![Examples of scaling relationships for AI (red), platform (green), and service startups (blue): (a) the estimated amount of revenue vs. the number of employees scales sublinearly and (b) the total amount of funding vs. the number of employees scales superlinearly strengths are so vastly different for the startup groups, we believe it is an insightful variable in terms of group comparison. We included two further analyses by (i) splitting up the data geographically into Europe, Northern America and Asian-Pacific and (ii) examining different business categories of AI startups and included the results in Appendix 2. The regional analysis regarding revenue (i) reveals close to no differences for service startups but larger differences for platform and AI startups: The beta coefficient is smallest for Northern America (AI: 0.23; platform: 0.18) and highest for Asian-Pacific (AI: 0.42; platform: 0.44) with Europe in between (AI: 0.29; platform: 0.25). Since the average value of the target variable is close to similar for the three categories this hints to connection of employees and revenue that is much smaller in the US. For funding, the regional differences are much smaller. The business category analysis for AI startups (ii) reveals that in the FinTech and the Health Care sector employees and revenue are connected most closely. Since these sectors are heavily regulated, market entry and data sharing prove to be especially difficult [9, 57]. Regarding the funding, the more traditional sectors of Analytics and E-Commerce are less likely to gain more funding when having more employees. We ran several robustness checks to rule out other explanations. First, we cutoff revenue outliers at different top percentages to control for possible distortions by the most performant startups. For a 10% cutoff, the slope for every startup group dropped similarly by around 3% (AI startups) to 5% (service startups). Second, grouping the startups by age and running the same analyses shows that both the predictive quality and the slope β rise with a higher startup age, which is in line with the theory that bigger](https://www.researchgate.net/profile/Matthias-Schulte-Althoff/publication/344493810/figure/fig1/AS:943629447213056@1601990200306/Examples-of-scaling-relationships-for-AI-red-platform-green-and-service-startups_Q320.jpg)
Content uploaded by Matthias Schulte-Althoff
Author content
All content in this area was uploaded by Matthias Schulte-Althoff on Oct 06, 2020
Content may be subject to copyright.
A preview of the PDF is not available
A Scaling Perspective on AI Startups
Abstract and Figures
Digital startups’ use of AI technologies has significantly increased in recent years, bringing to the fore specific barriers to deployment, use, and extraction of business value from AI. Utilizing a quantitative framework regarding the themes of startup growth and scaling, we examine the scaling behavior of AI, platform, and service startups. We find evidence of a sublinear scaling ratio of revenue to age-discounted employment count. The results suggest that revenue-employee growth pattern of AI startups is close to that of service startups, and less so to that of platform startups. Furthermore, we find a superlinear growth pattern of acquired funding in relation to the employment size that is largest for AI startups, possibly suggesting hype tendencies around AI startups. We discuss implications in the light of new economies of scale and scope of AI startups related to decision-making and prediction.
Figures - uploaded by Matthias Schulte-Althoff
Author content
All figure content in this area was uploaded by Matthias Schulte-Althoff
Content may be subject to copyright.
... AI startups have ample opportunities for growth, provided they can scale their value offerings with resources at their availability. In our context, growth is defined as a relevant measure of firm size, and scaling is the relationship between multiple measures of size, such as financial, human, or digital resources (Cottineau, 2017;Schulte-Althoff, Fürstenau, and Lee, 2021). AI startups may benefit from the increased modularity, flexibility, and malleability of digital infrastructures, (Henfridsson, Mathiassen, and Svahn, 2014;Yoo et al., 2012), particularly design flexibility and design scalability, to facilitate rapid growth (Henfridsson, 2020;Huang et al., 2017). ...
