The Data Hostage Crisis

Part of the Judgement Layer series on decision infrastructure.

Where does the data live that feeds a judgement layer?

The modern data stack debate (Snowflake vs. Databricks, lakehouse vs. warehouse) is largely a Silicon Valley sideshow. It describes maybe 30,000 organisations globally.

SAP alone has 200,000+ customers. Add Oracle, Microsoft Dynamics, Infor, and you’re looking at the vast majority of enterprises above €100M revenue. They don’t have lakehouse architectures. They have ERP systems, Excel, and meetings.

For these companies: can we get our own data out?

The Tightening Grip

RISE with SAP is marketed as cloud transformation. But it has a second effect: SAP manages the environment. The database sits inside SAP’s cloud perimeter. The data extraction pipelines that enterprises built over the past decade need to be renegotiated, not just migrated.

This isn’t adversarial. It’s rational business strategy. SAP is doing what any platform does when it reaches sufficient scale: moving from selling software to selling the operating environment. Oracle is doing the same. So is Salesforce.

But for anyone trying to build an independent reasoning layer, the implications are significant. Your ability to reason about your own business becomes mediated by the access your system of record vendor permits.

Every major vendor is reinforcing this position with AI features. SAP has Joule. Oracle has AI agents. Salesforce has Agentforce. Each is useful. Each is also a strategic move to ensure that AI-driven insights flow through the system of record, not around it.

The result: the vendor that stores your transactions increasingly controls the terms under which you can reason about them.

Read-Only as Independence

The architectural answer is deceptively simple: a judgement layer must be read-only by design.

It never modifies source data. The system of record stays authoritative for transactions. We reason across it.

Read-only access preserves independence. It lets you reason freely across data from any source, in any combination, without asking permission from any single vendor. SAP data, Salesforce pipeline data, market feeds, internal spreadsheets: all become inputs to reasoning, none becomes the gatekeeper of it.

This is not a technical novelty. It’s a political position disguised as architecture. The moment your reasoning layer can also write to the system of record, the vendor has a legitimate argument for controlling it. Read-only removes that argument entirely.

The Portability Question Nobody Is Asking

The data portability debate in enterprise software is well-established. Regulators and procurement teams understand that organisations should be able to extract their transactional data. GDPR pushed this further for personal data. The conversation exists, even if execution remains difficult.

But there’s a second portability question that almost nobody is raising: who owns the accumulated judgement?

When an organisation runs a judgement layer for two years, the most valuable asset isn’t the data it ingested. It’s the calibration patterns it built. Which assumptions held across quarters. Which decision-makers were consistently directionally right. Which reasoning templates produced good outcomes in which contexts. Which risk thresholds proved too conservative or too aggressive.

That calibration layer is organisational cognition in structured form. It took hundreds of decisions to build. It cannot be reconstructed from the transactional data alone, because it captures the reasoning between the data points, not the data points themselves.

If the judgement layer is built inside the system of record vendor’s perimeter, that cognition belongs to the vendor’s platform. Switching ERPs has always been painful because of data migration. Switching away from a vendor that holds your accumulated judgement is worse: you lose not just your records but your organisation’s learned reasoning.

This is why architectural independence matters at the judgement layer even more than it matters at the data layer. Data can be exported and reimported. Calibrated judgement, if it’s entangled with a vendor’s proprietary AI stack, cannot.

The Harder Question

The industry conversation about AI in enterprise is focused on features: which copilot, which agent, which assistant. The structural question matters more.

Who gets to build the reasoning layer on top of enterprise data, without being beholden to the platform underneath?

The enterprises that answer this question with architectural clarity (read-only access, vendor-independent reasoning, portable judgement) will compound their decision-making capability over time. The ones that let their reasoning layer get absorbed into their system of record will find, in three years, that they’ve outsourced their organisational thinking along with their data management.

The data hostage crisis isn’t about data anymore. It’s about who owns the organisation’s capacity to reason.