Pillar 2: Data Readiness

Minimum Viable Data

With priorities defined, Orion shifted NorthRidge into agent-specific data readiness, deliberately avoiding any notion of enterprise-wide data transformation. This phase was explicitly framed as minimum viable data, not data modernization.

Michael framed it clearly in the kickoff meeting: "We're not here to fix your data. We're here to determine whether the three agents you've prioritized can operate with the data you have today. If they can't, we'll define the minimum remediation required—not a transformation program."

Rather than convening a single cross-functional data forum, Orion ran a series of agent-specific working sessions, one agent at a time, each with its own business owner, veteran practitioners, system owners, and data stewards. These sessions were sequenced over multiple days to maintain focus and accountability (see Data Readiness Session Guide).

Session 1: Pre-QA Validation Agent

Jennifer Liu, QA Director, led the first session alongside two senior QA reviewers who had each processed thousands of reports. David Chen, NorthRidge's Data Steward, and the IT application owner for the Document Management System joined remotely.

The session began with a critical question: What data does this agent actually need to validate a survey report?

The QA veterans sketched the workflow on a whiteboard. A report came in. They checked it against 63 validation rules — boundary completeness, methodology documentation, regulatory compliance markers. Most of these rules were deterministic — they either passed or they didn't. The problem was that these rules existed only in the heads of experienced reviewers.

"We've got 63 rules, but they've never been written down in one place. They're in training materials, in email threads, in people's heads. That's our first data gap." — Jennifer Liu, QA Director

By session end, the group had mapped five data sources the agent would need:

• Survey Reports (Document Management System) — read only
• Validation Rule Library (QA SharePoint) — read only, but needs formalization
• Historical QA Records (SQL Server, 3 years) — read only
• Property Classifications (Master Data Store) — read only
• State Regulations (External Regulatory API) — read only

The critical finding: the data was accessible, but the validation rules themselves needed to be extracted from tribal knowledge and formalized into machine-readable format. This was a focused remediation task — not a data modernization program.

Session 2: Field Note Normalization Agent

Mike Torres, Field Operations Manager, brought a different energy to his session. He arrived with printed examples of field notes — handwritten abbreviations, voice-transcribed observations, regional terminology that meant different things in different offices.

"Look at this," he said, holding up two notes describing the same type of boundary marker. One said "IP w/ cap" and the other said "iron post capped." The same thing, written completely differently.

The session revealed a messier data reality:

• Field Note Corpus — scattered across mobile devices, voice recordings, and handwritten forms
• Terminology Dictionary — did not exist
• Report Templates — existed but weren't consistently followed

"Our field teams have been using ChatGPT to clean up their notes before submitting. They're doing it because we've never given them a better option. That's shadow AI, and it's a compliance risk we didn't know we had." — Mike Torres, Field Operations Manager

The data readiness score for this agent was lower than the Pre-QA agent, but the path forward was clear: build a terminology dictionary as part of the implementation, and design the agent to learn from corrections over time. The group agreed this was feasible without enterprise transformation.

Session 3: Exception Routing Agent

Sarah Martinez, VP Operations, took ownership of the final session. This agent had lower volume but higher stakes — routing complex cases to the right senior experts.

The data discussion surfaced an unexpected challenge. NorthRidge had historical exception data, but the criteria for what made a case exceptional had never been formally documented. Different senior reviewers had different mental models.

"Tom looks at property value first. Maria looks at property type. I look at the client history," Sarah admitted. "We've never aligned on a single framework."

The session pivoted from data mapping to a brief criteria alignment exercise. Within an hour, the group had sketched the first draft of a unified exception framework:

• High property value (>$2M)
• Complex property type (multi-use, commercial, historical)
• Client with prior disputes or compliance issues
• Unusual methodology requirements

This framework didn't exist before the session. Creating it was part of the Pillar 2 work — not a separate initiative.

The Named Data Owners

Critically, each data source was paired with a named human owner responsible for access decisions. If ownership could not be established, the agent scope was adjusted rather than forcing new data processes.

Data Source	Named Owner	Access Authority
Survey Reports	Sarah Martinez, VP Operations	Full read access for QA purposes
Validation Rule Library	Jennifer Liu, QA Director	Full read/write access
Historical QA Records	Jennifer Liu, QA Director	Read access with audit logging
Property Classifications	David Chen, Data Steward	Read-only access via API
Field Note Corpus	Mike Torres, Field Operations Manager	Read/write for normalization
Exception Criteria	Sarah Martinez, VP Operations	Full ownership, evolving

Data Readiness Outcomes

By the end of three sessions, NorthRidge had a clear picture of data readiness for each agent:

Agent	Readiness Score	Key Gap	Remediation
Pre-QA Validation	11/12	Validation rules not formalized	Rule extraction workshop (2 weeks)
Field Note Normalization	8/12	No terminology dictionary	Build dictionary during implementation
Exception Routing	9/12	Exception criteria undocumented	Criteria alignment completed in session

No enterprise data transformation was required. Each gap had a focused, bounded remediation path. Total elapsed time: three days of sessions plus one week of documentation.

This approach allowed NorthRidge to evaluate whether the prioritized agents could operate within the data landscape as it existed today, without triggering a large-scale consolidation effort. It reduced risk, increased trust with IT and data teams, and ensured that when agents moved forward, they did so with clear boundaries and human accountability — not implicit assumptions.

---