... Despite progress in transfer learning and fine-tuning for broader applications, the sensitivity of AI algorithms to data input variations still constrains their eased replication, as data changes can significantly alter outcomes (e.g., Chen, Zaharia, and Zou, 2023;Kaddour et al., 2023;Sahiner et al., 2023). In general, AI startups appear to have a distinct growth pattern compared to other digital startups due to their reliance on AI-specific resources that are hard to scale (Burström et al., 2021;Chui and Malhotra, 2018;Linde et al., 2020;Paleyes, Urma, and Lawrence, 2022;Schulte-Althoff, Fürstenau, and Lee, 2021) and therefore constraining organizational scaling efforts (Giustiziero et al., 2021;Haefner et al., 2023;. So far, we still lack an empirically grounded understanding on how to orchestrate resource bundles within AI startups to benefit from design flexibility and design scalability, which is sensitive to AI specific challenges and also allows rapid reactions to changing circumstances. ...
... Generally, startups in a digital setting have two scaling modes: due to a flexible design ventures can react rapidly to changing contexts because their digital infrastructure is not pre-defined (Henfridsson, Mathiassen, and Svahn, 2014), and due to a scalable design startups may replicate offers at low costs due to the eased replication of digital artifacts (Henfridsson, 2020;Shapiro and Varian, 1998). Still, given the different set of resources which AI startups are built upon, the configuration of resources for scaling in the context of AI seems to differ and remains a difficult task (Haefner et al., 2023;Schulte-Althoff, Fürstenau, and Lee, 2021 ...
As AI startups increasingly enter various domains and markets, they promise high economies of scale and scope. However, the different technical foundation of AI startups compared to conventional digital startups challenges design flexibility and eased design replicability, e.g., due to unbalanced datasets, customized model development, resource-intensive procedures and constant monitoring. We adopt an action design research approach, drawing on several AI startup batches from an AI accelerator and the existing body of knowledge, to provide an empirically grounded answer on how AI startups can orchestrate their resource bundles to benefit from flexibility and replicability of designs. Our findings suggest an actionable process that helps to reflect on activities and resource bundles in an AI product pipeline, to orchestrate bundles for facilitated design flexibility and design replicability. By providing design principles tailored to the resource orchestration for scaling AI startups, we advance the discourse towards a nascent design theory.
... AI startups are digital startups, where AI is a central aspect of its business model (Schulte-Althoff et al. 2021). In the domain of AI startups, data is a critical asset, often described as the fuel for artificial intelligence technologies (Pumplun et al. 2019). ...
... Scaling, therefore, refers to how individual components of a system react when its size changes (West 2017). The ability to scale accordingly is a crucial point for AI startups, especially concerning fundraising, productivity, and the distribution of new product and technology innovations (Schulte-Althoff et al. 2021). In comparison to other types of startups, AI startups seem to be able to expand known types of scaling, such as by being capable of transferring machine learning models to other business applications (Agrawal et al. 2019). ...
This study examines data management and scaling in AI startups. It uncovers the distinctive scaling dynamics of AI startups, where data emerges as a central resource. Through qualitative research and interviews with 13 AI startups, operational complexities and strategic approaches in data management are unveiled. The research employs a practical Framework of Data Conceptualisation for AI startups, which we developed based on our findings, enriching both scientific understanding and entrepreneurial theory in the AI domain.
... While data is considered homogeneous, the creation and growth of ventures is more influenced by the specificity and lack of interoperability of tools (von Briel et al., 2018). However, data has become critical to the value creation of digital ventures (Abbasi et al., 2016), especially as more companies produce machine learning-based products and services that rely heavily on data (Iansiti & Lakhani, 2020;Schulte-Althoff et al., 2020). Empirical studies suggest that data has an impact on the creation and growth of ventures because it enables ventures to adapt more quickly to market changes and ultimately create superior customer value (Gregory et al., 2020;Huang et al., 2017). ...
... Conditions that inhibit such free combinations of digital resources, such as external regulation (Kimjeon & Davidsson, 2022), might therefore impair venture creation. Recently, increasing attention has been paid to the effect of combining digital data on venturing, for instance, its effect on venture growth (Huang et al., 2017;Schulte-Althoff et al., 2020). Digital innovation and digital entrepreneurship research have, however, predominantly focused on data in consumer-facing firms where most data that is being combined is transaction data owned by singular firms or platforms, e.g., considering financial transactions between providers and users (Huang et al., 2017) or investors and borrowers (Gomber et al., 2018), potential partners for online dating (Davidson & Vaast, 2010), ride sharers and users (Frey et al., 2019), or spectators and e-race drivers (Jarvenpaa & Standaert, 2018). ...
