How TikTok used AI to overcome scale-scope trade-off

For decades, the trade-off between scale and scope has posed a major strategic challenge to businesses and academic researchers alike. An industrial firm that makes one product, be it electrical devices or automobiles, often faces physical capacity constraints or limited markets for its core product – so these firms expand into related and even unrelated sectors, deploying surplus resources, in order to grow through scope as well as scale.

What’s known as the resource-based view (RBV) of such a trade-off says that a firm’s internal resources and capabilities are the key drivers to competitive advantage. This view was championed by Edith Penrose in her much-cited 1959 book, The Theory of the Growth of the Firm, and by many scholars since.

The digital era allowed rapid scaling up through network effects, but often required hyperspecialisation – which prompted its own sort of tension between such specialisation to increase scale and diversification to achieve greater scope. What’s more, even successful digital firms can run up against constraints posed by non-digital resources: Uber’s expansion into food delivery, for example, stretched the key (non-digital) resources of drivers and vehicles, posing significant opportunity costs and diluting the customer experience for passengers.

How ByteDance’s AI strategy challenges the resource-based view

New research co-authored by Shahzad (Shaz) Ansari, Professor of Strategy and Innovation at Cambridge Judge Business School, sheds new light on this scale-scope dilemma through an examination of ByteDance, parent company of hugely popular short-form video platform TikTok, and how the firm used artificial intelligence to challenge traditional thinking in this area.

“Our research shows that the cross-fertilisation at ByteDance through AI turns conventional RBV logic on its head,” says Shaz. “Diversification at ByteDance doesn’t only draw on AI but makes it stronger, thus making diversification a source of resource advantage rather than an outcome of such resource advantage as traditionally explained.”

The research co-authored by Shaz, published in Strategic Management Journal (SMJ), follows several other recently published studies on AI implications that Shaz has been involved with. Research in the journal Long Range Planning outlined a framework designed to bridge differences of opinion on how AI can best be deployed in the circular economy sphere, based on purpose, strategy and governance. An article in Harvard Business Review focused on the automotive sector found that AI has a clear advantage over humans when it comes to data analysis and predictive modelling in crucial areas such as product design and optimising supply chains. Research in the International Journal of Business and Management called for AI partnerships between smaller firms and tech giants like Microsoft and Google to have guidelines on ethics, power imbalances and data control.

Shaz has been named Associate Editor at the SMJ, where he plans to focus on qualitative research as the only Associate Editor so specialised in that area. “There’s little doubt that further research I will contribute to SMJ as an editor or author, beyond the ByteDance study, will further examine the implications of AI on all aspects of management,” he says.

How ByteDance used AI to learn across video, news and e-commerce

While many companies now use AI to boost efficiency or improve product customisation, ByteDance went a step further: AI helped TikTok create short-form video and text content that refreshes very quickly. AI also allowed the company to translate learning from customers across domains, so models trained on one domain such as videos informed other domains such as news feeds and e-commerce – thus improving the underlying models and creating what Shaz and his co-authors term a “’reinforcing cycle of growth’”.

“Prior (research) highlights that resources which can be (re)deployed at low marginal cost are fungible and scale-free and how knowledge as resource grows through diverse application and recombination,” says the research co-authored by Shaz. “Yet even such resources do not inherently improve with redeployment. We show that AI’s self-learning property allows it to strengthen through use: learning in one domain enhances its predictive performance in others, provided the data are accessible, structurally related and supported by organisational integration.”

The research is authored by Feng Wan of International Business School in Zhejiang, China, Tianxi Yang of University of International Business and Economics in Beijing, Xianwei Shi of Shanghai Jiao Tong University, Ke Rong of Tsinghua University in Beijing, and Shaz Ansari of Cambridge Judge Business School.

News and information were the focus of ByteDance’s early days

While consumers in the UK and other countries largely know of ByteDance through TikTok, the research by Shaz and colleagues outlines the firm’s more varied history. When founded in 2012, ByteDance principally offered the Chinese news and information platform Toutiao, and then developed AI recommendation algorithms to match users with other content such as short videos, social networking and office tools as it expanded beyond consumers into workplace offerings.

It was initially not all smooth sailing, as ByteDance faced issues such as inaccurate recommendations and the failure of short videos to start quickly. “In response, it continually reconfigured, centralising AI to enable its reuse while decentralising product experimentation to domain units. Through this adaptive interplay, the scale-scope trade-off was transformed into a reinforcing cycle of growth.”