Featured faculty: Tony Ke and Liao Chenxi
Written by Putro Harnowo

Generative AI, or GenAI, has rapidly changed how consumers find products. AI chatbots like ChatGPT, DeepSeek, and Claude AI have become the go-to tools for daily inquiries, much like Google in the old days, while traditional search engines like Google and Bing are stepping up by displaying AI-generated answers at the top of search results.

No surprise that American software company Adobe reported traffic from AI sources to retail sites surged by almost seven times in 2025 compared to the previous year. This shift has propelled a new digital strategy called generative engine optimisation (GEO), which aims to ensure the product shows up when users ask an AI a specific question.

Since GenAI is trained on massive amounts of data, companies can increase their visibility by creating synthetic content, or favourable product mentions across articles, blogs, social media, online forums, and more. The goal is to let AI models learn that their products are desirable, thereby increasing the likelihood of being recommended.

“A particularly powerful GEO strategy is deliberately crafting synthetic content so that GenAI learns and reproduces desired associations between consumer questions and product offerings,” says Liao Chenxi, Associate Professor of Marketing at the Chinese University of Hong Kong (CUHK) Business School.

For instance, a brand covertly creates blogs or articles on the top 10 products in the market but disproportionately highlights its own products, or posts a question on forums like Quora or Reddit and then uses another account to give a detailed answer recommending its own products.

Not to be confused with fake reviews, synthetic content is strategically created to influence GEO by feeding information that can be easily parsed by AI engines without explicitly targeting consumers. When content becomes misleading to consumers or includes fabricated claims, such as fake reviews, it can be considered deceitful.

In fact, the abundance of fake reviews has driven regulators in the US and the UK to ban them altogether. While it is not meant to mislead consumers, at its core, synthetic content aims to influence GenAI, which some consumers may consider as manipulation.

In her new study titled Synthetic corpus and consideration manipulation in generative engine optimisation, Professor Liao finds that, despite helping increase visibility, firms caught red-handed deploying synthetic content are perceived by consumers as lower quality. However, banning it entirely may backfire, as synthetic content can help consumers discover good products.

“AI engines work like a ‘black box’, but some consumers have an expectation that content manipulation exists on the internet, so they do not always blindly follow what GenAI recommends.”

Professor Liao Chenxi

Mapping player interactions with GenAI

In collaboration with Tony Ke, Professor of the Department of Marketing, and Xu Xiaoyan of Southwestern University of Finance and Economics, Professor Liao uses game theory to examine how firms strategically deploy synthetic content to influence consumers and GenAI, and how consumers respond to the firms’ actions.

The game-theoretic model in the study involves a firm selling a product of either high or low quality, GenAI, and consumers. Quality products naturally attract plenty of positive reviews, so a high-quality firm can rely on genuine, organic content. A low-quality firm has little organic content, so it must create synthetic content to make its product appear more attractive.

GenAI treats all content, whether organic or synthetic, as input data for its recommendation system. If the content provides sufficient evidence of a useful product, GenAI will recommend it. Consumers choose the product based on perceived quality, but are unaware of its true quality.

Before deciding, consumers go through a three-stage purchase funnel. The first is the consideration stage, in which consumers discover products. In traditional digital marketing, search engines and ads serve as beacons to discover the products, but consumers nowadays turn to GenAI.

Next, in the evaluation stage, consumers typically check other sources to confirm product quality. Consumers who consider GenAI recommendations then conduct independent evaluations. While some consumers fully rely on AI, the model focuses on how AI influences the options consumers consider. The same content that drew the AI’s attention serves as evidence for consumers to decide.

However, consumers are not entirely fooled. Some still cross-check with third-party sources and verify product quality through their own inspection. If they find out the content recommended by GenAI is synthetic, they will perceive the featured product negatively, regardless of its actual quality. GenAI would not be blamed, as consumers use it primarily to find products.

“After initial discovery with the help of AI, consumers have many independent channels to evaluate the product,” says Professor Liao. “AI engines work like a ‘black box’, but some consumers have an expectation that content manipulation exists on the internet, so they do not always blindly follow what GenAI recommends.”

In the final stage, consumers have formed expectations about the product’s quality. When the perceived quality matches personal preferences, consumers are more likely to buy it.

To manipulate or not to manipulate?

Synthetic content may successfully trick AI models, but it is not always effective on consumers. If consumers are good at spotting it, high-quality firms can keep their reputations by avoiding it, but risk being thwarted by synthetic content from low-quality rivals. High-quality firms can still leverage synthetic content, but the benefits depend on other market players.

If low-quality firms promote extremely poor products, allowing synthetic content is detrimental as it can harm consumers. For high-quality firms, the reputational damage from being caught using synthetic content becomes so severe that they risk being mistaken for having awful products, too.

However, when low-quality rivals promote moderate-quality products, synthetic content from high- and low-quality firms broadens market demand and helps all products be recommended by GenAI without risking severe reputational damage if consumers find out. High-quality products that lack organic buzz can benefit from a synthetic nudge to be endorsed by GenAI.

Ultimately, given that being spotted using synthetic content may lower consumer trust, Professor Liao emphasises that the strategy should be applied wisely. For firms with genuinely high-quality products, another way to improve GEO is to optimise organic content by encouraging satisfied customers to leave reviews and fostering genuine discussions on social media and online forums.

“When real consumers spontaneously praise a product online, it naturally provides quality data that AI engines incorporate in their algorithms, increasing the recommendation probability,” she adds. “Companies absolutely need to pay attention to GEO to stay visible, but it doesn’t mean that it can substitute traditional organic signals. In essence, GEO is a marketing tool for visibility. To build real trust, firms must make sure their product quality matches their claims.”

The post Can you truly trust AI recommendations? first appeared on China Business Knowledge.

Featured faculty: Philip Zhang Renyu and Jenny Jin
Written by Joanne Madrid

Large language models or LLMs, the AI systems behind chatbots like ChatGPT, are increasingly being used as stand-ins for humans in behavioural research. Dubbed synthetic surveys or silicon sampling, people can now ask AI to complete surveys at a fraction of the cost and time of human participants.

AI startups like Simile, Aaru, Artificial Societies, Uxia, and many others offer human behaviour simulations to predict market trends, consumer preferences, and even political views. This synthetic survey is a direct application of social simulation, in which LLMs act as humans in experimental settings.

Yet there is a catch. Critics say that social simulation may not be a reliable substitute for human respondents, as AI sometimes struggles to capture nuanced opinions and shows bias due to its vast training data. Although LLMs have successfully reproduced simple experiments, they often fail to behave like real humans in complex scenarios.

According to Philip Zhang Renyu, Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School, such limits may stem from a deeper structural challenge in how LLMs understand and reason about their tasks.

“LLMs in behavioural and social simulations can fail not only because of flawed reasoning, but also because of the limitations in how they interpret their tasks,” he says. “Even if they understand the tasks, their reasoning process may not align with human thinking.”

