A big part of this is ensuring that data sources can be trusted, that the flow of data is monitored and governed. Isn't this just Data Stewardship? Well, partially. For instance, ThoughtSpot takes a "semantic-layer first" approach, which enables them to power the natural language processing with ease.Ī term that's gotten increasing use over the past few years has been Data Stewardship. Different platforms will handle this in different ways.
This data model is usually embedded into the analytic / semantic layer of most modern BI platforms (In Tableau this typically occurs at the extract layer), but this does no have to be the case! This abstraction layer just hast to live somewhere - it could be a homegrown solution of tightly controlled documents and databases - but usually the BI platform is the right solution and it should be the core function of any BI platform to make the implementation of the semantic layer easy. Much of BI work takes place here, deciding on the scope of data model that will serve the business while working within the technical constraints of available storage and compute resources. Those technical definitions are then translated into data model built using the technical and business definitions laid out above.This is usually done in consultation with a data engineer, who can help address gaps and answer questions. The business intelligence practitioner translates those KPIs, metrics, and dimensions into technical requirements, providing the database fields required and any transformations or aggressions occurring to that field.This is typically done with consultation from a business intelligence practitioner who can assist with building precise definitions that can be translated easily into technical requirements. subject matter experts) provide precise business definitions for the metrics, KPIs or dimensions they require to answer their business questions. This typically happens in a 3 part process.
In fact, Business Intelligence solutions will not fully succeed unless this semantic layer is properly maintained and governed, because this is where trust is created. It falls on Business Intelligence to take ownership of this problem. And unfortunately most analysts will do the work without pushback and the executive that started things in motion will not understand the efforst they've dedicated to this task. Why isn't there agreement on sales numbers? Is the problem at the report level? At the database level? Was it an analyst error? Can we trust any of our data? This can often set in motion hours of meetings, testing, and diagnostics while analysts from different departments (who may have never worked together) try to cobble together an answer that will satisfy their bosses. This creates layers upon layers of headaches that quickly add up, especially at scale. This occurred because there was no centralized semantic layer that clearly defined what Sales means to the business and instead both analysts are likely pulling data their own way, with their own queries, and call them the same thing, which means their can be no common truth. The analysts are using different rules to define the same metric (Sales). Well, yes, they might be using the same data, but the financial analyst must include product returns but the marketing analyst removes product returns. How could this be? Aren't they using the same data? Many executives take this as a given and are shocked when they find out the Sales numbers in marketing do not match finances numbers. As stated in the definition, the goal is to create a shared version of truth. This is part is often overlooked or undervalued because failure in this area more directly impacts the front-line analyst and the work and value created by this part is usually not understood by executives, at least until something goes wrong. of creating a shared version of truth via an analytical or semantic layer grounded in business rules and needs. I like to refer to this part as the Information Stewardship. For this article I'll discuss the second part of the definition I laid out in my first article on BI. Last week I wrote about Information Design, which is probably what most people think of when they think of Business Intelligence - the dashboard or the report they end up using or building.