Automation is the new buzzword in 2022. Forbes Technology Council listed automation as a critical focus for enterprises to reduce workload and error.
But that is not enough. With the pervasive use of AI and analytics, some enterprises take one step further to automate the automation. The idea is to democratize technologies among knowledge workers by bringing automation into complex tasks like data processing and building ML models. So, businesses can drive end-to-end automation by using automatically developed analytics.
At the recent CDOTrends Digi-Live! Summit Series, data analytics experts and technology practitioners discussed odds and ends to bring the vision of automating analytics into reality.
Automate to democratize
One reason automation is catching attention, according to Suganthi Shivkumar, vice president for Asia at Alteryx, is a result of the post-pandemic isolation era. In addition to enabling business resiliency during the crisis, automation has drastically raised the awareness of democratizing data.
IDC’s study indicated that 94% of Asia Pacific business and digital leaders agreed data fluency is essential for their organizations. But only 19% are considered an expert in this area.
“What drives this data fluency? The answer is simple, the more people who are on board in this data journey, the more mature the organization becomes and the more return it gets,” said Shivkumar. “It has become a non-negotiable need for data analytics to be a pervasive workforce skill set.”
Aiming to empower general knowledge workers to be involved in the data analytics journey, enterprises are looking for tools to bring end-to-end analytics automation. By bringing business domain expertise closer to the data and analytics processes, Shivkumar said enterprises could also create value faster.
“We increasingly see our clients spend millions on the exclusive few — the data scientist and elitist community — but still struggle to see the value,” she said. “The bulk of the people that drive insights and breakthroughs are the citizen data scientists or the knowledge workers, who typically don’t have access to the data analytical tools.”
Citizen-level data scientists
This is precisely what the job search platform Hiredly is looking for in its data analytics journey.
“As a start, we don’t need Ph.D.-level data scientists; we need a citizen level,” said ThenHui Chong, chief technology officer at Hiredly. “How can we enable the non-technical business users to understand data? It’s very important to build this data-driven mindset within the organization.”
Chong added many organizations practice data analytics using manual processes. To simply understand a pattern or identify a trend, business users must request data from different systems. He noted there are great opportunities for automation.
“A platform that gives users a single point of access… including (the automation of) how we collect data, setting the ETL process from different data sources, will be convenient,” he said. “In this case, the on-boarding or adoption of the data-driven culture could be much faster.”
A healthy dose of skepticism
While it is an admirable ambition to democratize data analytics, we are still far from bringing end-to-end automation of analytics into reality.
“We need to have a healthy dose of skepticism around exactly what can be automated,” said Lee Sarki, head of data analytics (life & health), AP, Middle East, and Africa, Munich Re.
As a provider of reinsurance, primary insurance, and insurance-related risk solutions, Munich Re’s business heavily depends on its risk assessment and analytic models. Sarki added when the cost of model error brings financial consequences to the business, the use of automation and expectations from citizen data scientists require careful planning.
“We need to be realistic about citizen data scientists,” he said. “It’s one thing to talk about descriptive analysis, reporting, and Power BI; it’s a very different thing to talk about ML models and make automated decisions with real consequences.”
On top of having a realistic expectation from general knowledge workers to deliver value from data analytics on top of their daily jobs, Sarkin noted businesses also need to be aware of the implication of ML models governance and accuracy from citizen data scientists.
“(AI) Bias, fairness, ethics, and transparency are not a soft topic,” he said. “It is very much a risk management and reputation issue.”
Levels of data scientists
Nevertheless, all the panelists agreed businesses should still embrace a data-driven business culture and nurture citizen data scientists. Sarkin added that different types of domain knowledge workers could contribute differently to the organization’s data science strategy.
At Munich Re, he said a global analytic curriculum training is available as part of the company’s consistent effort to build a data-driven culture. These training sessions also allow the client-facing teams to explain their risk assessment models.
For more advanced domain experts like the insurance actuaries, Sarkin said they are in a better position and have a more technical background to be involved in the development of risk assessment models.
Deep knowledge of different data source systems is another essential expertise. He said before model development, the analytics team needs to ensure data integrity with a granular level of understanding of different data sets and how they are related and inter-connected.
“As much as we need to bring in the domain experts, sometimes with an underwriter, they are not necessarily a domain expert in the source systems and the business processes,” he said. “If you don’t really get a grasp of that, everything downstream is impacted.”
Among the expert-level data scientists, they focus on developing risk models, pricing models, and predictive analytics. As the center of excellence in data analytics, Sarkin said the team is also responsible for the accuracy, transparency, and governance of different ML models.
“Within my team, we have responsible AI capabilities dashboards and framework,” he said. “We monitor the (models’ performance towards) data science KPIs and translate them into business and risk implications.”
Human augmented automation
Meanwhile, the expertise level of data scientists varies between businesses. Shivkumar from Alteryx noted that some organizations have a higher tolerance for model errors. These organizations can benefit from using automation to bring business domain workers close to analytics.
She added the automation tools range from data engineering design, data loading pipeline, and orchestrating data flow. Combined with a no-code interface, these tools allow business analysts to drag-and-drop data blocks to develop models and analyses.
Sarkin also agreed automation plays an increasingly important role in model deployment, validation, and maintenance. He said this is when MLOps tools are helpful to test and verify the ML model's accuracy and support the maintenance and retraining.
“Retraining and managing models in an agile way is important,” he said. “If you cannot intervene early and retrain (models), the business will be under threat. I believe that MLOps technology is now part of risk management; it’s no longer just process management.”
Despite the rise of automation, panelists agreed humans must be in the loop to augment the development and the use of analytics to ensure its integrity.
“We must have a risk culture around AI to appreciate consequences to error. We must have a trajectory that’s sustainable and responsible; otherwise, AI will get a bad rep unnecessarily,” Sarkin concluded.
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/Koonyongyut