[AI IN ACTION] AI in finance, medicine and retail: Boon or bane?

Industry

Published: 24 Oct. 2023, 07:00

[AI IN ACTION] AI in finance, medicine and retail: Boon or bane?

Dozens of sorting robots occupy the first floor to sort the packaged items accordingly based on its shipping regions at Coupang's fulfillment center in Daegu. [COUPANG]

With the arrival of advanced technologies like AI and big data analytics, some experts have been quick to predict that they will wipe out the jobs held by human professionals. The actual development has been unfolding in a more nuanced way in Korea.

Take robo-advisers for example. When the use of the system built upon systematic trading algorithms was approved in limited segments in 2016, industry insiders believed it could threaten the jobs of fund managers and brokers with its promise of balancing expected returns with risk to devise the optimal portfolio.

Seven years since its introduction, however, the actual return rates of the investment products managed by robo-advisers proved disappointing, plunging major robo-advisory startups into financial difficulties or even being put up for sale.

The downbeat performance underscores AI's limited capabilities in the sectors that human professionals find daunting to crack.

Still, in some sectors like medicine and retail, they flourish, but their position is supplementary to the human workforce, rather than being a replacement.

A screen capture of a robo-adviser portfolio [DECEMBER & COMPANY, SHUTTERSTOCK]

Underwhelming returns
Robo-advisers are increasingly shunned by skeptical investors as the rate of return proved to be sluggish in recent years.

Return on investment for robo-advisers programmed to be “neutral risk” was 2.43 percent in the second quarter, hovering below the 4.94 percent of the Kospi 200 index, according to quarterly data by Koscom’s Robo-Advisor Center.

Koscom is a state-run financial IT solutions company and the system manager of the Korea Exchange.

Investment returns for the same program in the first quarter was 5.01 percent, lower than the 10.63 percent for Kospi 200.

Introduced in Korea in 2016, robo-advisers use AI algorithms to make investment recommendations or actually trade for the customers, who input their interests and risk appetite for the program to process. Assets available include everything from U.S. equities to gold.

The market initially grew centered on fintech startups, but banks and brokerage firms have increasingly jumped into providing the service.

Assets under management on the service totaled 778.46 billion won ($575 million) at the end of September, down 57 percent from the same month a year ago, data by Koscom showed.

The drastic decline is attributed to some banks stopping to provide the data to Koscom starting in August, the institution explained. Excluding banks, the assets under management for robo-advisers accelerated 38 percent in the same period.

Some startups failed to make it through amid the weak stock market.

December & Company, a robo-adviser operator Fint launched in 2019, is in the process of being sold to an alternative investment firm Forest Partners after it fell into capital impairment.

Its deficit in the first quarter was 74.72 billion won, up 65 percent from a year earlier.

The deficit at Fount, a robo-adviser startup that provides solutions to banks and insurers, jumped more than 50 percent in 2022 from a year earlier to 34.1 billion won.

“The robo-adviser algorithm does not raise profits as much when the market is strong, but it defends well instead,” said Jeong Su-hyeon, a spokesperson for Fount.

“Robo-adviser’s strong points are revealed when it is managed in the long term, so it would not be accurate to conclude that the return of the service is low based on short-term indicators,” said Song In-sung, CEO at December & Company. Song added that robo-adviser’s portfolios should not be compared directly with stocks since it also includes other kinds of assets, like bonds.

Some experts think otherwise.

“The outcome shows that robo-adviser operators lack the capacity to organize portfolios and rebalance it timely according to market situations,” said Lee Sung-bok, senior research fellow of the financial services industry at the Korea Capital Market Institute (KCMI).

“In economics, at least five years is given to evaluate an asset management capacity. Despite having been active for seven years, the robo-adviser market still hasn’t reached 3 trillion won, meaning that it has not been able to satisfy consumer needs.”

It’s difficult to systematically apply robo-advisers, which uses machine learning algorithms, on the stock market, according to Lee Jeong-hwan, an associate professor at Hanyang University’s College of Economics and Finance.