Data has become an indispensable input, throughput, and output for the healthcare industry. In recent years, omics technologies such as genomics and proteomics have generated vast amounts of new data at the cellular level including molecular, structural, and functional levels. Cellular data holds the potential to innovate therapeutics, vaccines, diagnostics, consumer products, or even ancestry services. However, data at the cellular level is generated with rapidly evolving omics technologies. These technologies use scientific knowledge from resource-rich environments. This raises the question of how new ventures can use cellular-level data from omics technologies to create new products and scale their business. We report on a series of interviews and a focus group discussion with entrepreneurs, investors, and data providers. By conceptualizing omics technologies as external enablers, we show how characteristics of cellular-level data negatively affect the combination mechanisms that drive venture creation and growth. We illustrate how data characteristics set boundary conditions for innovation and entrepreneurship and highlight how ventures seek to mitigate their impact.
Supplementary Information
The online version contains supplementary material available at 10.1007/s12525-023-00669-w.
... The CE certification process alone can take up to a year or more, challenging business plans and go-tomarket times (MedDev Compliance Ltd. 2024). As shown in a recent meta-analysis, at least the time-consuming process of AI development is mitigated by startups having relatively more employees in relation to revenue compared to traditional software service or platform technology companies (Schulte-Althoff et al. 2021). Given this, the more competitive investment environment and the importance of access to proprietary data into account, high-risk AI startups might face even more challenges in the very early design and funding phase. ...
... CrunchBase sources, updates, and validates data daily through four synergetic activities, making it a reliable data source (Ferrati and Muffatto, 2020;Schulte-Althoff et al., 2021). More than 4,500 global investment firms update their own profiles, which allow CrunchBase to access the most up-to-date data. ...
Developing desirable consumer IoT products becomes the challenge for emerging businesses. Lack of clear understanding about the functions and desirability of such products has led to a lower level of consumption than was expected. The purpose of this paper is to propose a value-based framework for product desirability, and to examine value propositions in terms of product value, product features, and user experiences by considering emerging businesses. Data from 982 companies was extracted from CrunchBase. Desired value factors, and product features companies seek the most to develop desirable products were identified. Functional value was offered more frequently than emotional value or social value. Safety, interactivity, and connectivity are the most significant features considered by companies. Companies should consider the emotional and social aspects alongside the focus given to functional aspects. The proposed framework, and the results obtained could be important for companies to develop desirable products addressing consumer preferences.
... The revenue-employee growth pattern of AI startups is found to be similar to that of service startups (Schulte-Althoff et al., 2021). This indicates the potential for scalability and growth in the AI startup ecosystem. ...
This chapter examines the potential benefits of integrating artificial intelligence (AI) and automation within startups. Through a literature review, the study unveils how AI and automation reshape various aspects of startup operations, from market insights and customer engagement to efficiency enhancement. Ethical considerations and the evolution of human-AI collaboration are highlighted, emphasising responsible integration. Case studies showcase AI's role in augmenting human capabilities. The study suggests several promising research directions. These include exploring ethical implications, identifying industry-specific applications, examining scalability, and ensuring regulatory compliance. It culminates in practical recommendations for startups, advocating tailored adoption of AI tools, data-driven strategies, and fostering human-AI synergy. This research endeavour underscores AI and automation's potential to drive startups' success by offering ethical, innovative, and efficient pathways to navigate the evolving technological landscape.
... The emergence of different electronic devices like smartphones, tablets, laptops, smartwatches, etc., has contributed to the software industry's manifold growth, resulting in a multitude of software startups [9]. The success of deep learning, a sub-field of ML, in a wide range of applications resulted in the adoption of ML in multiple companies and pushed the growth of startups rapidly [44] leading to billions of dollars of contribution to the economy [12]. ...
