20-PM1-03: ST1.1 & ST1.2 - The Data Science of Startups / How Artificial Intelligence is Reshaping Business Models
In recent years, several open startup databases have emerged, following the creation of Crunchbase in 2007, now one of the leading sources of information about the entrepreneurial ecosystem (Dalle, den Besten & Menon, 2017). Such databases contain detailed information on startups from all over the world, ranging from recording the identity of a company’s founders to information concerning the startups’ funding rounds or textual description of its activity. Typically, Crunchbase currently contains information about over 670 000 organizations, 780 000 individuals, 220 000 funding rounds, 92 000 investors and over 6.3 million pieces of news relating to the entrepreneurial ecosystem worldwide (in the US, 52 000 organizations, 261 000 individuals, 110 000 funding rounds, and 29 000 investors).
These new datasets have first been used by entrepreneurs and investors, but in recent years they have rapidly gained recognition in the academic world, all the more so as they allow researchers to benefit from the methods, tools and techniques associated with Data Science. Liang et al. (2016) and Sharchilev (2018) have for instance worked on predictors of startup funding or success, Gastaud et al. (2018) have developed innovative visualization techniques and metrics for innovative ecosystems, Bhamidipaty and a team of researchers at IBM (2018) have addressed the task of textual similarity matching, while DeSantola et al. (2017) and Ratzinger et al. (2018) have studied financial milestones, in relation respectively to the first female board member and the education of startup founders.
These studies are further in line with the recent increase in cross-fertilization between several disciplines associated with Data Science. They create a context where management sciences can benefit from tools developed in computer science, statistics or physics, in order to study various complex phenomena associated with innovation, R&D and entrepreneurship, typically issues related to startup funding, investor networks, temporal evolutions, etc.
In addition, the growing amount of data on startups and innovative ecosystems available has also become of special relevance for decision-makers, who expect these new endeavours to help them address the innovation challenge, and who have signaled their interest with respect to the exploitation of the new datasets in order to better inform innovation policies.
In this context, this track will welcome contributions and bring together researchers working on these topics and all the matters relevant for the new and emerging "data science of startups"
Buyouts and innovation incentives : The case of the 2008 crisis
Mines Paristech, France
The importance of start up innovation on growth has become more and more evident. Here I will look at how expectations of being acquired changed in the setting of the 2008 crisis and how this in turn affected start-up innovation decisions.
There is a recent literature on declining business dynamism (see Decker et al (2017), Gourio et al (2016)) that attempts to explain the economic slowdown with the decline in firm entry. I contribute another dimension to this literature with type of innovation and a focus on the financing and buyout channel.
The financing and entrepreneurial innovation literature is covered quite well by some surveys from Lerner (2010) and Kerr and Nanda (2009, 2014).
There is also a literature on mergers and innovation however they typically study the post merger effect on innovation (ex. Seru (2014), Sevilir and Tian (2015)). Furthermore, the innovation measures in this literature are usually a count, or citations weighted count, of patents. I am instead focusing on the pre-merger effect on innovation originality.
The finance and innovation literature focuses on the "push" effect of financing (aka more financing results in more innovation). Instead, I will consider the financial aspect as an indirect pull factor. The type of innovation is chosen to maximize the likelihood of buyout which is influenced by access to financing.
How did buyout expectations change during the 2008 crisis and what effect did the unconventional policies have on this? How does the expectation of being bought out affect a start-up's innovation decision?
I disaggregate the financing effect into two parts. First I use financing variables as regressors in a Poisson regression on the number of buyouts. I then use the predicted number of deals as a proxy for the expectation of being bought out for a new entrant in the second stage. Specifically, I look at the innovation originality of new entrants regressed on this expectation of buyout with financing variables and sector controls.
The innovation measures are built from patent data in the Patstat database maintained by the European Patent Office. This is a comprehensive database of patent applications and publications starting from the mid 19th century and covers all major countries. Patstat also includes data on publication citations, grant information, application authenticator office, priority patents, technology codes, applicant and inventor information, abstracts, legal information, etc.
