Science fiction has always made machines look incredibly smart. With a push of a button here, turning of a dial there, feeding in a few lines and the job is done. In reality, we are moving towards it, but we are not quite there yet. Automated machine learning (AutoML) is a step in that direction, where a machine teaches itself how to come up with a solution, making technology easy to use and accessible to more people. It empowers businesses to develop customised models, which a data scientist would otherwise do by sifting through data, finding the right model, and feeding the machine the best-suited algorithm. Such is the demand to automate machine learning that the market is estimated to cross $14 million by 2030.

When automation needs to be automatic

Machine learning (ML) has propelled automation, but it needs expertise. Those applying ML tools need to know how to process data, which algorithms to use, and how models should be deployed. Using the tool itself can be a daunting task. AutoML emerges as a solution making the process more user-friendly without a specialised data scientist. In the telecom industry, for example, some firms used AutoML to build churn-management models that predicts which customers have a high risk of canceling their contracts. The idea to democratise AI engineering was one of Gartner’s top 12 strategic technology trends in 2022, which is also changing the role of data scientists and speeding up businesses. Study shows that nearly 80% of a data scientist’s time is spent preparing data for modeling. With AutoML data preparation, data scientists can be free of the laborious time-consuming work to analyse the output of the machines or drive newer innovations with AI/ML models.

Why AutoML is a step up

Companies using AutoML have experienced increased productivity. For instance, a financial services firm recorded a revenue climb from 1.5% to 4% by using AutoML to handle pricing optimisation. AutoML ensures:

· More efficiency – It reduces the training time of machine learning models thus speeding up processes.

· Cost reduction – It needs less resources to complete the work in a given period. A faster more efficient system automatically increases productivity compared to hand-coded models.

· Ensures optimisation – It tunes hyperparameters and manages a model’s future use. And reduces the risk of human error that ML may have.

Where AutoML is being used

# Fraud detection – It can improve the accuracy and precision of fraud detection models. US-based Concensus Corporation was able to improve 24% fraud detection models, reduce false positives by 55%, and cut down deployment time from a month to 8 hours.

# Healthcare – In research and development, where it can analyze large data sets and draw insights aiding in medical diagnosis management. US-based Evariant got more one-to-one interaction with clients that helped improve their customer service and spiked their return on investment

# Real estate – Singapore’s Ascendas Singbridge Group used AutoML to figure out their parking lot efficiency and recorded a 20% increase in revenue with more accurate predictions on parking lot usage

# Email marketing – Denmark’s One Marketing was able to minimise spam for customers, improve mail open rate by 14%, click rate by 24% and ticket sales by 83%.

A work in progress

The challenges of AutoML stems from the fact that it is a relatively new field and all tools are not fully developed. Hence, the expectation that we set out with it should be measured. AutoML is still not a replacement for human expertise, it should be seen as more of an assistant to data scientists and an ally of people without the knowledge of deep learning. It is also not super-fast as AutoML uses algorithms to select and optimise machine learning models, which can be time-consuming and labour-intensive in the beginning. While it can select the best machine learning algorithm for a task, it cannot always optimise the required hyperparameters.

The process requires intervention to understand the attributes of domain-specific data, to define prediction problems, and to create a suitable training data set. These steps can drag out and needs to be rechecked continuously by domain experts and data scientists, delaying the AutoML systems from being completely automatic.

How to get into the fast lane?

With these bumps in mind, it is critical to start the AutoML journey right.

Review your needs – The first step is to decide which tasks are for AutoML practitioners and which need data scientists.

Upskilling is the key to keeping up with technology – Those handling AutoML tools should have some training and understanding of how data science works

Set reasonable goals – Being aware of the limitations of AutoML can help optimise it as well. Standardised tasks can be handed over to machines but specialised knowledge will always come from the human mind.

Looking back, the leap technology takes in a decade is ginormous, but it all begins one step at a time and AutoML is one of those important stepping stones to get on the automation bandwagon.