Easing Talent Crunch With AutoML: A Real-World Case

Image credit: iStockphoto/metamorworks

Every company understands the value that a data scientist brings to the table. The problem is finding them. This is why more companies are deploying augmented analytics and AutoML tools and steering some data scientist roles toward the business domain users.

This trend to democratize AI engineering was also one of Gartner’s top 12 strategic technology trends in 2022. It is changing the role of data scientists and speeding up businesses into the journey of a data-centric enterprise.

It is what happened at global automotive distributor and retailer Inchcape. At the recent Chief Digital & Data Officer Asia Summit, Inchcape’s chief data scientist and global head of data science Ram Thilak shared his experience of driving self-service data science within the organization.

Keeping up the sales margin

The shortage of data scientists is real. While most companies struggle to meet the demand for data science and AI modeling, Thilak said Inchcape tried to curb this demand by upskilling business users and to turn towards automation tools.

With a team of less than 15 data scientists to serve 40 markets worldwide, Thilak said if the company relied only on the data scientists, its data and analytics strategy would never be successful. So, the team empowered more business executives to use a dashboard to access information and make predictions.

“We want to encourage every business user to start looking at [analytics] modeling,” he said. “We started in Asia as a pilot [program].”

The program focused on supporting executives that manage sales and operation planning to tackle the company’s pressing challenge — managing monthly stocks. Inventory management in the automotive industry is a unique challenge.  The sheer volume of vehicle models, colors, and features brings endless options for customers. And with more consumers spending time researching a vehicle before purchasing, its unavailability means sales lost. Meanwhile, the dealer could have hundreds of cars for sale, taking up inventory space and reducing the profit margin.  

By guiding executives that manage sales and operations to use self-service analytic tools and incentives — with initiatives like the “eight stock predictions”, an exercise that forecasted demand and supply trends — the company was able to contain excess stocks within three years.

“It’s about empowering them and ensuring that they can make the decision, so they don’t have to knock at the doors of the data scientist team for [every] decision,” said Thilak.

Right tools for the right audience

To support Inchcape in democratizing analytics, Thilak said AutoML and other self-service machine learning (ML) tools played an important role.

“I’d say first is [to look for] low-code/no-code [environments],” he said when asked about evaluating analytics and ML tools. “Secondly, how intuitive it is and how easy it is for people to use.”

He also suggested providing tools within an environment that is familiar to the users. “There are tools I called ‘Excel on steroids.’ It allows you to do a powerful version of the pivot table or connect three to four excel sheets to get insights. These are probably the tools we have for employees.”

“Right tools for the right audience,” added Jon Teo, domain expert for data governance & privacy, APAC and Japan at Informatica. He said it is essential to provide familiar tools to the users. “We’ve seen examples of people downloading the data or printing dashboards. These are different forms of self-service. It’s not always high-end.”

With most AutoML tools offering the engineering and selection of analytics models, he suggested businesses support users with better data governance. By adding access to data sources and ownership of the analytics environment, business users can better understand the background to ensure data integrity.

“From the governance standpoint, we are interested around making sure people understand the data [source and terms of use],” he said.

Creating the right environment

Training is another central element in democratizing AI modeling. But according to Thilak, cultivating an environment to drive curiosity with data is more important.

“It’s never easy to quantify qualification, but what you can do is to create that curiosity within the organization and [offer] guidelines,” he said. “At least when they first hit the roadblock, they are not afraid.”

At Inchcape, Thilak’s team also develop a mentor program for business users, providing more hand-holding support and encouragement.

Nevertheless, he added that ensuring a technically ready environment is equally essential. The environment will be prepared to proceed to the modeling stage with proper documentation of the processes and a sanity check on data integrity.

Drift, outlier, and governance

Market changes and business objectives often follow. This means AI models will also require adjustments. The biggest challenge when automating data science and AI modeling is governance and the ability to identify outliers.

This was what happened at Inchcape during COVID-19. When the market was shocked and the sales turned stale, none of the forecasting models could predict or make appropriate projections.

“We [the data scientist team] tried to encourage and give tips to the business users, how they can contain the effect,” said Thilak. “It’s important to empower everyone, but at the same time have a governance structure to ensure that things don’t go out of hand.”

Teo added some tools are available to assess and monitor data drift. He also suggested developing data drift scores and sharing them, allowing data consumers to understand the data asset better.

The risk of data drift and AI outlier is real as more of our businesses and the government rely on it, noted Lawrence Liew, director for AI Innovation at AI Singapore. This national program aligns industry players, academia, and the government to catalyze, synergize and boost the country’s AI capabilities.

Liew added data is the foundation and the program’s guidelines also focus on ensuring AI governance through enabling data accuracy.

“If data is not clean or accurate, even BI projects will fail,” said Liew. “Having clean and accurate data is key to business survival.”

Sheila Lam is the contributing editor of CDOTrends. Covering IT for 20 years as a journalist, she has witnessed the emergence, hype, and maturity of different technologies but is always excited about what's next. You can reach her at [email protected].

Image credit: iStockphoto/metamorworks