Professor Zhang’s latest study, Improving behavioural alignment in LLM social simulations via context formation and navigation, co-authored with Jenny Jin, Assistant Professor in the same department and their PhD student Kong Letian, proposes a two-stage method to improve LLMs’ ability to simulate human behaviour.

Beyond clever prompts

Most AI users engaged in prompt engineering have already recognised the importance of clear, specific instructions, but many popular prompts still fall short in social simulations. Therefore, the study aims to find whether a more comprehensive approach is more effective.

Specifically, Professor Zhang and the team turn to a classic Human Problem-Solving Theory developed by Herbert Simon and Allen Newell. Both celebrated scientists, with the help of programmer Cliff Shaw, developed Logic Theorist, the first computer programme designed to perform automated reasoning that marked the birth of AI.

Simon and Newell argue that humans solve problems by forming a mental picture of the task and selectively searching for solutions within it. Inspired by this theory, the team then develops a two-stage framework to better align LLMs with human behaviours in social simulations.

The first stage is context formation, or explicitly specifying experimental design so LLMs can understand the tasks and their background by defining their goals, what information matters, and how all the pieces fit together. The second is context navigation, or guiding the reasoning process so LLMs can further strategise, update their assessments, and weigh the best options.

For LLMs to behave like humans, Professor Zhang notes that the two-stage framework, context formation and context navigation, should be clearly defined so the AI understands the task and its role perfectly.

“Even really smart AI still needs a crystal-clear instruction to make good decisions. While AI sometimes can reason autonomously in simpler tasks, context navigation remains crucial in complex scenarios to choose the most relevant behavioural signals.”

Professor Philip Zhang Renyu

Comparison with popular prompts

Current popular prompting techniques can improve task performance but are not always suitable for social simulations. For instance, chain-of-thought prompting focuses on step-by-step instructions, yet if the AI misunderstands the rules and setup, it often yields inconsistent results. Few-shot prompting guides the AI model with examples, but it may not be reliable when the social experiment differs from the examples.

Tree-of-thought prompting asks the AI model to explore several paths to choose the best solution, but it fails to determine which one is most human-like. Meta-prompting can organise reasoning by designing a reusable prompt, but still requires humans to specify the setting and the standard for a valid result.

“Context formation provides a systematic way to explain the rules and setup of an experiment to the AI models, which helps it understand and adapt to specific scenarios rather than relying on a few examples or generic instructions,” Professor Zhang says.

“Context navigation directs the AI models selectively towards the crucial information and strategic considerations, rather than requiring an open search across reasoning paths. These stages help the AI figure out why it still mismatches human behaviour and see if it misunderstood the setting, if its reasoning was incorrect, or both.”

Putting the method to the test

To evaluate the framework, the team tested it across three existing social experiments with publicly available human data to directly compare the method against real human choices. The instructions in these experiments were then given step by step, as in chain-of-thought prompting, to the most capable LLMs at the time — GPT-4o, GPT-5, Claude 4 Sonnet Thinking, and DeepSeek-R1 — to see how far they replicated human behaviour.

The first human experiment found that the waiting time to get a product can signal its quality. When the same instructions in the experiment were fed to LLMs, all models boiled the problem down to a cost analysis and ignored social cues humans perceived.

Adding context formation that some queuing buyers already knew the product’s quality helped the LLMs understand the instruction, but still could not replicate human behaviour. Only after providing context navigation, that queue length can be a meaningful cue for quality, did the LLMs start to act like human shoppers.

The second experiment asked humans to choose whether to invest in a project without knowing its true quality. Some project owners offered investors a refund if they failed to reach full funding, which human investors viewed as a signal of a good project.

However, even after being provided with investment rules, all LLMs significantly funded fewer good projects than humans. Only when the LLMs received clear context navigation to interpret the refund scheme as a positive signal did they replicate human investors.

Not every situation demanded both steps. The third experiment simply found that human purchases decreased as the price increased. Oddly, upon receiving the instructions, LLMs bought more when the price rose, and as the cost soared further, they eventually reduced their purchases.

When LLMs were given a clear explanation of the experimental setup, they successfully replicated human behaviour. Here, providing context formation was sufficient. The contrast points to a useful rule of thumb: simple choices often need only context formation, while more complex scenarios involving several parties require a two-stage method.

As AI models improve, explicit guidance on how to reason may matter less for simpler tasks since the models can work it out for themselves, but this doesn’t mean that context navigation becomes less important. “Even really smart AI still needs a crystal-clear instruction to make good decisions,” Professor Zhang adds.

“While AI sometimes can reason autonomously in simpler tasks, context navigation remains crucial in complex scenarios to choose the most relevant behavioural signals.”

The post Can AI really mimic our decision-making? first appeared on China Business Knowledge.

Featured faculty: Kim Keongtae and Michael Zhang
Written by Pan Jingyi

AI-powered coding assistants are changing how software is built. These tools can be highly effective in helping programmers and software developers, even those with less experience, to write code in a programming language. Almost everyone can now make an app or software from scratch by simply asking an AI chatbot, a practice known as vibe coding.

OpenAI’s release of Codex in 2021 was the first AI model powerful enough to build entire programmes from simple text prompts. The company then partnered with GitHub, a platform where developers worldwide can store and collaborate on their software code, to launch GitHub Copilot as an autocomplete tool. In early 2023, Copilot was transformed into Copilot X, a full AI coding assistant.

Nowadays, Copilot X is not the only tool in the market. Other major AI companies, such as Anthropic and DeepSeek, as well as big tech like Google, Amazon, and Meta, have also launched their AI-assisted software development tools. As millions of developers start relying on AI, burning questions loom: have these tools changed how developers code, and can they actually boost productivity?

“AI technology may increase the developers’ productivity and push the limits of what developers feel able to attempt, but these gains come with trade-offs for code quality and originality.”

Professor Kim Keongtae

Kim Keongtae, Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School, collaborates with Michael Zhang, the Wei Lun Professor of Business AI in the same department, their PhD student Li Xinyu, and Francis Joseph Costello of Nova School of Business and Economics, to seek answers. In the study titled Exploring altered open source software development patterns in a time of generative AI, they examine 1,350 open-source software developers on GitHub before and after Copilot X was released.

The analysis finds that CopilotX significantly helps occasional contributors who were previously less active on GitHub. These casual developers participated in 48 per cent more open-source projects and became far more likely to experiment with new programming languages after using the AI tool.

Meanwhile, highly active or more seasoned developers who used Copilot X updated their code 12 per cent more often and contributed to six per cent more projects compared to those who didn’t use the tool. AI not only helps developers work faster but also spreads their effort across a larger set of projects and switches between different projects more easily.

Contributing to open-source projects sometimes requires developers to learn an unfamiliar codebase and figure out how the software is structured before making a small update. For occasional contributors, these barriers can be enough to stop them from participating at all.

“AI coding assistants can lower these barriers and make the first step easier,” says Professor Kim. “After conversational AI coding assistance became available, developers became more willing to explore new programming languages and projects.”

