Data science and especially machine learning and AI are often presented in the media as these hard-to-achieve results of large, expensive, high-risk projects, executed by highly trained (and highly paid) specialists. Truth is, over the past 5 years we have seen incredible maturation of ‘democratized AI’ – tools that make solving many machine learning problems easily within reach of many many data and business analysts, like the Alteryx Intelligence Suite. This is an excellent development.
Igor Holas is works as a data consultant at Inviso by Devoteam and is the Head of AI for the Devoteam Group internationally.
Truth is, over the past 5 years we have seen incredible maturation of ‘democratized AI’ – tools that make solving many machine learning problems easily within reach of many many data and business analystsIgor Holas, Head of AI
A bit of context
Machine learning (ML) is a great enabler of our data future. In many instances we see businesses on a data journey run into one, or a number of problems that are hard to solve without reaching for statistical or ML solutions. Perhaps a customer has a data-quality issue preventing them from conducting a neat merge between data sources – requiring a ‘fuzzy match’. Or they need to process a large volume of unstructured data – e.g. a large amount of news articles. Or maybe they just need a good guess about what their customers will do next – for example what item will they place in their basket next. We solved all of these problems for customers in the last two years. In all these instances machine learning, statistics, or AI offer a way to arrive at a ‘very likely’ answer to these questions and enable the client to keep moving.
Oftentimes, I see that media coverage of AI and ML still focuses on hard-to-solve bleeding-edge use cases that are extremely high-risk, high-cost to solve. But to be honest, extremely few businesses have such problems with such extreme needs.
The problems most of us run into, and can derive immense business value from, are extremely low risk and – dare I say – trivially easy to solve. In fact, solving these common problems is so low risk, we have solved them for customers under ‘no cure – no pay’ principle – and have gotten paid.
In 2021, you may no longer need consultants to solve problems with AI.
The prospects of what I am about to write, might be bad news for me as a data scientist, but as of 2021, you may no longer even require the services of a data scientist (or a data science consultant) writing code to solve your common ML problems – instead you can opt to partner with a service provider to let you (or your business analysts) build ML solutions to your data problems in incredibly easy-to-use low-code ‘point and click’ way.
The magic trick is called Auto-ML. Auto-ML is a term used to describe a tool that, given some basic information (data to learn on, clear indication what it is we are aiming to solve, and a set of correct answers) will find the best solution … automatically.
Now, in some cases, a skilled data scientist can likely build a better, more precise model, but they would need weeks of work to do this .. all the while, Auto-ML model will have a good solution ready in minutes of work and perhaps a few hours of waiting for the training process to finish. Even better, the Auto-ML services do not require the services of a Python programmer and best of all, if you do find out you require more precision than the Auto-ML model provides, a skilled data scientist can use the Auto-ML as a great starting point from which to improve your model further.
Auto-ML, and low-code platforms that make use of it, are enabling analysts who understand the principles of machine learning, and can navigate a tool like Alteryx, PowerBI, Tableau or Looker, to prepare their dataset, submit it for training, and deploy the finished model to start bringing value to their business – all while using graphical interfaces they will find rather familiar. It is some extremely powerful stuff.
Oftentimes, I see that media coverage of AI and ML still focuses on hard-to-solve bleeding-edge use cases that are extremely high-risk, high-cost to solve. But to be honest, extremely few businesses have such problems with such extreme needs. Igor Holas, Head of AI in the Devoteam Group
Platforms to consider: DataRobot, Alteryx, and Google.
First platform I want to mention is one of the granddaddies of Auto-ML – DataRobot. DataRobot has spent nearly a decade building an easy-to-use tool for analysts to build Auto-ML models, deploy them to production, and manage them as they serve their business. The tool is incredibly easy to use and, despite being very capable, manages to always clearly indicate the recommended next step so the user does not feel lost. I am a fan.
What makes 2021 the year of ‘democratized AI’ for me, however, is that two major players: Alteryx and Google have released new generations of their guided ML products showing that the market is maturing rapidly.
Alteryx has been releasing various ‘guided machine learning’ features for a while now, and they are currently grouped in its Intelligence Suite that uses a simple UI to guide a business user to either build a very powerful predictive model or let the Automatic mode perform Auto-ML on the data. (We have seen a preview of even more powerful Auto-ML features coming to Alteryx later in 2021, so stay tuned for more). Embedded in the Alteryx workflows, these predictive steps can radically expand what questions one can ask from their data.
Finally, Google has repackaged its AI offerings into Vertex AI, an incredibly powerful, yet easy-to-use AI platform. While Google’s general approach to user experience means the tool does not hold one’s hand quite as tightly as DataRobot, Vertex AI, like DataRobot, lets a user go from loading data to the system to having a functional prediction model in just a few clicks.
What makes Google Vertex AI so powerful is that it is part of the extremely powerful Google Cloud Platform which combines best in class tools like BigQuery, AI services, Cloud Functions and more.
With Vertex AI, embedded in GCP, you can perform Auto ML with ease, but you can also easily have your data be processed by one of the many Google AI services like translation, video analysis, image analysis, or Document AI (for which Devoteam is a development partner to Google). In fact one of the key value propositions, for me, of Vertex AI is that you can so easily move between the various Google AI offerings. All within the same platform, using the same interface. What is more, because of the way Google prices its Google Cloud Platform products, Vertex AI is likely to be a fraction of the cost of either DataRobot or Alteryx.
Our approach to AI at Inviso by Devoteam
At Inviso and across the entire Devoteam, we have helped a number of customers solve their data issues with data automation and machine learning, and we have helped many more be proficient with tools like Alteryx or Google Vertex AI, so they can build these solutions themselves. Whether you have one ‘simple’ data issue you need solving, or you are considering a major data journey, we can not only help you get started, help you get there, but we can also help you choose the best tools to meet your goals and train you how to use them yourself.
Also, don’t be shy to reach out to me or one of my colleagues in Inviso by Devoteam if you want to discuss your data and AI needs and vision.