The acquisitions database I use is Thomson SDC Platinum which contains details on M&A deals collected from news articles and public security filings. The information consists of target and acquiror names, address, immediate and ultimate parents, industry codes, deal announcement date, effective date, withdrawal date, whether the firm is a financial firm, deal value, percent of shares acquired, percent of shares sought, number of bidders, source(s) of funding, etc.
Since there is no common firm identifier between these datasets, I develop a fuzzy string matching algorithm to merge them based on firm names.
First, some descriptive statistics show average firm originality rising until 2008 and then clearly starting to decrease. When we look at the smaller sample of just target firms, this fall is much steeper.
Preliminary results show that a lower long term rate increases the expected number of buyouts. The short term rate also has a negative coefficient however the effect becomes less significant when we use different lead years on the dependent variable.
In the second stage, the expected number of buyouts has a negative and significant effect on the originality of a new entrant's innovation choice. This is robust to different innovation measures and different time lags. This implies that start up innovation decisions are affected by their expectations of being bought out.
As an extension, I look at specific sectors, such as the software sector and ICT sector, with a logit regression on firm pairs. Regressing innovation proximity and other spillover controls on whether the firm pair is in a buyout deal shows that proximity indeed has a positive effect on buyout likelihood.
Contribution to Scholarship
The main contribution is a novel mechanism for the growth slowdown after the 2008 crisis. Specifically, I investigate the change in start-up innovation incentives due to the influence of other firms through the buyout channel.
Contribution to Practice
This study sheds light on the impact of asymmetric policies and has implications for policy makers in financial intermediation and the field of anti-trust. It also exalts the importance of start-ups and their entry "position".
This study contributes to the literature on data science in the start-ups ecosystem and, more broadly, innovation and regulation. The use of two large datasets is one connection as is a focus on the software sector in an extension.
Decker, R., J. Haltiwanger, R. Jarmin, J. Miranda, 2017. "Declining Dynamism, Allocative Efficiency, and the Productivity Slowdown," American Economic Review, American Economic Association, vol 107(5), pages 322-326, May.
Gourio, F., T. Messer, and M. Siemer, 2016. "Firm entry and macroeconomic dynamics: A state-level analysis" American Economic Review, American Economic Association, vol. 106(5), pages 214-218, May.
Lerner, J., 2010, "Innovation, Entrepreneurship and Financial Market Cycles" STI Working Paper
Kerr, W. and R. Nanda, 2014 " Financing Innovation" Harvard Business School Entrepreneurial Management Working Paper No. 15-034.
Seru, A., 2014 "Firm Boundaries Matter: Evidence from conglomerates and R&D activity" Journal of Financial Economics, Volume 111(2), pages 381-405
Sevilir, M. and X. Tian, 2015 "Acquiring Innovation", Working paper
Does standardization matter to make AI start-ups attractive to venture capital? Evidence from China
Technical University of Denmark, Denmark
Artificial intelligence (AI) research is young, and the AI industry is even younger. During the past few years, we have witnessed a global booming of start-ups with the wide range of technologies and business models, making the AI industry into a fertile ground for investment, technology development and business creation.
Venture capital (VC) plays an important role for innovation and entrepreneurship by investing start-ups of high growth potential (Popov & Roosenboom, 2013). Venture capitalists judge the attractiveness of a start-up for investment by evaluating the team, business concept, technology, market, finance, and competitive landscape of each case with various selection preference and meanwhile subject to moral hazard (Kanniainen & Keuschnigg, 2002).
As AI is a new industry, which is undergoing an ongoing development of relevant technologies, it becomes a lucrative target for VC. Meanwhile, AI industry is high of uncertainty both in technology and market, making it highly risky for VC. A crucial factor of industry structure that matters for VC to mitigate risks is standardization, which affects technology diffusion and the way in which start-ups can benefit from technological change (Katz & Safranski, 2003; Wang et al., 2016). The literature clarifies key functions of standardization (Tassey, 2000).