Abstract—Context: On top of the inherent challenges startup
software companies face applying proper software engineering
practices, the non-deterministic nature of machine learning
techniques makes it even more difficult for machine learning
(ML) startups. Objective: Therefore, the objective of our study is
to understand the whole picture of software engineering practices
followed by ML startups and identify additional needs. Method:
To achieve our goal, we conducted a systematic literature review
study on 37 papers published in the last 21 years. We selected
papers on both general software startups and ML startups.
We collected data to understand software engineering (SE)
practices in five phases of the software development life-cycle:
requirement engineering, design, development, quality assurance,
and deployment. Results: We find some interesting differences
in software engineering practices in ML startups and general
software startups. The data management and model learning
phases are the most prominent among them. Conclusion: While
ML startups face many similar challenges to general software
startups, the additional difficulties of using stochastic ML models
require different strategies in using software engineering practices to produce high-quality products.
... CrunchBase sources, updates, and validates data daily through four synergetic activities, making it a reliable data source (Ferrati and Muffatto, 2020;Schulte-Althoff et al., 2021). More than 4,500 global investment firms update their own profiles, which allow CrunchBase to access the most up-to-date data. ...
Abstract: Developing desirable consumer IoT products becomes the challenge for emerging businesses. Lack of clear understanding about the functions and desirability of such products has led to a lower level of consumption than was expected. The purpose of this paper is to propose a value-based framework for product desirability, and to examine value propositions in terms of product value, product features, and user experiences by considering emerging businesses. Data from 982 companies was extracted from CrunchBase. Desired value factors, and product features companies seek the most to develop desirable products were identified. Functional value was offered more frequently than emotional value or social value. Safety, interactivity, and connectivity are the most significant features considered by companies. Companies should consider the emotional and social aspects alongside the focus given to functional aspects. The proposed framework, and the results obtained could be important for companies to develop desirable products addressing consumer preferences.
Mobile health startups develop innovative, sensor-based solutions that continuously collect health data. To generate added value from these large amounts of data, an integration of the solutions into the healthcare system is essential. In this context, the collaboration between interdependent healthcare stakeholders is required which can be enabled by structures considered as digital ecosystems. To understand the conditions for ecosystem participation, more specifically the incentives and disincentives for data openness, we conducted 30 interviews with four healthcare stakeholder groups in Germany and analyzed the data using a Grounded Theory approach. As a result, we developed a conceptual model that describes the integration of mobile sensor-based health solutions into digital health ecosystems. Thereby, we improve the understanding of incentives and disincentives for data openness on the collective ecosystem level, the ecosystem-stakeholder-group level, and the individual user level. Practically, we contribute by outlining important market entry barriers for mobile health startups.
Some of the world's most profitable firms own platforms that exhibit network effects. A platform exhibits network effects if, the more that people use it, the more valuable it becomes to each user. Theorizing about the value perceived by users of a platform that exhibits network effects has traditionally focused on direct and indirect network effects. In this paper, we theorize about a new category of network effects-data network effects- that has emerged from advances in artificial intelligence and the growing availability of data. A platform exhibits data network effects if, the more that the platform learns from the data it collects on users, the more valuable the platformbecomes to each user. We argue that there is a positive direct relationship between the artificial intelligence capability of a platform and the value perceived in the platform by its users-a relationship that is moderated by platform legitimation, data stewardship, and user-centric design.
Taking three recent business books on artificial intelligence (AI) as a starting point, we explore the automation and augmentation concepts in the management domain. Whereas automation implies that machines take over a human task, augmentation means that humans collaborate closely with machines to perform a task. Taking a normative stance, the three books advise organizations to prioritize augmentation, which they relate to superior performance. Using a more comprehensive paradox theory perspective, we argue that, in the management domain, augmentation cannot be neatly separated from automation. These dual AI applications are interdependent across time and space, creating a paradoxical tension. Over-emphasizing either augmentation or automation fuels reinforcing cycles with negative organizational and societal outcomes. However, if organizations adopt a broader perspective comprising both automation and augmentation, they could deal with the tension and achieve complementarities that benefit business and society. Drawing on our insights, we conclude that management scholars need to be involved in research on the use of AI in organizations. We also argue that a substantial change is required in how AI research is currently conducted in order to develop meaningful theory and to provide practice with sound advice.
