Data Literacy and governance at Scale
The short version
Transformed informal, undocumented data practices into a formal governance structure — creating accountability,
accessibility, and consistency across a 900+ person organization managing one of the most complex data programs in the
nonprofit sector.
---
What Existed Before
When I arrived, the organization had two things working in parallel that weren't quite working together.
On the constituent side — donor data, supporter records, finance — governance was mature. There were policies, owners,
and processes. On the programmatic side — the shelter data that powered our analytics and products — governance was
largely informal. A hand-selected group called the Mission and Metrics Advisory Team had been making important
decisions for years: defining what counted as a shelter, establishing which metrics were appropriate in which
contexts, building the shared language the organization and industry ran on. They had no official authority, but their
decisions stuck. That's a credit to the people involved. It's also a fragile way to run critical infrastructure.
As Shelter Pet Data Alliance scaled and data began flowing in automatically through API integrations, the informal
model broke down further. When data had come in manually, human eyes caught anomalies. Automation removed that check.
And when something did go wrong, tracking down who was responsible for a given shelter could take days.
---
What I Built
The first intervention was simple but foundational: we identified a relationship owner and a data owner for every
single shelter in our dataset. Clear accountability at the record level.
From there we built outward. We documented 20+ SOPs that had never been written down. We centralized data dictionaries
and business rules that existed in people's heads or in files nobody knew to look for — cataloging 86 business terms
in a shared data dictionary — and published them on the intranet so anyone could find them without knowing to ask. We
established the Shelter Stats Governance Committee — a formal body with actual authority over definition changes, SOP
approvals, and oversight of how data is handled across the organization. The Mission and Metrics Advisory Team, which
had done years of important work without any formal standing, was ultimately sunset. The Governance Committee replaced
it.
We also built automated anomaly detection to replace the manual review that had disappeared when data collection moved
to API integrations — catching errors before they reach stakeholders. And we created a formal data sharing agreement
policy, establishing the organization's first governance layer for how data moves between internal and external
partners.
We made deliberate prioritization decisions along the way. When we brought on a governance leader, the original
mandate included building a full data catalog. We pivoted — data quality was the more pressing need, and our existing
catalog was sufficient for data engineering purposes without requiring a larger investment that wouldn't touch the
business directly. Knowing what not to build is part of governance too.
One of our most significant ongoing efforts is de-siloing shelter-level data that has accumulated across the
organization over time — survey responses, animal-level records, pre/post programmatic engagement results, data
collected by individual strategists and programs for specific needs. The goal is to centralize it at the shelter level
so the analytics team can see across all of it, not just the slice any one program collected.
---
What Changed
Senior leadership and our team spend significantly less time navigating individual data questions — because the
answers exist, they're findable, and they're consistent. In 2025 alone, our quality systems ran 1,303 data checks and
made 237 corrections before data reached a single stakeholder. When a definition is in question, there's a committee.
When an SOP is unclear, there's a document. When something goes wrong with a shelter's data, there's an owner.
None of this is governance for governance's sake. Every piece of infrastructure we've built is in service of faster,
higher-quality decision making — getting the right information to the right people quickly enough to actually change
outcomes for animals.
The next frontier is training for data producers: anyone entering data into our constituent management software, our
shelter management software, or accessing insights from our data visualization tools. Building literacy at the point
of data creation is where the long-term quality gains are.