The hidden costs of easy coding

While productivity rose, a noticeable increase in copy-pasted code appeared. Although more projects are being started, they are not necessarily leading to more original work. “AI technology may increase the developers’ productivity and push the limits of what developers feel able to attempt, but these gains come with trade-offs for code quality and originality,” Professor Kim adds.

“Copied code is not always harmful, as some reuse is normal in software development. The concern is that if developers insert larger blocks of code without fully understanding how they interact with the rest of the project, the project may accumulate technical debt or a shaky foundation.”

The build-up of hidden problems could make software harder to maintain, debug, or improve later. It can also lead to compatibility issues, inconsistent coding styles, and more rework down the line. For project managers, this finding calls for treating AI-generated code as a draft, not as final output.

AI can boost careers

Open-source software development relies on voluntary effort, so developers have to contribute their personal time. While some programmers and software developers participate for their own interests, GitHub functions much like a public portfolio to showcase recent work and collaborations. Employers also often look at GitHub profiles to assess software developers’ skills and experience.

Therefore, to further examine whether changes in how developers worked with AI affected their careers, the researchers matched developers’ GitHub histories with their LinkedIn profiles and found encouraging signs. Developers with access to Copilot X were more likely to experience internal transitions and promotions in the following six to 12 months.

It appears that increases in open-source contributions and proficiency with new technologies are associated with positive career outcomes. “Visible output on GitHub can also signal employers that a developer is adapting to new tools and can leverage them to deliver more work,” Professor Kim says.

“However, the benefits of AI coding tools for career improvement are temporary. When only a few developers know how to use these tools really well, they’ll be noticeably more productive than others. As these tools become widely used, simply using AI won’t make a developer stand out as much anymore.”

How can developers stand out in the AI age?

Given the short-term benefits and the increasingly prevalent copied code, Professor Kim warns developers that AI should only be used as a learning partner and assistant. Although AI may accelerate software development, it cannot replace the developer’s responsibility to understand and evaluate the system. “Take the AI tool as a tutor rather than a shortcut. The human developer should remain responsible for the final judgement.”

This advice is especially crucial for beginners, as less experienced software developers are more likely to use code suggested by AI tools, even if the code is not very good or has mistakes.

RELATED ARTICLE

How retail investors use GenAI to navigate stock markets

As technologies evolve rapidly, AI tools can play a vital role in ensuring software quality and in helping spot critical issues during development, including reviewing updates that might introduce risks and identifying areas that need more testing. Highly skilled developers can leverage these tools to validate code outputs, establish clear coding rules, and integrate code into complex systems.

Ultimately, Professor Kim anticipates that the more sustainable career advantage will likely come from higher-level capabilities, such as understanding software’s underlying problems, evaluating AI output, designing maintainable systems, and coordinating human-AI workflows.

“The career advantage from knowing how to use AI coding tools will not vanish but shift towards those who can use it in more advanced, strategic, and complex ways.”

The post Is AI creating more productive but dull programmers? first appeared on China Business Knowledge.

Featured faculty: Chen Zhi
Written by Ellis Ng

Today’s AI chatbots are fueled by large language models (LLMs) trained on vast expanses of human data, from books, news articles, company reports, and social media conversations. However, as corporate reliance on these tools grows, so does the threat of embedded biases and critical errors.

An October 2025 report from the BBC found that nearly half of AI-generated responses contain inaccuracies. Last March, an engineer at Meta accidentally leaked sensitive data after following a suggestion from an AI agent. These high-profile cases prompt a critical question: What if these powerful AIs, trained on everything humans have ever created, also pick up human flaws?

“LLMs don’t just mirror human biases, but also often amplify them,” says Chen Zhi, Associate Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School. “Our experiments show that LLMs consistently replicate the classic too-low or too-high ordering bias well-documented in humans, ordering too little in high-margin scenarios and too much in low-margin ones.”

A study Professor Chen co-authored with Liu Jifei and Zhong Yuanguang of South China University of Technology, Large language newsvendor: Decision biases and cognitive mechanisms, tests LLMs on the “newsvendor problem”, a decision-making problem in which an AI must stock a resource before knowing the actual demand. The results show that AI often overshot the mark and performed worse than humans. The smarter the model, the more likely it was to act irrationally.

The paradox of intelligence

Professor Chen and the team ran multi-round experiments with 15 decision rounds each: LLaMA-8B, GPT-4, and GPT-4o. Each AI had to decide how much product to order, just as a store owner would try to guess what customers would buy and how much. After each round, the AI received feedback on what customers actually demanded and how much profit it made.

The entire process was repeated for three scenarios with varying customer demand and conditions: one with no guidance, one with a mathematical formula for calculating the optimal order quantity, and the last with a special “no-risk” scenario in which every order placed was guaranteed to turn profits.

“The most sophisticated model by the time we conduct our study, GPT-4, actually showed the greatest irrationality,” Professor Chen says. “It would correctly compute the optimal order quantity in its reasoning, then talk itself out of it through elaborate ‘risk management’ adjustments, even in settings with zero financial risk.”

“The most sophisticated model by the time we conduct our study, GPT-4, actually showed the greatest irrationality.”

Professor Chen Zhi

Each model failed for a different reason. For instance, GPT-4 over-ordered by 70 per cent more than humans in low-margin scenarios. These errors are not due to the AI being cautious and overthinking, but stemming from deep-seated information-processing, as its deliberation undermined its own best decisions, even when there was no chance of losing money.

Meanwhile, LLaMA-8B lacked the computational power to consistently apply rules, leading to erratic choices. GPT-4o was built for speed, stuck closely to simple formulas, and made steady decisions. “The three models illustrate three distinct failures: over-analysis (GPT-4), rigid heuristic adherence (GPT-4o), and fundamental computational limitations (LLaMA-8B),” Professor Chen adds.

The researchers call this the paradox of intelligence, where more brainpower does not guarantee better results. All models were anchored by their first scenario, carrying the bias even after many rounds, but also fixated on the most recent orders. In the most volatile scenario, LLMs reacted to the latest data point by changing their orders all the time. By comparison, humans did this fewer than four times out of ten.

Such errors can be traced back to how the models read and process text. LLMs take in information one piece at a time, and this sequential reading spills into decision-making. The models either cling to whatever they encountered first or overreact to whatever arrived last.

Giving AI clear rules and more adjustments

There are several practical strategies to curb these biases, according to Professor Chen. Providing clear and structured prompting is highly effective. When GPT-4o was given the optimal formula and clear rules, it applied them directly and achieved near-perfect results with minor deviations.

“Providing explicit optimal formulas significantly improved performance,” he adds. “Well-designed prompts can constrain bias without requiring any modification to the models.”

Another way is to choose LLMs carefully by matching them to the task rather than defaulting to the most powerful option. “Model selection matters enormously,” he says. “Efficiency-optimised models can outperform more complex ones on well-defined optimisation tasks, so managers should match an AI model to the task rather than defaulting to the most capable option.”

