IT‑IS‑Networking
Technology › Data & Records

Business Intelligence & Performance Analytics

Architecture direction and standards over an enterprise analytics platform, held separately from the group that built the reports on it.

Also known as BI, Analytics, OLAP, Data Warehouse, Microsoft BI Stack, Dimensional Modeling

Left to right is time. Each row is one of the 4 disciplines this one worked alongside, ordered so kindred disciplines sit next to each other rather than alphabetically — a landscape only means anything if neighboring rows belong together. Height is the work running in that domain in that year, weighted by what failure would have cost — life-safety ×3, a statutory obligation ×2.2, the estate itself ×1.5, an evaluation ×1. Every project declares which of those it was, so each ridge traces to an entry you can open and disagree with. Only the work that used this technology is counted, so what shows here is the company it kept rather than the record around it. Height and color are on the record-wide scale rather than this page’s own, so a modest estate draws a modest landscape and any two of these pages can be read against each other.

Analytics is where an estate’s numbers stop being operational data and start being evidence, and the architectural question is not which chart library to use. It is who gets to decide what the numbers sit on.

At the City of Jacksonville the enterprise architecture group held authority over all technology, business intelligence included, and what it directed was the substance: the platform analytics would run on, how that platform integrated with the rest of the estate, and the standards it had to meet. Delivery sat with a separate standing Business Intelligence Group under its own manager, building and operating inside that direction.

Keeping those two apart is the whole discipline. A reporting practice that owns its own platform choices becomes a second architecture nobody documented — one that pulls from the systems of record on terms it set itself, and that nobody can reconcile against the estate it claims to describe.

The platform was the Microsoft stack: SQL Server Analysis Services cubes, SQL Server Reporting Services, and a SharePoint front end, served at its own enterprise hostname with development and quality assurance siblings alongside production. That last part matters more on an analytics platform than anywhere else, because the output is a number somebody will act on. A report served from the wrong environment is not a broken page. It is a wrong figure with an authoritative URL.

The dimensional work underneath it is its own body of method — third normal form to star schema, dimension and fact design, surrogate and foreign keys, hierarchies, degenerate dimensions, fuzzy grouping, slowly changing dimensions, and extract-transform-load design on full against incremental refresh. Those are decisions about what a question can be asked of the data later, made before anyone knows which questions will be asked.

And the reporting layer is not an appendix to a model. A chargeback model an agency cannot read a report out of is a spreadsheet with ambitions: the dispute it was built to make possible never happens, because nobody can see the number clearly enough to argue with it.

Also searched as

  • ETL
  • ELT
  • data warehouse
  • OLAP
  • dimensional modeling
  • star schema
  • analytics platform