So many buzzwords, so much confusion. Automated analytics, artificial intelligence (AI)-driven BI, and automated machine learning (AutoML), aren’t these terms describing the exact same thing? NO. Although these technologies may sound or look alike in popular BI solutions, how they work behind the surface is significantly different. In this article, I’ll explain and compare key differences to help you understand when to use each one.
Automated Insights, the Company
Let’s begin by clarifying Automated Insights. Automated Insights is a company that picked a fantastic name. Automated Insights created Wordsmith, a natural language generation platform. Automated Insights creates human-sounding narratives from data, making it easy to produce real-time, written analytics, personalized reports, and stories at scale. The Wordsmith platform is utilized by companies to increase organizational data literacy and broaden adoption of business intelligence software.
Automated Exploratory Data Analysis
Unfortunately for Automated Insights, the combined terms “automated” and “insights” are frequently used to describe automated exploratory data analysis. Making things a bit more confusing, automated exploratory data analysis in modern BI tools often includes charts with natural language descriptions.
Automated exploratory data analysis typically runs aggregate queries and summary statistics on dimensional data models to generate charts. You usually cannot control what is run. Most vendors implement it as a black box. Automated exploratory analysis results include totals, averages, trends, outliers and descriptions.
Automated exploratory analysis is great for getting quick, high level insights from historical data. It suffers from limited forward-looking, predictive modeling capability due the use of dimensional data. Beware of the limitations and proper use of these types of machine generated insights found in most modern BI tools.
A dimensional model, also referred to as a star or snowflake schema, contains facts surrounded by descriptive data called dimensions. It enables accurate reporting over time and improves aggregation query performance. Dimensional design patterns were born in the 90s but still fuel most modern reporting solutions. Tableau, Qlik, TIBCO Spotfire and Power BI all use dimensional models.
Beware of fundamental design differences between BI and machine learning data models when using automated analytics.
Dimensional models are NOT designed for predictive modeling, machine learning, and AutoML. If your BI tool automates machine learning using a dimensional model, your results will most likely NOT be accurate.
Artificial intelligence (AI)-driven BI
Artificial intelligence (AI)-driven BI or AI in BI is used to describe a plethora of scenarios. If you hear a vendor using this term, it usually means that they embedded machine learning technology. Like automated exploratory analysis, embedded machine learning is a black box. You won’t be able to see how or what is being done in the background.
Examples of embedded AI in BI that you cannot control includes algorithms to automatically join two tables, detect data types, make recommendations for charts, reports, or data corrections, or create alerts on data value changes. Some vendors use AI-driven BI terms to describe automated exploratory analysis. Other vendors display machine learning model results that you can see, build, explain, and control within a BI tool. To sort through what AI in BI means, you will need ask for more details in how it was implemented and if you can control it.
Automated Machine Learning
Automated machine learning (AutoML) is the process of automating the manual end-to-end steps to develop, monitor and manage a machine learning model. Results are forward-looking answers called predictions with probabilities and explanations for a specific question. AutoML requires a machine learning ready dataset.
The depth and breadth of AutoML capabilities vary extremely widely across open source and current vendor offerings. Many vendors claim to offer AutoML but don’t automate most of the steps, model monitoring or model management. Some vendors stretch the truth a bit by calling manual drag-and-drop solutions with job schedulers AutoML. Only a few vendors offer explainable, transparent AutoML that is not a black box.
Depth and breadth of AutoML capabilities vary extremely widely across offerings.
Since AutoML is used for answering predictive questions with data, it is important that you know what you are doing, understand how it works, and see what drives different outcomes. You also need controls, auditing and governance. You don’t want your company to star in the next AI gone wrong headline news story.
To understand the magnitude of options for AutoML to avoid being bamboozled.
- Watch demos and ask questions
- Take a test drive with a trial
- Delve into automation step details
- Ask to see guardrails in action
- Check for feature engineering
- Review model management and monitoring
- Read the documentation
The 2019 Gartner Magic Quadrant for Data Science and Machine Learning Platforms and Critical Capabilities reports are also good reads. Be sure to read the commentary section for each vendor.
Machine Learning Datasets
Unlike other analytical techniques, machine learning algorithms rely on carefully curated data sources. You’ll need to organize your data into one “flattened” analytical table of variables.
Where columns in BI applications are called dimensions, columns in machine learning datasets are referred to as features. Features and feature engineering are the creative part of machine learning that requires knowledge of the data and the business.
Feature engineering improves model performance and accuracy. Some AutoML solutions do not perform automated feature engineering at all while others only automate basic feature creation. The best AutoML solutions in the market have deep feature engineering capabilities that you can visually explore and tune to generate high quality models. To learn how to prepare your data for machine learning and AutoML, please read my white paper on that topic or watch the related data prep webinar.
When to Use What
In the traditional BI to self-service BI evolution, I saw numerous reporting data messes generated by non-technical users with good intentions. Usually I got the call from a line of business sponsor that realized their team needed professional help to get them back on track since reports were no longer accurate or grew too difficult to maintain.
If you don’t have time to do it right, when will have time to do it over? -John Wooden
Over time bad data models can become expensive problems. This holds true for self-service BI, analytics automation and AutoML. Don’t lose trust with your stakeholders by underestimating the learning curve for implementing automation that looks simple in cool demos.
Novice users with good intentions become a far greater risk when dealing with machine learning and AutoML problems. Identifying suitable machine learning use cases, properly preparing data for automation, understanding good and bad results, auditing, and change management while continuing to keep a model healthy requires training and on-going mentoring from experienced experts.
Don’t let AI, automation and AutoML marketing hype confuse you. Do your homework. Learn more about how these technologies work. Get the details on what is automated and what isn’t. Most importantly…use the right tool for the job.
- Use Automated Exploratory Data Analysis to get fast, high level summaries of historical data analysis with dimensional data models
- Use AI in BI that you can’t control for a simple tasks, productivity boost while understanding the risks and limitations
- Use AI in BI that you can control to deploy governed, explainable, trustworthy AutoML models to the masses within BI tools
- Use AutoML to get predictions and probabilities with machine learning datasets
For more information on AI in BI versus AutoML, check out From Business Intelligence to Machine Intelligence: How AI Will Impact BI or visit me at one of the upcoming industry events.