Fine-tuning also offers a promising solution. Professor Chen explains that fine-tuning LLM, where the model is further trained on a more specific dataset to adapt to a particular task, could improve performance. “Finally, human-in-the-loop oversight remains essential, especially for detecting cases where a model computes the right answer but then overrides it with heuristic adjustments,” he adds.

Do we really need smarter AIs?

Professor Chen and his fellow researchers have continued testing newer LLMs and expanded their experiments to other scenarios. He notes that biases persist in newer LLMs, but the driving factors are not always the same as those seen in the initial study. Some models have continued to overthink like GPT-4, but in different ways, and new kinds of errors have appeared.

“Firms should not assume that a more advanced or expensive model will produce better operational decisions,” he says. “AI biases are systematically shaped by their underlying architecture, and the specific contours of these biases evolve as model architectures change.”

When models compute the right answer but then talk themselves out of it, humans can intervene before those decisions get implemented and compare the model’s reasoning against its final output. Clear decision checkpoints, where AI outputs pass through rules-based checks before implementation, can also serve as a practical safeguard.

“Since LLMs can intensify human biases, the economic consequences of unsupervised deployment can exceed what we would expect from human decision-makers alone, so organisations need systematic monitoring for bias amplification,” he adds. “The goal is not to eliminate AI from the decision process but to design systems that leverage AI’s computational strengths while constraining its vulnerabilities.”

The post Does smarter AI generate more human errors? first appeared on China Business Knowledge.

Featured faculty: Jesse Yao
Written by Putro Harnowo

For the first time, Meta will eclipse Google as the largest advertising platform on earth. The social media company is poised to claim more than US$243 billion in revenue this year, edging out Google’s US$240 billion. Artificial intelligence (AI) has optimised Meta’s algorithms to better match ads to its 3.56 billion daily users while engaging them with short videos across Instagram and Facebook.

Both giants have invested heavily in AI, yet neither can sit on its laurels since other big techs have also ramped up their AI-driven algorithms to refine their social media ad targeting. Advertisers pay digital platforms to ensure their message reaches the right users and receive a payoff if those users make a purchase. Without being targeted with an ad, a potential consumer would not be aware of the product and may never buy it.

Ad targeting creates value for businesses and consumers when advertisers accurately reach their target audience. AI technologies like machine learning and large language models come in handy to help platforms collect information about users’ behaviours and demographics, such as age and location, and crunch the data to predict their likelihood of purchase.

At this point, it may be scary to see how social media uses AI-fueled algorithms to match their user profiles with the diverse needs of advertisers. However, Jesse Yao, an Associate Professor at the Department of Marketing at the Chinese University of Hong Kong (CUHK) Business School, argues that perfect targeting is impossible.

“When competition is strong, companies have a high chance of targeting the same pool of individuals, especially if their algorithms use similar mechanisms,” he says. “Companies will definitely develop more sophisticated algorithms to improve their targeting ability, but data privacy regulations will continue to limit their ability to perfect their targeting.”

Data privacy laws restrict businesses from collecting personal data that could identify an individual, so businesses must implement measures to decouple users from their real identities. Consequently, algorithms cannot achieve 100 per cent accurate targeting or certainty that someone is interested, and even if they did, the targets would no longer be high-quality.

How do businesses deal with flawed targeting

Since perfect targeting is arduous, advertisers often ponder whether to focus on “precision” to carefully target only a small number of very interested groups or on “recall” to cast a wider net to reach almost everyone who might be interested. The drawback is that prioritising high precision may miss out on other individuals who are also interested but were not identified, while prioritising high recall may waste resources.

A paper titled Algorithmic targeting and the precision-recall tradeoff, tries to navigate this dilemma. In the study, Professor Yao collaborates with Ganesh Iyer at the University of California, Berkeley and Zachary Zhong Zemin at the University of Toronto to use game theory, a mathematical study of strategic decision-making in which the outcome for each party depends on the choices of all involved.

“When competition is strong, companies have a high chance of targeting the same pool of individuals, especially if their algorithms use similar mechanisms.”

Professor Jesse Yao

The study highlights that the algorithms used by many platforms may be highly similar. While the algorithms used by the platforms are proprietary, their underlying machine learning techniques are similar, especially when they use public data and the same data analytics tools or AI models. People also typically search for products on multiple channels, signalling their interest in different platforms.

When algorithms identify individuals with a high probability of purchase, there is a big chance that other competitors are also eyeing the same targets. Even across distinct apps like TikTok, Facebook, and Amazon, there would still be significant overlap, and advertisers end up targeting similar groups with their competitors. This is often why, when you Google a brand, you might see an ad for the product you searched for on social media, and then see more ads from other brands.

“Companies can benefit a lot from being the only one that targets interested people, but will benefit less from competing for the same targets,” says Professor Yao. “To soften competition, they strategically target fewer people who are moderately interested, or lower both recall and precision. This way, the targets still have a reasonable chance of making a purchase when seeing ads, but without as much costly head-to-head competition.”

What if businesses tailor their algorithms differently?

Undoubtedly, targeting fewer people with moderate interest is not ideal, as it could also result in a pool of duds. Hence, advertisers are motivated to develop unique algorithms and analytics tools to differentiate their predictions. “The more unique, proprietary data a company uses, the less likely they are to target the same people as their competitors,” Professor Yao adds.

Advertisers nowadays can either work with digital platforms or build their own custom algorithms, but this approach is costly. It may also indirectly help their competitors by making the markets less crowded with ads, without the competitors even needing to spend a penny, so this strategy must be executed carefully.

Some may consider combining public and proprietary data to improve predictions and reduce costs. However, data privacy regulations in many jurisdictions require companies to choose either public or proprietary data for a given consumer, but not both at once.

Combining data sources also creates new personal data profiles, which require new consent or a re-evaluation under most data privacy regulations. Some digital platforms even explicitly forbid data scraping or combining user profiles for commercial targeting, especially if the data becomes personally identifiable.

Modern targeting in ever-competitive markets

Given that there is no easy way to avoid targeting overlap, Professor Yao suggests that advertisers strategically adjust their precision and recall while keeping tabs on the rivals’ digital campaigns. “Companies should not only think about the potential customers but also consider the strategic response from their competitors.”

When ad costs are low, they may consider targeting a larger pool of people to reach as wide an audience as possible or to achieve high recall. When ad costs are high, they should be more selective and target only those genuinely interested. To minimise overlap, advertisers should continue seeking unique segments they can own, rather than competing for the most obvious targets all the time.

For Meta, perhaps its unique segments are its short videos and cross-platform ecosystem, but for advertisers, its vast user base may increase overlap with their competitors. While the study does not specifically analyse Meta’s success, its framework provides a useful lens on how developing proprietary data analytics and strategic differentiation can gain a competitive edge.

There are plenty of fish in the sea, but if all the boats have trawlers, a wise fisher is those who know where to cast their line.

