Data Silos Are An Accountability Failure

Data silos persist in matrix and divisional organisations because no one is accountable for sharing — not because systems fail to connect. Here is how to fix that.

A division head sits on clean, well-structured pipeline data. The finance team needs it for the quarterly plan. The data exists. The integration works. Nothing moves.

That is not a technology problem.

The case for blaming the tools is weaker than it looks

Talend and SAP both argue that centralising data into a cloud warehouse or unified platform is the primary fix. That position has real weight behind it. When two divisions run genuinely incompatible schemas, assigning an accountable owner does not make the data comparable. The technical gap is real, and governance rules applied to structurally incompatible data produce friction without resolution.

But the argument breaks down at the point of explanation. If technical incompatibility were the primary driver of silo persistence, organisations with modern integrated platforms would not have silo problems. They do. Ackerman's 2000 study on collaborative work showed that shared work breaks down on social rules before it breaks on tools. The presence of compatible systems does not guarantee sharing. A cloud warehouse solves access. It does not change whether a division head wants the finance team reading their pipeline numbers.

What actually keeps silos alive

Ho, Lam, and colleagues studied coordination across government silos and found that what improved coordination was clear political direction, monitoring, and remedial action by leaders — not system replacement. O'Reilly and colleagues connect cross-unit friction directly to how organisations split authority and reward local performance over collective outcomes. When a division head's bonus ties to their unit's numbers, sharing those numbers with a team that might challenge them is not irrational caution. It is a rational response to the incentive structure you built.

Nomad Data puts it plainly: incentives for collaboration stop one team from acting as gatekeeper for data that others need. Salesforce makes the same point from a different angle — shared ownership and data standards do more work than raw access when teams need the same facts.

The accountability gap has a specific shape. No one owns what happens when data does not move. No agreed definition exists for what "customer" or "revenue" means across units. No consequence follows from withholding. Acceldata's governance research argues that governance must cross system boundaries end to end — which only makes sense if governance is doing independent causal work, not just sitting on top of a technical fix.

Three things leaders can change without replacing a single system

Name an owner for each critical data domain, not a team, an individual. Give that person explicit authority to resolve definition conflicts across units and make the ownership visible in how performance gets reviewed.

Align the definitions before the next planning cycle. Bring finance, sales, and operations into a room with the specific fields that caused the last reporting conflict. Write down what each term means. Publish it. This sounds embarrassingly simple, which is why almost no one does it before a platform migration and almost everyone does it after one fails.

Attach a consequence to non-sharing. Not a punitive one — a structural one. If a division's data does not flow into the shared reporting layer by a defined date, their numbers do not appear in the board pack. The absence is visible. That visibility does the work that monitoring alone cannot.

Where this leaves the technology question

The technology-first answer is not wrong. It is incomplete. A minimum technical baseline matters — data that is genuinely incompatible across systems needs transformation before governance rules have anything to govern. But buying a new platform before fixing the accountability structure means the new platform inherits the same political geography as the old one. Ho et al.'s finding holds: coordination improved through direction, monitoring, and consequence. The sequence matters.

I have watched three organisations run Snowflake migrations while leaving division incentives untouched. All three had the same silo problems eighteen months later, with better-organised data inside them.

Start with who is accountable for what happens when data does not move. The system question comes after that answer exists.