Business Intelligence, Dashboarding and Machine Learning. How to choose the right business analysts toolset? — Broscorp

Business Intelligence, Dashboarding and Machine Learning. How to choose the proper toolset for your business?

Business
06.12.2021

As more and more businesses get started with AI and machine learning algorithms, there is a growing request to integrate the results of those algorithms into data visualization tools. 

Machine learning brings dramatic change to analyzing business KPI, as it’s possible to perform not only descriptive (i.e., “what happened”) analysis but also prescriptive what will happen. Together with “what-if” analysis, this type of analysis is crucial for business decision-makers.

This article aims to give a starting point in the research of the proper Business Intelligence toolset to support Machine Learning and Artificial Intelligence propagation in your organization. In this article, we will talk about existing integrations used by most popular BI vendors and try to understand which problems could be solved with each type of integration.

Business Intelligence, Machine Learning, Artificial Intelligence and Big data management

Types of the integrations

Built-in models/AI features

Most of the popular business analyst toolset come with pre-built AI features, like functions or specific chart types that allow users to configure some AI model on top of user data.

Some examples are:

  • Tableau Forecasting feature
  • PowerBI Forefasting
  • PowerBI Anomaly Detection

This is the most simple path to get Machine Learning/Artificial Intelligence to your decision-making process and does not require any specific knowledge to use.

Business Intelligence, Dashboards, Machine Learning

Access to raw Python/R execution environments

This type of integration comes very handy if you have data scientists and want to visualize the models’ output immediately. 

Machine learning models could be trained and executed using standard Python or R framework, providing results straight to the dashboards.

Some examples of this type of integration are:

— Tableau TabPy

  • This feature allows to the evaluation of any python code from within tableau functions. So, training or executing ML models is smooth and convenient.

— PowerBI custom Python queries

  • One downside of PowerBI python/R queries is that it works on the user desktop only. You can export the results, but to refresh such a data source, you need to do it only on the same host where it was developed (using a personal gateway).

— Server Side Extensions in QlickSense

  • I like the extensibility of the Qlick a lot, and the idea of extension is excellent. The downside is that in case you can’t find ready-to-go solutions, it requires some effort to develop the extension.

Full cycle ML/AI support

Systems allow to train new models and get the predictions part of the typical BI and built-in transformations dataflow.

For now, I only can refer to PowerBI as the only system with such feature. This is possible due to close integration with Azure platform and I like the implementation a lot.

Other 3rd party services integrations

With this type of integration, BI tools provide you with access to 3rd party service that can give back the results of some pre-trained models.

Some examples: 

  • Qlick AWS SageMaker integration
  • PowerBI and Azure ML integration
  • Looker and Google Data Studio to BigQuery ML Integration
  • This is not completely BI platform integration, rather it’s an integration of the ML models training and execution into a warehouse, which BigQuery does.
  • It allows you to start with some ML experiments and visualizations almost at no cost.
  • While I mentioned looker and GDS, BiqQuery (and hence its ML feature) is available for many other platforms making it a decent choice to start with.

Recap

If you require something more advanced than basic forecasting, take a look at the built-in Machine Learning capabilities of the platforms/tools:

  • AutoML for PowerBI
  • BigQuery ML for google platform

Those options are a good trade-off between the quality, flexibility, and development time.

If this is not enough and you want to give your data scientists complete freedom in building the models, then take a look at the systems allowing to execute random python/R code right from the platform:

  • PowerBI desktop Python Queries
  • Tableau TabPy service
  • QlickSense server-side extensions

Or, choose a platform with integration with cloud solutions like AWS SageMaker or Azure ML. While this approach has the same benefits of building the most complex models, it’s combined with the simplicity of usage from most BI tools.
The crucial point is that all of these solutions only make sense when tight with ETL and Data Warehouse solutions. “Broscorp” collected vast amounts of experience dealing with different platforms, languages, and tools.  We can help you choose the right BI tool, perform design, architecture, and technical implementation of such a solution.


The author of the article: Mykhail Martsyniuk, Solution Architect at “Broscorp