The post If everyone uses AI, who stands out? first appeared on China Business Knowledge.

Featured faculty: Jiang Wenxi and Gao Zhenyu
Written by Putro Harnowo

The launch of ChatGPT in late 2022 by OpenAI was a turning point for artificial intelligence (AI) worldwide. While the chatbot was not officially available in China, enterprise partnerships with Microsoft Azure and workarounds such as virtual private networks and third-party proxies have helped, to some extent, the ChatGPT moment reach the country.

However, that access closed when OpenAI decided to withdraw completely from the country in 2024. Another AI giant, Anthropic, released Claude in 2023 but never made it available to the Chinese market. Although it seems that American tech firms are pulling away, Chinese firms are equally hesitant to rely on foreign AI technology.

No wonder that DeepSeek’s debut in early 2025 sent another shockwave, the DeepSeek moment, and dramatically jump-started China’s AI adoption. “Technology adoption is not purely economic, but one deeply intertwined with strategic priorities,” says Jiang Wenxi, Professor of Finance at the Chinese University of Hong Kong (CUHK) Business School.

In a paper titled AI sovereignty, Professors Jiang and Gao Zhenyu, Associate Professor in the same department, in collaboration with Fudan University’s Ren Haohan, Wang Kemin, and Wu Yuezhi, examine more than 28,000 detailed dialogues between Chinese listed firms and their investors from January 2022 to June 2025 on investor interaction platforms. Investors frequently inquire whether firms use or plan to use specific AI models.

As expected, ChatGPT and DeepSeek were discussed far more frequently than other models. The launch of both also triggered notable adoption spikes, and a consistent pattern appears: Many Chinese firms were hesitant to use AI at first, but once DeepSeek became available, they accelerated.

From an economic point of view, DeepSeek is significantly cheaper than ChatGPT, so this could be one of the main reasons for avoiding foreign AI models, but the study notes that cost has never been a problem. Rather, there is a bigger concern.

Data is the lifeblood of AI

Firms that are consistently reluctant to adopt foreign AI but show a greater preference for domestic AI often use data-related terms and keywords in their annual reports. This suggests that the data is highly valuable to them and any leaks or breaches will result in dire consequences.

AI tools often require sending data to external servers, which raises the risk of sensitive data leakage. Since Chinese firms handling critical data are subject to regulations on data handling and processing, sending data to a foreign AI’s servers could pose a compliance risk.

“Priority of safeguarding data security makes firms wary of foreign AI models that could inadvertently funnel proprietary or sensitive data out of China,” says Professor Jiang.

Data is a fundamental resource for training AI models. Without vast amounts of data, it would be impossible for AI models to perform analysis and generate outputs and predictions. Businesses also increasingly see data as a critical asset, much like oil or other resources, that can give them a technological edge.

Therefore, China has issued a series of laws, including the Cybersecurity Law (2017), the Data Security Law (2021), and the Personal Information Protection Law (2021), that provide a comprehensive framework for handling data. These laws require data to be stored on servers physically located within the country.

Cybersecurity Law laid the foundation for data protection and set broad rules for network security and critical information infrastructure. Data Security Law imposes strict rules on data collection, storage, use, and cross-border transfer. The Personal Information Protection Law dictates how personal data must be handled, including requirements for transferring outside of China.

Similar to many other countries, China’s data protection laws borrow heavily from the EU’s General Data Protection Regulation. While the EU does not impose a blanket requirement to keep all data locally, it sets very strict rules on how its citizens’ data must be protected, regardless of where the company processing the data is located.

“Technology adoption is not purely economic, but one deeply intertwined with strategic priorities.”

Professor Jiang Wenxi

More than just supporting local technology

China produces more top AI researchers and talent, but the US leads in AI models and the critical chips for AI training. The discussions captured on investor interaction platforms also acknowledge this. Investors understand that a preference for domestic AI models may place Chinese firms at a competitive disadvantage.

Regardless, Chinese firms are keen to use domestic AI models to avoid compliance risks. “Although the best Chinese AI models still underperform the US ones, the gap is quite small and may not be significant when applied to specific business scenarios,” Professor Jiang adds.

Further analysis indicates that Chinese firms do not completely ignore foreign models but strategically limit their use to low-risk domains, such as customer service, translation, and other non-critical functions.

Moreover, the market supports this approach. When Chinese firms, especially those in strategic national sectors, announced their AI adoption, their stock prices rose significantly. Conversely, when these firms said they were using foreign AI, their stock performed worse or at least stayed unchanged.

The future of AI ecosystems

AI has become a foundational utility, much like the internet, and has driven exponential investment in China and the US, the dominant players in AI development. Alas, the two superpowers have locked in a head-to-head competition for years.

Most recently, the US has banned the sale of its advanced AI processors to Chinese firms and their subsidiaries, while China intervened in the acquisition of a Chinese-founded AI firm, Manus, to prevent the outflow of proprietary technology and AI talent to the US.

The available data and privacy frameworks determine how AI models are trained, which then shapes the next wave of AI innovations. Since data security influences AI adoption and development, this loop may lead to AI models from two dominant powers evolving differently.

“The development of AI technology may evolve into two separate approaches and ecosystems,” Professor Jiang says. “However, Chinese models are open source, so firms can fine-tune them locally to meet specific needs with greater data control for developers to adopt. This gives some hope that the separation might not be exacerbated.”

Ultimately, with the growing calls for global AI governance to safely advance the technology, collaborative frameworks may emerge to mitigate the risks of a fragmented ecosystem and ensure that the benefits of AI technologies are shared universally.

The post When data security drives AI preference first appeared on China Business Knowledge.

Featured faculty: Wu Fan
Written by Joanne Madrid

Retail investors traditionally rely on financial news, analyst reports, and online forums to interpret the market. With generative artificial intelligence (GenAI) rewriting the industry playbook, investors are rapidly deploying it into their decision-making. Deloitte expects the share of individuals using GenAI for investment advice to reach 78 per cent by 2028.

“GenAI significantly lowers retail investors’ mental workload by integrating complex data and offering personalised information aggregation much faster than traditional methods,” says Wu Fan, Assistant Professor of Accounting at the Chinese University of Hong Kong (CUHK) Business School. “Yet little is known about the dynamics of retail investors’ interactions with GenAI.”

Having partnered with one of China’s largest GenAI platforms to analyse more than 1.7 million stock-related queries, Professor Wu finds that users typically start with simple questions, such as asking for stock recommendations or market conditions.

When GenAI answers convey a positive signal, and users perceive it as useful, the queried stock tends to yield higher returns on the same day. This effect is stronger when users react to those answers, such as giving a thumbs-up or sharing them with their networks.

However, Professor Wu warns, this doesn’t mean GenAI accurately predicts or even causes market movement. “GenAI aggregates and echoes existing market sentiment rather than exerting an independent influence on trading behaviour. When users trust GenAI’s answers, they might also act on them, and this collective interest could push the stock price up.”