Few studies have addressed the impact of standardization on the attractiveness of start-ups to VC. Given the booming of and surging amount of VC investment into AI industry, a gap needs to be filled by researching how VC makes investment decision while AI standardization is still in its early phase.
How does the level of standardization in sub-sectors of AI industry influence the attractiveness of AI start-ups to venture capital investment in the Chinese context?
Sub Q1: What are the measures for standardization in AI industry in China
Sub Q2: Impact of standardization on VC investment in AI startups
This study uses quantitative methods to analyze data about VC investment in AI start-ups in China. The data on VC investment in China includes information on 1714 Chinese startups in the AI industry and their VC capital received during 1996-2018. The measures on standardization in AI subsectors are established based on the "White paper on AI Standardization" published in 2018 by the Chinese National Digital Technology Standardization Research Institution. We use regression models and survival analysis to, respectively, test the impact of standardization on the total amount of VC and how many rounds of VC each startups received.
The empirical data is one of its kind and first being used in research, as far as we know.
The preliminary results show that: (1) AI startups in the subsectors that are of high levels of standardization received higher volume of VC investment than those in the sebsectors that are of low levels of standardization; (2) AI startups in the subsectors that are of high levels of standardization were less likely to receive as many rounds of VC investment as those in the sebsectors that are of low levels of standardization. These findings are very interesting because they reveal that VC investors might view the level of standardization as a mixed signal of technology maturity (thus money is invested with more confidence when it is highly standardized) and market opportunity to explore (thus money is continuous invested through many rounds to explore uncertain market when standardization is not in place or at very early stage of development).
Contribution to Scholarship
This study first contribute to the literature on the relationship between standardization and innovation in a specific manner, as it focuses on the AI industry. Second, it shows the influence of market turbulence in a fast-growing industry, AI, which featured with a large number of startups, on the decision making of VC investment might be mitigated by the level of standardization in various degrees across different subsectors of AI industry. Thus, this study also contributes to the literature on the role of VC on growth companies.
Contribution to Practice
This study makes practical implication for VC managers and AI startups, respectively, to understand how to decide on investment strategy (and make the right portfolio) and how to attract VC investment. AI startups can use these insights as complementary knowledge to the design of their business models.
This study addresses an important question about how attractive an AI startup is to VC investment, which is crucial for its survival, growth and design of business model. This study also reveals the variety of AI standardization in China, a fertile playground for AI technology, capital, and market.
Kanniainen, V., & Keuschnigg, C. (2003). The optimal portfolio of start-up firms in venture capital finance. Journal of Corporate Finance, 9(5), 521-534.
Katz, J. A., & Safranski, S. (2003). Standardization in the midst of innovation: structural implications of the Internet for SMEs. Futures, 35(4), 323-340.
Popov, A., & Roosenboom, P. (2013). Venture capital and new business creation. Journal of banking & finance, 37(12), 4695-4710.
Tassey, G. (2000). Standardization in technology-based markets. Research policy, 29(4-5), 587-602.
The Business of Artificial Intelligence: AI Start-up Firms
1New York University, United States of America; 2Boston University, United States of America
New machine learning techniques have led to an acceleration of “artificial intelligence” (AI). Numerous papers have projected substantial job losses based on assessments of technical feasibility. But what is the actual impact and does this impact disproportionately affect start-ups?
The literature provides inconsistent results on quantifying the impact of AI-related technologies on the labor market. Frey and Osborne (2017) base their estimate on a technical evaluation of “automatability,” the technical feasibility of automation. Partially automating a job can increase employment in that occupation or industry as well as decrease employment (Acemoglu and Restrepo 2018). In the recent empirical literature, Arntz et al. (2017) modify the basic estimates of Frey and Osborne (2017) to account for the partial nature of automation. However, they do not consider the adoption of the technology or the labor demand effects that might cause automation to increase employment. Several papers have looked at the impact of robots on employment, with differing results (Acemoglu and Restrepo, 2017; Graetz and Michaels, 2015; Dauth et al., 2018).
