Faster horses: Collaborative AI innovation between incumbents and startups

Lara Anne Hale Assistant Professor Department of Management, Society, and Communications Copenhagen Business School Dalgas Have 15 2000 Frederiksberg, Denmark Mail: lh.msc@cbs.dk, Phone: +45 42421774  Response to the Call for Papers for Reshaping Work: AI@Work Division: Business 

Faster horses: Collaborative AI innovation between incumbents and startups 

“If I had asked them what they wanted, they would have said faster horses!” – attributed to Henry Ford 

Introduction  Amidst the blossoming of technological development of the fourth industrial revolution, so-called ‘Industry 4.0’ (Schwab, 2017), organizations are seeking to integrate the potential benefits of artificial intelligence (AI) into their businesses. The idea is that organizations should become smarter and more efficient (Marr, 2018; Kaplan & Haenlein, 2019). The McKinsey Global Institute suggests that by 2030 70% of all businesses will have adopted at least one form of AI; but also that an “S-curve pattern of adoption and absorption of AI is likely” (Bughin et al., 2018). On the one hand, how incumbent businesses can adopt AI into their core activities is not always clear. On the other hand, the number of AI startups disrupting existing markets is exploding. It is widely acknowledged that incumbents struggle to innovate — at least quickly — and that startups innovate quickly while struggling to scale (Hockerts & Wüstenhagen, 2010; Weissbrod & Bocken, 2017; Slot, in press), but collaboration between incumbents and startups for joint innovation is increasing, especially in Europe. And whereas it is not uncommon to ask how AI disrupts business, there is less known about how collaboration can disrupt AI. This research examines the collaborative innovation between a building industry incumbent, the VELUX Group, and an AI startup, Airboxlab, to ask the following question:  How can the collaborative innovation between incumbents and AI startups shape the development of AI?  1  Research Case  Despite the substantial adoption of AI suggested by the McKinsey Global Institute and the building industry’s keen interest in developing AI-driven systems, the actual integration of AI into the industry is painfully slow. In their 2018 global survey on ‘PropTech’ (property and real estate technologies), KPMG finds that 66% of respondent organizations do not even have a business- wide vision or strategy for digital and technological innovation (KPMG, 2018), let alone specific innovation regarding AI. In response to this challenge, the Urbantech program12was launched at BLOXHUB in Copenhagen, Denmark. The program seeks to stimulate innovation for sustainable building through corporate and startup engagement and runs August through October 2019 with the aim of launching a pilot. Mid-2019, Rainmaking gathered ten mid-growth, product-ready startups and matched them with the innovation needs of the corporate partners. One such pairing was the VELUX Group, who is interested in AI for facilitating healthy indoor climate, with Airboxlab, who offers AI- and machine learning-mediated healthy indoor climate technology (through their cloud-based monitoring system Foobot3). This partnership is the focus of the research.  Methodology  The research methodology combines an organizational ethnography and action research, the latter applying specifically to the Urbantech program. The ethnography is ongoing within the VELUX Group from January 2018 (concluding December 2020) and concerns smart building business models. It is primarily based on observations — with one to two visits per week at the VELUX Group headquarters and note taking (72 field notes and 42 internal documents) (Strauss & Corbin, 1998) –, and four case studies built upon 60 interviews with 68 interviewees. This methodology is useful for understanding processes within an organization (Langley, 1999), such as processes of technological innovation. In the Urbantech program, my role switched from observer to active participant for the dual purposes of research and organizational problem solving (McKay & Marshall, 2001). The research purpose concerns collaborative innovation between the two parties; whereas the problem concerns reciprocal learning and, ideally, joint application of Airboxlab’s solutions to a VELUX Group-relevant pilot. The action research has been based on four workshops lasting 1.5 – 4 hours each between 27 August and 18 October 2019 (present day), reflective discussions within the VELUX Group, as well as the Airboxlab CEO’s and my participation in Techfestival’s Corporate x Startup Summit4 on 5 September 2019, focusing on the emerging European joint technological innovation strategies between corporates and startups. Interactions, direct quotes, and reflections were recorded for all workshops and meetings.  