Highly active stock-related queries often signal a few things: a large number of shares are being traded, some informed traders have better information about certain stocks than others, and the gap between the highest stock price a buyer is willing to pay and the lowest price a seller is willing to accept is widening.

Most common queries from retail investors

In a paper titled, How stock market participants use generative artificial intelligence: Evidence from user-platform interaction data, Professor Wu, Frank Ecker of the Frankfurt School of Finance and Management, Li Xitong of HEC Paris, and Li Yilan of ESSEC Business School investigate how Chinese retail investors begin their GenAI journeys. The data shows 40 per cent of users ask just one stock-related question, but only 8.5 per cent follow up with more than 20 queries.

For retail investors who ask more questions, their requests gradually move from general to more specific queries, such as financial statement analysis, assessing the impact of news events, and comparing competitors. This pattern is the most common among financially knowledgeable users, who tend to use GenAI for deeper analysis.

“Retail investors’ journey evolves from passive information consumption to a more targeted information extraction,” Professor Wu adds. “Investors initially use AI to provide clear basic information and then transition to deep-dive analytical needs.”

GenAI delivers the strongest results when asked to compare, organise, and interpret specific financial information. “Tasks involving structured reasoning and synthesis, such as financial statement analysis, seem to yield the most informational benefits,” he adds.

“GenAI aggregates and echoes existing market sentiment rather than exerting an independent influence on trading behaviour.”

Professor Wu Fan

Do retail investors ask the right questions?

The study also spots a missed opportunity. Retail investors rarely use GenAI to summarise company filings and disclosures, one task where AI excels and could add value.

“Retail investors may not fully understand what GenAI can do, or they simply experiment with the technology out of curiosity without deeply exploring its capabilities, so they still struggle to formulate effective prompts to get the summary they want,” Professor Wu says. “Some may also worry about the accuracy of GenAI summaries and prefer to rely on traditional sources.”

Retail investors often prefer information that is already summarised by intermediaries rather than having GenAI create it from scratch. As more sophisticated investors tend to focus on analytical queries, summarisation may also fall into basic tasks they quickly move past, or not be seen as the most efficient way to obtain the insights they seek.

What triggers GenAI queries

Retail investors typically don’t go straight to GenAI for analysis, but often get their first inspiration from reading or watching the news. User queries rise around major corporate events, such as earnings announcements and performance forecasts, and particularly surge only after such events make headlines.

“GenAI does not completely replace the information funnel,” Professor Wu says. “Platform query volumes still closely track contemporaneous media coverage, suggesting that users often still rely on traditional channels to initiate their research.”

GenAI queries peak during trading hours, as users are likely looking for quick checks of time-sensitive information. After hours, a significant number of queries are still submitted, but they are more likely to be for in-depth research.

User queries also tend to decrease when companies publish reports covering longer or broader topics. More detailed disclosures and performance forecasts are associated with fewer GenAI queries, suggesting that when companies provide investors with enough context upfront, there is less need to seek it elsewhere.

Users will also stay engaged when earlier queries about market signals align with actual stock performance, indicating that perceived accuracy builds trust and repeat use. Surprisingly, users react negatively or show lower interaction to long or complex answers, but respond positively to concise, opinion-driven, and direct responses.

“If answers from GenAI simply mimic the density of traditional analyst reports, retail investors may disengage,” Professor Wu adds.

Preparing for AI-assisted investing

As GenAI adoption grows, Professor Wu suggests that companies rethink how they communicate with investors and make disclosures easier to process by both investors and AI tools.

“Firms must comply with regulatory requirements to provide sufficient, relevant and timely information to the market, but they can also consider structuring disclosures in more machine-readable formats to better facilitate AI-assisted processing,” he says.

Given that users want quick answers and GenAI platforms are good at extracting summaries, corporate disclosures should be clear, concise, and use a consistent style to make it easier for AI tools to identify and extract relevant information. The report should also highlight key takeaways and, instead of just presenting numbers, include an explanation of what they mean and why they changed, so GenAI can analyse the context to provide more helpful explanations.

In terms of content, the report should directly address common user questions about the business outlook, financial performance, and operations, to help GenAI provide better information and potentially reduce the need for users to ask follow-up questions.

“Keep in mind that, when using GenAI, users must remain cautious regarding hallucinations or factual errors in AI-generated responses.” Ultimately, GenAI can be a powerful research tool, but it works best when paired with critical thinking, not as a replacement for it.

The post How retail investors use GenAI to navigate stock markets first appeared on China Business Knowledge.

Featured faculty: Li Hongfei
Written by Pan Jingyi

Childhood friends Queenie Fan and Day Lau started their handbag brand, Cafuné, a decade ago in Hong Kong. The label has grown its retail footprint across Asia, yet success did not shield Fan from the kind of scepticism many women founders know well.

“When you start your own business at a young age, people are not always willing to take you seriously, especially in the leather industry, which is typically more traditional and male-driven,” Fan said in an interview.

Her story is far from unique. Katherina‑Olivia Lacey, a co-founder of a Singapore‑based tech startup Quincus, had investors question her role during a seed funding round. A 2025 report by the Organisation for Economic Co-operation and Development also found women are 25 per cent less likely than men to receive bank loans to fund their businesses.

“Different ways of asking questions change the way AI defines successful entrepreneurs.”

Professor Li Hongfei

The myth of the hard-charging leaders who win through aggression and swagger, typical masculine traits, is still alive and prevalent in modern culture. An artificial intelligence (AI) model trained on a myriad of stories and human writing to converse and answer questions, called a large language model, may learn and accept that skewed worldview.

However, Li Hongfei, Assistant Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School, discovers that AI can be wiser than humans.

“Our study found that large language models generally exhibited a balanced gender perception of entrepreneurship, with a slight preference for feminine traits. However, algorithms may show a notable bias towards masculinity in specific scenarios, or when you use a different approach to ask the questions.”

AI advocates gender equity

In a study titled Detecting gender stereotype biases against women entrepreneurs in large language models, Professor Li, along with Cao Xian of Illinois State University, Xu Qingyu of the City University of Hong Kong, and Zhu Ruoqing of the University of Illinois Urbana-Champaign, asked ChatGPT, a major language learning model, to describe characters of successful entrepreneurs from selected words: dominance, forcefulness, aggressiveness, assertiveness, warmth, sensitiveness, emotiveness, and expressiveness.

The results are encouraging. Rather than choosing purely masculine traits, such as dominance and aggressiveness, the model consistently selects three words for both successful men and women entrepreneurs, namely assertive, expressive, and warm. While assertiveness is traditionally seen as masculine, the other two are closer to feminine traits.

“This contrasts with historical and popular literature that overwhelmingly portrays successful entrepreneurs with stereotypically masculine traits,” Professor Li says. “This pattern might reflect programmed efforts by ChatGPT to minimise unethical gender bias in descriptions of successful entrepreneurs.”

