AutoML platforms offer “no-code” AI development, but the devil is in the details.
For most Business Intelligence professionals, the world of AI and Machine Learning(ML) seems a bit out of reach. The challenge is not so much in whether the technology is useful or not, but rather in the effort required to add AI and ML technology to their BI stacks. The requirement from business users is undoubtedly there. Whether it’s to predict customer churn, model marketing campaign performance, forecast sales, identify clients at high risk of defaulting on receivables, or countless other applications, adding AI/ML to your BI stack can provide immense value. The average BI professional’s problem is that although they are highly skilled at manipulating data and creating sophisticated visualizations, applying the additional data optimization and statistical mathematics necessary to build effective AI/ML models is not within their skill-set.
The promise of AutoML is straightforward and powerful. Simply put, it promises to take the difficulty out of transforming data into AI/ML algorithms. However, BI professionals’ challenge is that even AutoML fails to deliver on the promise of providing a real “No Code” AI development environment. The reason for this is simple – While AutoML platforms do an admirable job of automating the process of optimizing ML-ready data and selecting the most appropriate algorithms for any given use-case, they fail to deliver true end-to-end automation. To understand the deficiencies of traditional AutoML platforms, we must first understand the nature of AI/ML development. While there are multiple stages in the development of AI/ML models, traditional AutoML platforms concern themselves only with the final part – the optimization of input data and the selection and optimization of the ML algorithms. Enterprise data, however, is hardly ever “ML ready.”
As manipulated and carefully choreographed as it is, most BI-ready data is also seldom ready for ML. There are two significant hurdles that BI teams must overcome, even with traditional AutoML platforms: First, consolidating the most appropriate, most relevant columns from the hundreds, often thousands of tables of data in a data warehouse into one, single flat table. This process, known as Feature Engineering, is time-consuming, repetitive and requires a great deal of trial and error as teams test specific hypotheses of which tables might best fit a given model – only to go back and try again. The second roadblock is that a fair amount of “AI-focused” data prep must also occur to make the data more “AI friendly.”
No-Code AI: Almost There
Beyond pure “AutoML” platforms exist a different type of products – so-called “No-Code” AI platforms. Several providers all share a similar approach: A workflow-driven approach that requires a user to create a visual “flow” using symbols and connectors to essentially “instruct” the system. These products are certainly more straightforward to use than building tables manually and then applying AutoML processes, but they lack in one significant area: Speed of development. The problem is that although simple workflows are easy to build and conceptualize, the reality is that most AI/ML models require large, very complex, and sophisticated workflows that quickly become unwieldy and create a whole new set of challenges of their own. In a perfect world, a platform should connect to a central data set, accept a few simple parameters, and provide ready-built AI/ML algorithms that are ready to implement. That’s certainly not possible with any workflow-driven No-Code platforms.
AutoML 2.0: True No-Code AI Development
AutoML 2.0 platforms are a new entry into the field of AI/ML development that promise to provide a radically new experience in the way BI teams can develop, test, and deploy AI/ML models. With AutoML 2.0, users simply connect to an existing data set – even a complex data warehouse – and the system leverages AI to automatically discover and evaluate table combinations that are likely useful in building AI models. This “AI-Powered” feature engineering is at the heart of how AutoML 2.0 platforms work and can dramatically accelerate and simplify the process of developing AI and ML models. dotData clients have been able to go from “no AI experience” to having models in production in as little as six weeks. The power of AutoML 2.0 is that it finally begins to deliver on the promise of “one-click” no-code development with an environment that automates 100% of the workflow. AutoML 2.0 is giving BI teams the ability to satisfy user requirements for predictive BI systems without learning complex new methodologies or hiring armies of data scientists.
Here is an infographic that describes the differences between the manual machine learning approach versus no-code AI using AutoML 2.0.