Getting Started with Prism Datasets

Prism Analytics lets you bring external data into Workday and blend it with Workday-native data for reporting. The mechanics are approachable; the difference between a maintainable Prism estate and a swamp is dataset architecture.

Base datasets: land the data raw

A base dataset should mirror the source — same columns, same grain, minimal cleansing. It's tempting to fix data quality on the way in, but every transformation you bury in ingestion is invisible to the analyst debugging a number six months later. Land it raw; fix it visibly, downstream.

Derived datasets: transformations you can read

Derived datasets are where the modeling happens — joins to Workday data, calculated fields, filters, unions. Two habits keep them maintainable:

• Layer deliberately. Source-conformed first (clean types, standard keys), business logic second (joins, calculations), presentation last (final column names, filters for the report audience). Each layer has one job.
• Name for the consumer. A dataset called FIN_Claims_Monthly_Summarized tells a report writer what they're getting. One called Dataset_Final_v3_NEW tells them the estate is unmanaged.

Get the join keys right

Most Prism frustration traces to joins between external data and Workday objects. External systems rarely carry Workday IDs, so decide early what the durable key is — employee ID, cost center reference ID, supplier number — and enforce it at ingestion. If the source can't provide a stable key, that's a data contract conversation with the source owner, not something to patch with fuzzy matching in a derived dataset.

Refresh schedules and data freshness

Match refresh frequency to the decision the data supports. A daily operational dashboard needs a daily (or intraday) pipeline with monitoring; a quarterly trend analysis doesn't. Every dataset should surface its own load timestamp so consumers can see freshness instead of assuming it — and stale-data alerts belong on any dataset feeding a decision that matters.

Security is part of the model

Prism data participates in Workday security, and that's a feature — but it has to be designed. Decide which domains govern each dataset, whether row-level constraints apply (by company, region, or supervisory organization), and test with real user profiles, not just the integration account. Blended reports inherit the most restrictive intersection of their sources; surprises here are best found before go-live.

Start small, prove the pattern

The best first Prism project is one dataset family with a clear consumer: one source, a clean base-to-derived chain, one blended report someone actually uses. Prove ingestion, refresh, security, and reconciliation on that slice — then scale the pattern, not the exceptions.