AI companies have used several methods to prevent their systems from repeating sexist ideas, such as filtering out biased content and adding more examples of women in leadership roles. OpenAI, for instance, runs fairness tests to see how ChatGPT responds to a question when the user’s gender or name changes, ensuring its neutrality.

Women entrepreneurs win sometimes

In another experiment, Professor Li and the team fed ChatGPT with texts highlighting masculine traits, feminine traits, a mix of both, and neutral traits to subtly put the model in different mindsets. They then asked it to evaluate a business proposal from a gender-neutral founder to see whether exposure to certain traits would change how AI judges the proposal.

As in earlier results, AI shows little evidence of gender stereotyping. More specifically, ChatGPT does not undervalue feminine traits in general business evaluations and explicitly points out that the entrepreneur’s gender doesn’t affect the outcomes.

To make the test more realistic, the team asked AI to evaluate actual entrepreneurial ideas from Reddit, and it still doesn’t exhibit gender biases. In fact, the algorithms even give higher marks to ideas written in a warmer, more collaborative and people‑focused style, as in a more feminine tone.

When bias finally appears

The picture shifts when ChatGPT was asked to behave less like a conversation partner and more like a venture capitalist analysing investment opportunities.

In the last experiments, ChatGPT was assigned to weigh different personality traits and detailed business ideas, then determine an investment outcome. This test is closer to how AI is used in real life for screening applications, scoring founders, and ranking business opportunities using financial calculations.

Here, the result changes. ChatGPT favours more assertive, risk-embracing, and affirmative pitches that reflect masculine character. In other words, once the AI is pushed to think more technically, masculine traits carry more weight.

“Different ways of asking questions change the way AI defines successful entrepreneurs,” Professor Li says. “When being asked to behave like a data analyst or investor to calculate an investment score using a formula rather than simply ranking an idea, AI doesn’t ‘realise’ it is dealing with a gender question and treats the task as a mathematical exercise.”

Human involvement is crucial

As AI becomes a bigger part of entrepreneurial life — helping with everything from business plans to funding decisions — understanding its strengths and limitations is key. Large language models take the women’s side when describing success or rating ideas in a straightforward way, but when asked to behave like an investor analysing with formulas, AI quietly tilts towards a stereotypical pattern.

“We still need more research, using different methods, to really understand when large language models copy unfair biases and when they help challenge them,” says Professor Li.

For women founders like Fan, Lacey, and others building their ventures, AI can be a useful sounding board that does not automatically downgrade them for being women or for using a more collaborative and emotional tone. However, the study also reminds us that AI outputs should not be treated as completely impartial.

RELATED ARTICLE

Would you trust AI to decide your pay raise?

“When using AI to screen applications or analyse pitches, adding human oversight is still important to avoid unintentional prejudice,” Professor Li adds. “When you ask AI to help you make a decision, you need to understand what the rationale behind its decision is.”

Ultimately, given that gender bias is historically prejudiced human-generated data on which AI is trained, the discerning judgment of humans remains paramount. This vital insight reinforces the indispensable role of humans in ensuring that technology elevates fairness rather than perpetuates inequities.

The post Does AI perpetuate the boy’s club in startups? first appeared on China Business Knowledge.

Featured faculty: Robert Li Xiang
Written by Ellis Ng

“Travel changes you. As you move through this life and this world, you change things slightly, you leave marks behind, however small. And in return, life—and travel—leaves marks on you,” said the late American documentarian, Anthony Bourdain. Travelling is supposed to be an exercise for the soul.

However, tourist misbehaviour has constantly dotted headlines. Last year, anti-tourism protests erupted in Spain’s Canary Islands over tourists’ bad behaviour and overtourism, and similar strikes have since spread to Barcelona, Málaga, Athens, and other European cities. In Japan, the same problem has driven the city of Fujiyoshida to cancel its annual cherry blossom festival.

Most misconducts, like being noisy, littering, queue-jumping, and disregarding local customs, might seem minor, but they add up for locals. Traditional countermeasures, such as signage or staff patrolling to issue reminders, often fall short. Visitors easily overlook signs, and confrontations can flare up between tourists and staff.

As tourism boards and site managers search for better ways to manage visitors, Robert Li Xiang, the Fung King Hey Memorial Professor of Tourism Management and Director of the School of Hotel and Tourism Management at the Chinese University of Hong Kong (CUHK) Business School, suggests that robots can be an effective tool for this.

Equipped with artificial intelligence (AI), Professor Li notes that robots can act as persuasion agents to encourage positive behaviour. Far from replacing human interaction, they provide a non-confrontational means to foster responsible conduct and enhance the experience for both tourists and locals.

“Intelligent robots can influence behaviour through a few simple but powerful mechanisms. When a robot approaches a visitor or speaks to them, it creates a sense that ‘someone is watching,’ which can make people more aware of their actions,” he says.

Smart robots to encourage smart behaviour

A study titled Robot guardians: Mitigating tourists’ deviant behavior with intelligent robots documents the findings of Professor Li, as well as Zhang Mengyang, Pang Shuo, and Liu Na at Southwest Jiaotong University, and Shi Si at the Southwestern University of Finance and Economics in Chengdu, that robots outperform signage and humans when it comes to persuading tourists to behave.

In a field experiment at a busy intersection in southwestern China, among 8,148 observed tourists over three consecutive weekends, 17 per cent jaywalked when human staff were on patrol and 25.1 per cent when only signage was placed. This number dropped to 12.5 per cent when a robot was present.

“Robots feel less personal than human staff, so people may be more open to listening without feeling judged or embarrassed.”

Professor Robert Li Xiang

In other scenario-based experiments and AI-generated simulations with participants in China and the US, the results also show that intelligent robots are more influential in affecting tourist behaviour. The reason is that robots created a keen sense of being watched without the psychological tension that human intervention often triggers.

Professor Li grounds the findings in impression management theory, which holds that people control their behaviour when they feel they are being seen. Unlike surveillance cameras, robots can provide on-site intervention and activate self-awareness, but not as personally intimidating as a human figure, reducing the defensiveness that can undermine compliance.

“Robots feel less personal than human staff, so people may be more open to listening without feeling judged or embarrassed. This fusion, being noticeable but not confrontational, helps nudge people towards better behaviour,” Professor Li says.

Different looks for different messages

How a robot looks and speaks also needs to match. When a human-looking robot delivers a stern warning, it can feel socially threatening, and people get defensive. But if that same robot uses a touch of humour, the interaction feels lighter, and people are more willing to conform.

The underlying reason comes down to social threat and embarrassment. A basic-looking robot, on the other hand, doesn’t trigger much social pressure on its own, so a direct message works best. Humour from something that clearly isn’t human can make the message feel unserious.

Therefore, if its primary function is to intervene in deviant behaviour, a less human-looking robot with a more serious tone would be effective. “A simpler-looking robot may be more effective with a direct message, but a human-like robot can get away with using humour. Getting this combination right makes a big difference in how people respond,” Professor Li says.