There is no consensus in the literature on barriers to entry specific to start up firms which rely on data to train AI used in product creation and development.
Does this impact AI disproportionately affect start-ups? Does the nature of training AI with data create a barrier to entry for start-up firms?
Quantitative and conceptual. Econometric analysis of the data from the AI survey using OLS and logit regressions models. Theory building off of the economic concept of barrier to entry in regards to data.
Data Collection, completed survey of 180 WW AI start-up firms. Then we paired the firms' survey to other data sources, such as Crunchbase and the Census (US only).
There is some evidence that startups may face barriers to entry in certain major industries. The survey of AI start-up firms does not support alarmist predictions about the impact of AI. The actual effect of AI on jobs seems much more muted and varied by occupation. Only half of the start-up firms strongly agreed that labor cost reduction was a benefit to customers. And survey respondents replied that their customers are using AI to create jobs in certain occupations about as often as they use it to eliminate jobs.
Contribution to Scholarship
The paper contributes to scholarship in better understanding the extent of these barriers to entry for AI start-up firms, the nature of these firms' business models, and possible policy remedies.
Contribution to Practice
The results of the paper have implication on policies, which could provide assistance for AI start-ups that do not have access to the data needed to adequately train their products.
This paper closely fits the theme of the Track 1.2 on the data science start-ups
Accenture (2018). “It’s Learning, Just Not as We Know It.” Available: https://www.accenture.com/us-en/insights/future-workforce/transforming-learning.
Acemoglu, D. and P. Restrepo (2018). "Artificial Intelligence, Automation and Work." NBER Working Paper.
Arntz, M., et al. (2017). "Revisiting the Risk of Automation." Economics Letters 159: 157-160.
Autor, D. and A. Salomons (2018). "Is automation labor-displacing? Productivity growth, employment, and the labor share." BPEA Conference Drafts.
Autor, D. H., et al. (2008). "Trends in U.S. Wage Inequality: Revising the Revisionists." The Review of Economics and Statistics 90(2): 300–323.
Autor, D., Katz, L. “Changes in the Wage Structure and Earnings Inequality”. In: Ashenfelter O, Card D, editors. Handbook of Labor Economics. 3A. Amsterdam: North-Holland; 1999.
Athey, S. (2017). Beyond Prediction: Using big data for policy problems. Science, Vol. 355, Issue 6324, pp. 483-485
Bajari, P., et al. (2018). "The Impact of Big Data on Firm Performance: An Empirical Investigation." NBER Working Paper 24334.
Bessen, J. (2016). "How Computer Automation Affects Occupations: Technology, Jobs, and Skills." Boston University School of Law Working Paper No. 15-49.
Bessen, J. (2017). "Automation and Jobs: When Technology Boosts Employment." Boston University School of Law, Law & Economics No. 17-09.
boyd, d and Crawford, K. (2012). “Critical Questions for Big Data.” Information, Communication & Society, 15.5. 662-679
Brynjolfsson, E., McAfee, A. Race against the Machine. 2017
Dauth, W., et al. (2018). "Adjusting to Robots: Worker-Level Evidence." Working Paper.
Ford, M. . Rise of the Robots. London: OneWorld Publications. 2015
Frey, C. B. and M. A. Osborne (2017). "The future of employment: How susceptible are jobs to computerisation?" Technological Forecasting and Social Change 114: 254-280.
Goos, M., et al. (5863). "Job Polarization in Europe." American Economic Review Vol. 99: 58-63.
Graetz, G. and G. Michaels (2017). "Is Modern Technology Responsible for Jobless Recoveries?" IZA Discussion Papers No. 10470.
Hartmann, P. and J. Henkel. (2018)." Really the new oil? A resource-based perspective on data-driven innovation." Technical University of Munich working paper.
Jin, Wang, and Kristina McElheran. "Economies Before Scale: Survival and Performance of Young Plants in the Age of Cloud Computing." (2017).