“That is because machine learning fundamentally responds to news, and therefore, it is difficult to discover certain patterns,” Lee said.

Pensions raising hope
The industry, however, is raising hopes on eased regulations that would allow robo-adviser operators to be entrusted with managing retirement pension funds.

The Ministry of Economy and Finance said in July that it will push a regulatory sandbox that would allow robo-adviser operators to manage entrusted assets. It did not specify the date, but it could be as early as June next year, according to local media reports.

The retirement pension fund sums up to roughly 30 billion won.

More people would sign up for robo-advisers if the regulatory hurdle is lifted, according to Jeong from Fount. “That will increase the number of partners Fount provides the robo-advisory solutions to, which will be beneficial to our commission profit,” she added.

“The majority of people that signed up for retirement funds are passive investors that manage assets centered on risk free assets like savings,” said Yang Hoon-seok, team manager at Koscom’s Innovative Finance Technology Examination Team.

“Those people tend to be satisfied with a rate of return as low as 1 to 2 percent, and therefore, can be absorbed into the robo-adviser's customer pool, contributing to the market expansion.”

The regulatory approval could be a breakthrough for robo-adviser operators, but there are some conditions that need to be met.

“The rate of return for robo-advisers needs to be proven to be more effective than existing pension fund operators, like banks, to be able to pull in a wave of new demand,” said Lee from the KCMI.

“But based on its history, it’s doubtful whether the operators have the capacity to pull that off.”

Data-driven diagnosis
Medicine and health care are, by nature, inundated with ever-growing sets of data. And that makes these industries the primary domains where AI is making a deep-reaching impact.

AI has long become a vital component of modern health care, taking on various forms and functions, from analyzing extensive medical datasets to optimizing patient care and speeding up diagnostic processes.

As of 2022, the global AI health care market was valued at $15.1 billion and is estimated to grow more than tenfold to reach $187.95 billion by 2030, according to Statista.

Medical imaging, for instance, is one of the major areas where AI applications have proven to come in handy.

Medical imaging of VUNO's Med-DeepBrain [VUNO]

VUNO, a medical AI startup, is a prominent example.

The startup’s medical imaging tools are aimed at enhancing diagnostic accuracy and efficiency for medical professionals, using deep learning models.

Its VUNO Med-DeepBrain, which recently earned the U.S. Food and Drug Administration's (FDA) approval, parcellates the brain into more than 100 regions through deep learning and provides quantified information on the degree of atrophy in each region within a minute. This helps doctors diagnose patients with dementia stemming from major degenerative brain diseases, such as Alzheimer's disease.

VUNO’s other major focus is on biosignals.

Named as the first “breakthrough device” among medical AI devices approved by the FDA, the VUNO Med-DeepCARS is an AI-powered medical device that predicts a patient's risk of cardiac arrest within 24 hours based on data from four vital signs including breathing, blood pressure, pulse, and body temperature.

“The device is taking a step beyond such devices’ previous role as a diagnostic aide, and providing prognostic information to enhance patients’ safety,” said VUNO CEO Lee Yeha in a written statement to the Korea JoongAng Daily.

“It’s becoming an essential part of health care as the device helps to bridge gaps, especially in the case of a shortage of medical staff, and prevent unexpected cardiac arrests and acute deterioration in hospitals,” Lee added.

AI is also playing a role outside hospitals and medical facilities, often in the form of a chatbot for telemedicine services.

Dr. Now, a telemedicine service provider founded in 2020, has recently incorporated generative AI developed by a U.S. company into its chatbot service.

Dr. Now is utilizing Massachusetts-based Wecover Platform’s retrieval-augmented generation technology in its real-time consultation services, in which health care professionals answer users’ questions within five minutes.

“Dr. Now has been providing real-time, free medical consultation services since earlier this year, and through the cooperation with Wecover, the service has been upgraded to have better accuracy and speed,” said Jeon Sin-yeong, head of communications at Dr. Now.

