login. If SSO doesn't front the model providers and AI tools, there is no chokepoint — and no record.

Data classes (ASSUMED) — what raises the stakes

Class	Example	Sensitivity driver
Employee PII	HRIS, payroll	Standard regulatory
Tenant / resident data	Screening inputs, applications, demographics	Fair-housing exposure
Fair-housing-sensitive	Anything feeding tenant-screening or housing decisions	Discrimination & disparate-impact liability — Gracie's core risk
Customer (PM-company) data	Multi-tenant SaaS records	Contractual isolation duties
Source & secrets	Repos, keys, model credentials	Breach + model-egress risk

What the team does with this

This estate hypothesis pre-loads the engagement: connectors are staged against the assumed identity, cloud, and AI providers above, then run to measure. The delta between this modeled estate and the measured record is the assessment finding. Until connectors authenticate, every line here stays labeled ASSUMED / illustrative — and the first deliverable to Grace Hill is the corrected, measured version of this exact table.