The power of platform business models has grown as our economies become increasingly digital, but how companies address the challenge of platform growth to achieve a critical mass of users remains unclear. In this study, we take a business model (BM) perspective to understand how mobile payment platform providers go about addressing such a challenge. We studied how mobile payment providers engaged in innovation of their business models, and thus identified three pertaining aspects: rethinking the relationship management with retailers, creating partnerships with other actors in the payment ecosystem to complement and deliver the proposed value, and integrating and using front-end mobile technology. Furthermore, our study suggests that mobile payment providers need to adapt their role within the ecosystem to scale the platform, and that it will depend on their choice of scope of geographic availability. Finally, we suggest that mutual adaptation of BMs of platform-associated actors leads to improved diffusion of the platform offer, which also hints at the need for researchers to revisit innovation diffusion and technology adoption theories by acknowledging the importance of the BM of the offer side of technology.
Recent technological advances such as in genome sequencing have exploded bio data infra-structures including those that comprise of generic - anonymized or pseudonymized - data. As open data, the bio data infrastructures do not constrain the final application context for their data. Rather it is up to complementors, taking the role of digital entrepreneurs, to appropriate value from this data through their revenue streams while at the same time scaling their opera-tions and ventures. We undertake a qualitative explorative study of bio data ventures examining the tension of applying open generic genome data to specific contexts for customers while being able to scale their businesses. The study uses primary data from 26 interviews and secondary data to reveal six strategies that complementors use for value appropriation. We derive three mechanisms of appropriating value at different stages of the value chain for bio data analysis on open data infrastructures: data contextualizing, data decontextualizing, and data recontex-tualizing. The study sheds light to how bio data – which has received limited attention in information systems research – can be an important source of value appropriation in digital ecosys-tems.
As the platform business becomes more important, it is crucial to make adequate decisions and choices for strategies, considering influence factors in relation to the platform for each growth model. This study researched how to build a platform business in the IT industry from the perspective of a dynamic approach to understand how the platform growth model successfully enables business entities to enter the market and to continue expansion. Through 21 case studies, this research formulated the four stages of platform growth model: entry, growth, expansion and maturity, providing a conceptual framework to build a platform growth model ecosystem.
In industry after industry, data, analytics, and AI-driven processes are transforming the nature of work. While we often still treat AI as the domain of a specific skill, business function, or sector, we have entered a new era in which AI is challenging the very concept of the firm. AI-centric organizations exhibit a new operating architecture, redefining how they create, capture, share, and deliver value.
Marco Iansiti and Karim R. Lakhani show how reinventing the firm around data, analytics, and AI removes traditional constraints on scale, scope, and learning that have constrained business growth for hundreds of years. From Airbnb to Ant Financial, Microsoft to Amazon, research shows how AI-driven processes are vastly more scalable than traditional processes, drive massive scope increase, enabling companies to straddle industry boundaries, and enable powerful opportunities for learning—to drive ever more accurate, complex, and sophisticated predictions.
Digitalization has vastly increased the availability, the use, and the value of data. We show that this has implications also in the context of innovation, specifically, for basic research in the field of Artificial Intelligence (AI). While corporations in recent decades have generally shifted away from scientific research, this has not been the case in the field of AI. In AI, we show that a number of large corporations including Google, Facebook, and their Chinese counterparts hire leading researchers and publish increasing amounts of high-quality basic research. Conventional explanations of corporate science fail to fully explain why corporations would undertake this research and disseminate their results. We suggest that a central aspect of digitalization—the rising importance of data as a strategic resource—drives corporate participation in AI science and publication. Owning strategic data resources makes firms lead users of AI tools, gives them a novel comparative advantage over universities in doing research in AI, and constitutes a specialized complementary asset that facilitates value appropriation. We conclude that the phenomenon we observe reflects an overall shift in the sources of competitive advantage in AI, from exclusivity in technology to exclusivity in data.