Wecover’s technology addresses the issue of hallucination in generative AI models such as ChatGPT, as when the large language model generates false information, by integrating datasets specific to a certain area of expertise into the AI model.

Despite the recent strides made in the medical AI sector, major hurdles still remain, one being the lack of insurance coverage for AI-based medical devices.

“None of the domestic AI medical devices have yet been given a health insurance benefit code and officially entered the reimbursement system,” said VUNO CEO Lee.

The government has recently published a guideline on temporary registration for insurance coverage targeting “innovative medical technologies,” under which newly-developed technologies can be covered by insurance for a limited time period, in order to enhance accessibility to digital health care devices and services.

Hyundai Department Store's AI copywriting system, Louis, which was introduced in March, analyzed over 10,000 Hyundai Department Store ad slogans used in the last three years. This enables it to efficiently produce ideal ad copy and promotional event introductions. [HYUNDAI DEPARTMENT STORE]

AI copywriter helping human marketers
Previously, Hyundai Department Store's marketing manager Ha Jee-hoon spent over two hours daily brainstorming new ad slogans. But lately he has found more time to focus on crucial projects, as the company's AI copywriter now completes copywriting tasks in under 10 seconds.

While AI services based on simple algorithms, like chatbots and personalized product recommendations, have been prevalent, AI is now used to create advertising slogans customized for individual customers and to optimize product placement in stores and logistics centers. This trend has been adopted by leading players within the Korean retail sector, including Shinsegae, Hyundai, and Coupang.

In March, Hyundai Department Store added a new member to its team — Louis. But Louis isn't a human; it's an AI powered by Naver's HyperCLOVA AI language model. Louis has undergone rigorous training, processing over 6,500 times the amount of Korean language data than OpenAI's GPT-3. Furthermore, it focused its learning on more than 10,000 data points of advertising copies and promotional texts used by Hyundai Department Store that received favorable responses over the past three years.

“Previously, creating new ad slogans involved a lot of brainstorming, like reviewing past copy texts or similar event materials and adapting them for use,” Ha said. “However, with the introduction of Louis, we can simply input keywords like the target audience or age group, and it generates content, making the process much more efficient.”

Shinsegae I&C, the retail technology subsidiary of Shinsegae, introduced a fully-automated convenience store at the COEX shopping mall in southern Seoul in 2021 through a collaboration with Emart24 and the Ministry of Science and ICT.

The concept of checkout-free shopping is fascinating, an innovation that relies on some thirty cameras tracking shoppers, which can understandably raise privacy concerns. Shinsegae I&C assures that sensitive personal information is not collected.

“We use LiDAR technology to [analyze] purchases based on anonymized data,” said Han Su-woong, manager of the digital transformation team at Shinsegae I&C. “All in-store video data is immediately deleted upon a customer's exit to ensure privacy and data security.”

In addition, Shinsegae I&C has created a “store management platform” based on AI Vision technology. It relies on small wireless cameras placed on store shelves to track inventory and analyze sales, as well as AI-driven recommendations for product selection and placement that can boost sales.

Shinsegae I&C is currently testing the platform with an undisclosed European retail company with the goal of expanding its services internationally.

Coupang, the e-commerce giant known for its superfast 'Rocket Delivery' service, places high priority on optimizing its logistics centers to ensure swift deliveries to customers nationwide, even in remote areas.

Despite the seemingly haphazard arrangement of products throughout Coupang's logistics centers, the company strategically positions items to minimize workers' travel distances. This approach takes into account factors such as sales volume and timing for each product, all based on years of analyzing its extensive customer order data, and employing AI technology to enhance the efficiency of the goods' placement upon arrival. Employee routes continuously improve the data.

“Coupang's logistics operations are evolving into a cutting-edge digital environment, integrated with AI algorithms and robotic technology,” said Kang Han-seung, representative director of Coupang. “These investments not only enhance the work environment for employees but also significantly reduce their workloads.”

BY JIN MIN-JI, SHIN HA-NEE, SEO JI-EUN [jin.minji@joongang.co.kr]