Lambrecht, N. and C. Tucker (2017). "Can Big Data Protect a Firm from Competition." CPI Antitrust Chronicle(January).
Schwab, K. Fourth Industrial Revolution. New York: Crown Business. 2017
Sokol, D. Daniel, and Roisin Comerford. "Antitrust and Regulating Big Data." Geo. Mason L. Rev. 23 (2015): 1129, https://ssrn.com/abstract=2834611.
Susskind, R. and D. Susskind (2015). The Future of the Professions. Oxford, Oxford University Press.
Tucker, C. (2017). “Privacy, Algorithms and Artificial Intelligence.” Working Paper, MIT
Varian, H. R. (2010). "Computer Mediated Transactions." American Economic Review 100(2): 1-10.
Varian, H. R. (2014). Big Data: New Tricks for Econometrics. Journal of Economic Perspectives, Vol. 28, N. 2, Spring 2014, Pp. 3–28
Canadian Government Invests C$170M into the Future of AI. Aibusiness.com (March 30, 2017) https://aibusiness.com/canadian-government-invests-c170m-into-the-future-of-ai/
France to spend $1.8B on AI to compete with US and China. (March 29, 2018). https://www.reuters.com/article/us-france-tech/france-to-spend-1-8-billion-on-ai-to-compete-with-u-s-china-idUSKBN1H51XP
McKinsey Report (2017). Notes from the AI Frontier.: Applications and value of deep learning. https://www.mckinsey.com/featured-insights/artificial-intelligence/notes-from-the-ai-frontier-applications-and-value-of-deep-learning#part4
Edge AI driven technology advancements paving way towards new capabilities
1Husqvarna AB, Sweden; 2KTH
As industries are embracing AI driven innovation within operations and offerings, we need to understand different type of values these capabilities drive and significance of those experienced for different stakeholders. Based on experiments within a Swedish OEM, this paper explores capabilities enabled through Edge AI and study transformation driven within value typology.
While IoT enables connectivity and computing on our end devices (1), it has been around for some time, there are multiple theories, frameworks and research on opportunities, benefits and challenges around it (2). With IoT also came the possibility of Data Capture at the source which fuels the next chapter of technology with AI (3).
There are studies & literature worked upon constantly within AI detailing technology and modelling techniques.
With above two concepts combining together as Edge AI (4), introduces new capabilities to be self & environmental aware, which are then used as intelligence and smart devices within offerings of PSS.
Edge AI is just a technology unless we utilize it to deliver against perceived value (5) for our stakeholders and hence value models (6) and existing typologies needs to be looked into from this context together with what impact does innovation drive on the design of future products and services (7).
There are multiple research references on various value models & typologies from different stakeholders perspective like Customers, Suppliers, Partners, Businesses, etc. Also, AI is a big research area but primarily concentrates on modelling techniques. Only limited attention has been given to how AI Technology driven capabilities impact the value typology for different stakeholders.
Q1: How Edge AI creates new capabilities for the development of organizational Product Service Systems?
Q2: What types of values can edge AI bring to the Product Service System, and what significance can this hold for an enterprise?
There were three parallel projects initiated within a Swedish manufacturer with three different start-ups spanning from two to four months each. They utilized Edge AI but with different modelling techniques. Use cases for intelligence were identified for manufacturer’s garden equipment’s, data collected for the same through existing sensors on the equipment’s and fed into the modelling techniques of above start-ups. This yielded AI models, which then deployed back on the same equipment’s to exhibit intelligence and hence identifying capabilities that Edge AI enables. Above was followed by an investigation of value typology for stakeholders based upon Edge AI enabled capabilities.
Project 1 had the use case of estimating remaining useful life of the part; in this use case, collected data included Impedance, current & temperature over multiple field usage of the part on two different equipment’s. This data was then fed into Startup 1 Machine Learning engine which not only extracted patterns but also generated models with a footprint that was deployable on the part within the equipment.