“The key is finding the right balance. Robots need to be noticeable enough to catch attention and signal that behaviour matters, but not so intimidating that people feel uncomfortable or defensive.”

The future of smart and sustainable tourism

Smart robots offer a consistent, safer way to address daily misconduct, and the study opens new possibilities for smart tourism as authorities grapple with growing challenges in visitor management.

“For policymakers, robots provide a softer approach to public management, encouraging better behaviour through gentle reminders rather than strict enforcement,” Professor Li adds. “For businesses, robots can ease the burden on frontline staff, who often have to deal with difficult or confrontational situations.”

This approach could work in other environments too, including transport hubs, hospitals, shopping malls, campuses, and residential communities. “We’re already seeing early examples, for instance, robots helping with traffic guidance or reminding people of public rules in some Chinese cities.”

Looking ahead, Professor Li sees the most promising approach is likely a combination of robots and humans. Robots can handle routine reminders in a consistent, non-confrontational way, while human staff can step in for more complex or sensitive situations.

Still, organisations will need to address practical challenges before deploying robots in public. “On the practical side, organisations need to think about costs, maintenance, and whether the technology is reliable enough for daily use,” he says. “It’s also important to consider how robots will work alongside human staff.”

Privacy is another concern, especially if robots collect or process data. “Organisations should be transparent about how the robots are used and avoid making public spaces feel overly watched. Most importantly, robots should be used to give gentle reminders, not to shame or punish people.”

The post Could intelligent robots be the cure for bad tourists? first appeared on China Business Knowledge.

Featured faculty: Prasanna Karhade
Written by Putro Harnowo

Earlier this year, an X post from a Meta engineer went viral after her artificial intelligence (AI) bot ran amok, deleted emails, and wouldn’t stop, even after being ordered to do so. The engineer was testing the agentic AI that can execute tasks on its own. This AI agent promises autonomous decision-making to solve complex problems without human supervision, but the accident may prove it’s not entirely safe.

While businesses ponder the pros and cons of agentic AI, a safer choice called intelligent process automation (IPA) has been around for years. IPA bots combine AI technologies with robotic process automation to create AI bots. Their strength lies in low-code/no-code (LCNC) toolkits that enable users without coding skills to automate specific tasks, making them more secure and reliable in corporate environments.

Microsoft Power Automate is an example. It can perform repetitive tasks like sending emails or generating reports seamlessly within the Microsoft ecosystem. Software maker SAP also offers Build Process Automation to automate tasks integrated with other SAP products, and IBM’s Cloud Pak for Business Automation caters to larger enterprises seeking cloud-based automation. Many other tech firms provide independent IPA bots for specific needs.

However, despite hefty investments in AI solutions, many companies still struggle to deploy IPA bots widely across their workplaces. Prasanna Karhade, Associate Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School, argues that companies are still in the dark about how to implement AI solutions effectively.

“AI technologies are rapidly growing, but companies are sometimes stuck with the traditional ways of governing technology solely in the hands of the IT team,” he says. “Employees without coding skills can now leverage IPA bots to create and refine their own automated workflows, bolstering the democratisation of technology development. However, this innovation must also align with company IT policies.”

Professor Karhade notes that this new dynamic has forced businesses to revisit whether centralised technology management remains relevant and, if not, what the best strategies are to roll out AI tools that can be widely accepted across the organisation.

When AI tools are failing, and why

In a study titled, AI governance and the decentralisation of technology production: An investigation of AI-based IPA bots, Professor Karhade and his co-authors examine 176 IPA projects at a Fortune 200 US multinational IT services firm, particularly in the banking, financial services, and insurance domains, to identify the critical factors in AI tool adoption.

“Companies naturally want all employees to use the AI tools, so they demand high utilisation. Apart from that, the AI solutions must be repeatable to ensure broader application within the company beyond the initial use,” says Professor Karhade. “Therefore, utilisation and repeatability are the two key factors in AI governance.”

Having investigated 24 highly underutilised and 54 highly unrepeatable IPA projects, Professor Karhade and the team find that more than 83 per cent of these failed projects are imposed by top management without employee input. Conversely, among the 46 and 35 IPA projects classified as highly utilised and repeatable, respectively, employees are responsible for over 91 per cent of them. These successful projects have low coding intensity, meaning employees apply their own knowledge to deploy AI bots using the LCNC toolkits.

A bottom-up approach turns out to significantly improve the utilisation and repeatability of IPA projects, while a top-down approach results in the opposite. This finding provides a robust starting point for computational experiments to further identify other influential factors in determining the success of AI solutions.

“AI users will be more empowered as they can do many things on their own, but a centralised IT team still plays a crucial role in enabling and overseeing AI infrastructure.”

Professor Prasanna Karhade

The elements of success

The key to the success of an AI solution lies in how it starts, or what the researchers call the “genesis”. IPA projects initiated by employees yield successful adoption, but further analyses find that how the project is deployed, the amount of coding skills required, and the complexity of the tools also contribute significantly.

IPA bots are highly used when deployed by users using the LCNC toolkits, so they require very little coding. Process intricacy, or how many steps an IPA bot has to do to complete a task, is also crucial. Intricate processes are harder to automate, but when users leverage their knowledge to create bots with the LCNC toolkits, the bots are highly likely to be used.

Deployment, or how AI tools work and interact with users, can be divided into three types: attended, where users trigger the process; unattended or fully automated; and hybrid, where AI works alone but sometimes needs human help. Users are still likely to use attended bots in a top-down manner, but they’re not widely accepted.

AI bots may solve the problem at hand, but they are not necessarily repurposed widely. This is where repeatability matters. To create highly accepted bots for a wider user base, unattended AI bots are particularly preferable as they can handle processes autonomously.

Although an AI solution is all about democratising technology, when users implement an AI bot primarily for their own specific needs, the bot becomes too specialised to be reused by others. This contradicts the idea of decentralised technology, or, as Professor Karhade calls it, the limits of democratisation. When this happens, IT support is needed to ensure that the AI tools are reliable and flexible enough to be used across the company.

The evolving roles of the IT team

AI may have democratised technology adoption, but Professor Karhade underlines that the IT team is irreplaceable. “AI users will be more empowered as they can do many things on their own, but a centralised IT team still plays a crucial role in enabling and overseeing AI infrastructure.”

For companies, management should remain receptive to user-initiated AI projects and provide guardrails to enable seamless collaboration between employees and IT teams. This balanced approach will enable businesses to harness the benefits of decentralised technology while safeguarding operational integrity. Otherwise, an accident similar to what happened with Meta’s engineer could happen.

Professor Karhade believes the findings apply to broader contexts and industries dealing with large volumes of documents, such as retail, logistics, healthcare, and human resources. The core ideas revolve around how companies should rethink their technology management in the AI era and build a sustainable ecosystem that enables diverse employees to contribute to AI solutions.

The post Should you let employees build their own AI bot? first appeared on China Business Knowledge.