The Data Layer Is the Implementation

Half of every decision intelligence program is data work. The other half is decisions. The half most enterprises underestimate is the data work. The half most enterprises sequence wrong is also the data work.

Here is the sequencing failure we see most often. The implementation starts. Configuration runs in a sandbox with sample data. Data integration is scheduled as a parallel workstream, usually owned by a different team, often offshore. Sprint by sprint, the configuration team builds against sample data. Sprint by sprint, the integration team builds pipes against assumed schemas. The two converge at User Acceptance Testing — which is when the enterprise discovers, for the first time, that the sample data was clean, the real data is not, the assumed schemas were optimistic, and several of the master data attributes the configuration assumed do not exist consistently in the real source systems.

The program slips. The slip is blamed on data quality. The data quality problem was always there. The sequencing decision is what made it terminal.

We have inverted the sequence on every engagement we run. Data and integration are the first workstream, not the last. Sprint one runs against the customer’s real data. We do not wait for clean data — we discover the problems sprint by sprint, while the design is still cheap to change. This produces three structural advantages.

The middle layer — between the raw data lake and the decision intelligence platform — is where most enterprises lose six months. We do not. Not because we are smarter at data work. Because we ordered the work correctly.