Cross-agency data: why 70% of failures are semantic

September 29, 2026 · 5 min · Free, with full transcript

When two government agencies define "household" differently, one by mailing address and one by shared cooking facilities, no integration platform reconciles the numbers. This episode argues that roughly 70 percent of cross-agency integration failures trace to the semantic layer rather than the plumbing, and that the median time to reach a trusted cross-agency data agreement is 18 months. You get a worked example from benefits coordination, where food assistance uses an economic household and Medicaid often uses the tax household, plus the trap of mapping household_id to household_id. You leave with a sequencing rule for governance spend, a named-owner test, and a three-entity definition exercise. References include MIT Sloan Management Review and dbt Labs.

Listen to the podcast

5 min

Chapters

Key takeaways

  • Spend the first dollar on written definitions, not an integration platform, because mapping fields that mean different things creates agreement without substance.
  • Treat any vendor promise of interoperability in a quarter as selling the connection, not the consensus; plan against an 18-month median for a real cross-agency agreement.
  • Ship one agreed entity, such as household, case or resident, with both agencies signing the exact wording including edge cases, then publish it.
  • Name a single person who can say no on each definition instead of a working group, since shared ownership means the definition drifts.
  • Ask each partner agency to define your three most-shared entities independently in one sentence, and treat every disagreement as the actual project.
Read the full transcript

Host:Welcome back to Leaders Insights. The shared data infrastructure problem that quietly breaks every cross-agency program, and why it matters this week. Two agencies, same citizen, and they can't agree whether she lives in one household or two. Why does that break a program that cost forty million dollars to build?

Expert:Because the forty million bought pipes, not meaning. You can move data between systems flawlessly and still have both sides pointing at the word "household" like it's obvious. One agency counts a household by mailing address. The other counts it by who shares a stove and a fridge — that's the old census definition, shared cooking facilities. Same person, two truths, and now your eligibility numbers don't reconcile.

Host:That sounds like a definitions problem, not a technology problem. So why do CDOs — chief data officers, the people who own this — keep buying more technology?

Expert:Because technology has a demo and governance has a committee. You can show a dashboard to a legislator. You can't show them a six-month negotiation over what "active case" means. The article puts a hard number on this: in the cross-agency programs they reviewed, roughly 70 percent of integration failures traced back to the semantic layer — the shared meaning of the fields — not the plumbing. The plumbing works. The plumbing almost always works.

Host:Seventy percent. What does that number actually change for someone about to sign a vendor contract?

Expert:It changes where your first dollar goes. If seven out of ten failures are about meaning, then buying an integration platform before you've written down your definitions is spending on the 30 percent problem. MIT Sloan Management Review made this point years back and it still holds — the organizations that succeed with shared data spend disproportionately on governance up front, and it looks slow, and then they lap everyone who went fast.

Host:Give me the concrete version. One program, one specific mismatch.

Expert:Benefits coordination. Say a state runs food assistance and Medicaid — health coverage for low-income residents — off two systems. Food assistance treats a household as an economic unit that buys and prepares food together. Medicaid often assesses eligibility by tax household, meaning who's claimed on the same return. A grandmother living with her daughter is one household to one program and two to the other. When you try to auto-enroll across both, you either double-count her income or you miss it entirely. Either way she gets denied, or the state overpays, and both are audit findings.

Host:And no field-mapping fixes that.

Expert:None. You can map "household_id" to "household_id" perfectly and you've just connected two things that were never the same thing. That's the trap. Mapping gives you the feeling of agreement without the substance.

Host:There's a second figure I want to press you on. The article talks about how long this actually takes.

Expert:Eighteen months. That's the median they cite for a genuine cross-agency data agreement to reach the point where both sides trust each other's numbers. Not the tech build — the agreement. And here's what that number should do to your roadmap: if a vendor promises interoperability in a quarter, they're selling you the connection, not the consensus. dbt Labs has good material on defining metrics once and reusing them everywhere — worth reading — but they sell data tooling, so treat their timelines as best-case in a clean environment. Government isn't clean.

Host:So what's the CDO supposed to tell a governor who wants a win in twelve months?

Expert:Tell them the win is the definition, not the dashboard. Ship one agreed entity — pick "household" or "case" or "resident" — get both agencies to sign the exact wording, including the edge cases, and publish it. That single artifact is worth more than the integration, because every future program inherits it. You're building a foundation, not a feature.

Host:How do you stop the definition from dying in a committee?

Expert:Name an owner who can say no. Not a working group — a person. Shared ownership means nobody owns it, and it drifts back to "let's mapping it later."

Host:Last question. Somebody's listening, mid-project, integration budget approved. What do they do Monday?

Expert:Before Monday's standup, take your three most-shared data entities and ask each partner agency to write the definition independently, in one sentence. Compare them. Wherever the sentences disagree, that's your real project — and it was never going to be solved by more pipes.

Host:Sources for today's episode: The New Stack, Towards Data Science, dbt Labs (vendor — data tooling), MIT Sloan Management Review. That's your five. The full CDO track, structured and free of fluff, is at mba-training.com.

Read and go deeper