Project 2 had the use case of identifying lubrication requirements proactively on the equipment; in this use case, collected data included acoustic emission, Rotation per minute (IR) & temperature for one equipment used under various conditions. This data was then fed into Startup 2 Machine Learning engine which yielded patterns and generated models were deployable on the equipment.
Project 3 had the use case of identifying state of the equipment; in this use case, collected data included throttle angle, rotation per minute & temperature for two equipment’s used under various conditions. This data was then fed into Startup 3 Machine Learning engine yielding patterns, generating models and deployed on the equipment.
After deployment of all above models on respective equipment the capabilities enabled were identified through observations and experiment stakeholder interviews.
There were different capabilities enabled through Edge AI deployments on equipment’s during the experiments. Some of these has minimal impact on perceived value for stakeholders and some have much more, but nevertheless every capability is integral and important aspect of Edge AI value which lays foundation for other capabilities. Also, these capabilities do not necessarily follow a bottom up approach for value typology appropriation but can also be rather intertwined together.
The capabilities can be classified as: Data Efficiency (Intelligent filtering of data, not aggregating or tampering raw data, less bandwidth, reducing costs of transfer), Data Quality (eliminating noise from captured data at source, using soft sensing without hardware), Latency (real-time capabilities), Reliability (reduced connectivity concerns, monitoring & control functions), Integrity (security features, sustainability agenda, seamless deployment of services) and Personalization (self-learning equipment-s, safety offerings, individualistic add-on services).
The study is still being conducted, hence in coming coarse of time, these capabilities would be mapped against stakeholders perceived value topology as this research and paper continues.
Contribution to Scholarship
As highlighted in the literature gap, on one hand there is much active & published research on Edge AI Technology w.r.t. modelling and algorithms not much research can be found which inter-relates how AI Technology enabled capabilities impact the value typology and different stakeholders.
In step one, through a series of experiments, this work contributes towards identification of various capabilities enabled by Edge AI Technology when deployed on garden used equipment’s. Then as step two, this study also contributes by categorize these capabilities and outlining the value typology for perception models on stakeholders. As per above contributions, enabled capabilities & identified classes can be discussed for potential use to understand the effect on value typology for different stakeholders and thereby bridging the gap between technical & business research.
Contribution to Practice
During the experiments within this study, multiple Edge AI Models where created and deployed using different frameworks and implementation methods through different startups. Findings from the study can be used to create & design new Edge AI based Product Service System which delivers improved value to the stakeholders. This is done by mapping the outlined Edge AI capabilities to Product Service System design and hence exploit Edge AI capabilities.
With almost 40 billion connected devices already (about 5 connected devices per individual) and estimated to be doubled by 2025, the backbone of connectivity to cloud and these devices will be depleted and hence Edge AI will play an important role in analyzing the real-time incoming stream and hence reshaping business & value models.
(1)M. E. Porter and J. E. Heppelmann, “How Smart, Connected Products Are Transforming Competition”, Harv. Bus. Rev., No. November, pp. 64-89, 2014.
(2)I. Lee and K. Lee, “The Internet of Things (IoT): Applications, investments, and challenges for enterprises”, Bus. Horiz., Vol.58, No.4, pp. 431-440, 2015.
(3)M. Boehm and O. Thomas, “Looking beyond the rim of one’s teacup: A multidisciplinary literature review of Product-Service System in Information Systems, Business Management and Engineering & Design”, J. of Cleaner Production, Vol.51, pp. 245-250, 2013.
(4)Ali Keshavarzi and Wilbert vs den Hoek, “Edge Intelligence – on the challenging road to a trillion smart connected IoT devices”, IEEE Design & Test (Early Access), 13th Feb 2019.
(5)Michael G. Jacobides, Thorbjørn Knudsen and Mie Augier, “Benefiting from innovation: Value creation, value appropriation and the role of industry architectures”, Research Policy
Volume 35, Issue 8, October 2006, Pages 1200-1221.
(6)Per Lindstedt and Jan Burenius “The Value Model“
(7)Anthony W Ulwick “What Customers Want: Using Outcome-Driven Innovation to Create Breakthrough Products and Services”