By Naveena Davay Arunkumar, Lead Data Analyst, IT Asset Management, CareFirst BlueCross BlueShield
The problem wasn’t technical
The usual account of moving from a governance committee to an automated pipeline focuses on the technical rebuild — the dbt models, the validation logic, the reconciliation pipeline. That’s the easier half. The harder half is organizational: convincing a committee that’s owned data quality decisions for years that most of those decisions belong in code now, not in a meeting.
I lead data analytics for IT Asset Management at a large health insurer. I inherited a reconciliation program between our Flexera discovery data and our ServiceNow CMDB where roughly half of our asset records didn’t accurately reflect what was actually deployed. The governance model in place was the classic one: a steering committee met monthly, reviewed discrepancies that had been escalated to it, and decided which system’s version should take precedence.
It’s a reasonable model on the surface. It falls apart in an environment with more than one legitimate system of record — which describes nearly every regulated enterprise. Disagreement between systems isn’t the exception. It’s the baseline, and routing every instance of it through committee review simply doesn’t scale.
Why the obvious fix met real resistance
The fix was straightforward: build reconciliation into a Snowflake/dbt pipeline and turn validation rules into tests that fail the build rather than sit on an agenda. If a server showed up in Flexera’s discovery with no matching configuration item in ServiceNow, that went from a discrepancy awaiting next month’s review to a same-day notification.
Proposing this to the governance committee didn’t go the way I expected. The committee didn’t dispute that it would work — that part they accepted fairly easily. What I hadn’t anticipated was how strongly they pushed back against trusting automation over their own judgment, and I think I was wrong at first to treat that objection as purely emotional. There was a substantive point buried in it: automation doesn’t understand nuance, and there genuinely were cases where a person’s context caught something a rule-based test missed.
The approach that built trust
Instead of pitching the pipeline as a full replacement on day one, we ran it in parallel with the existing manual process for several months. Every discrepancy the pipeline caught was logged alongside what the committee’s process caught independently. This did two things. First, it gave the committee concrete, case-by-case evidence of gaps the automated approach caught faster — some of which had sat unresolved for months under the old model. Second, it surfaced the pipeline’s genuine blind spots, which we then fixed together with the committee’s input rather than around it.
That collaborative process mattered more than the raw numbers. It reframed the committee’s role from “final arbiter of every discrepancy” to “owner of the rules that decide what counts as a discrepancy” — a smaller role, but one the committee could recognize as still genuinely theirs.
What shifted, and what it took
Over about eighteen months, asset data accuracy moved from roughly 50% to 95%, and discovery coverage moved from 70% to 95% alongside it. I don’t think those numbers would have moved nearly as far, or as fast, if we’d tried to force the pipeline into place over the committee’s objections instead of building it with their input.
The change management lesson I’d pass on: budget more time for trust-building than for the technical build. The pipeline logic was, relatively speaking, straightforward engineering. Getting a room of experienced people to hand over a decision they’d owned for years took longer, and required different work entirely — showing evidence case by case, sitting with legitimate objections instead of arguing past them, and being honest about where the new system was still catching up to human judgment.
Naveena Davay Arunkumar is a Lead Data Analyst in IT Asset Management at CareFirst BlueCross BlueShield, specializing in data governance and asset reconciliation.