1 https://urbantechprogram.io/ 2 Made possible by the philanthropic partners Realdania and the Danish Industry Foundation; corporate partners VKR Holding and the VELUX Group, COWI, and EWII; supporting partners BLOXHUB, the Danish Architecture Center, the Danish Design Centre, and Amazon Web Services (AWS Activate); and the coordinator Rainmaking. 3 https://foobot.io/ 4 https://techfestival.co/event/corporate-x-startup/  2  Table 1: Workshop descriptions.  Workshop Date Location Participants Focus Workshop 1 27/08/2019 BLOXHUB Jacques Touillon (Airboxlab), Inouk Bourgon  (Airboxlab), Lone Feifer (VELUX), Peter Folbjerg (VELUX), Markus Holm Steffensen (VELUX), Thorbjørn Færing Asmussen (VELUX)  Initial Findings Though the program will continue through October, the initial findings are that in addition to the VELUX Group and Airboxlab working to innovate with each other through the pilot development, there is also negotiation over the potential applications of AI technology. These negotiations arise from the back and forth of discussing, as Jacques Touillon, CEO of Airboxlab, succinctly puts it: “How can we ‘augment’ the value of the existing product with value from the existing technology?” (Workshop 3). In this case, the existing product is residential housing solutions for healthy indoor climate, and the existing technology combines sensor systems and artificial intelligence with machine learning. However, the matter of user interface and system interaction with users becomes a point of dispute.  Planning, team formation  Workshop 2 04/09/2019 VELUX Jacques Touillon (Airboxlab), Inouk Bourgon  (Airboxlab), Markus Holm Steffensen (VELUX), Thorbjørn Færing Asmussen (VELUX)  Proposal development, VELUX research and experiment backgrounds Workshop 3 12/09/2019 Online call Jacques Touillon (Airboxlab), Adrien Ladond  (Airboxlab), Markus Holm Steffensen (VELUX)  Value proposition  Workshop 4 27/09/2019 Online call Jacques Touillon (Airboxlab), Inouk Bourgon  (Airboxlab), Markus Holm Steffensen (VELUX), Thorbjørn Færing Asmussen (VELUX)  Digital twin development, Airboxlab research and experiment backgrounds  3  Table 2: Entries relating to use and user interfaces.  Workshop User-relation Speaker (if specified) Workshop 1 “It’s mostly about the user experiences, and that’s  where things are falling short.”  The VELUX Group’s rather extensive innovation experience (e.g. Model Home 20205) supports that the way in which products are used is what accounts for the ‘performance gap’ — the gap between modeled results and actual results of building performance (Menezes et al., 2012). More specifically, with the RenovActive demonstration, researchers found that a key component of automated indoor air quality is user-informed priority settings (te Braak et al., 2019). However, according to Airboxlab, the data with which the AI agent is trained is composed of environmental and building data, but not user data. The VELUX Group has demonstrated that user consideration is essential for energy efficiency and air quality; but the AI reward system has not been developed to respond to this.  5 https://www.velux.com/health/future-housing  Jacques Touillon (Airboxlab)  Workshop 1 “Homeowners don’t want five different places to go to  adjust the settings of their house”  Thorbjørn Færring Asmussen (VELUX) Workshop 1 “The ‘user case’, if you expose it to users – that’s what  takes a long time.”  Lone Feifer (VELUX)  Workshop 1 Still working on showing the air quality data to users. VELUX side Workshop 1 “The only thing the end user is interested in is  yesterday, last week, my air quality was 90% healthy.”  Jacques Touillon (Airboxlab) Workshop 1 Chat bot based interation with the user is possible, but  they think it will only be used for a short period after setup  Inouk Bourgon (Airboxlab)  Workshop 2 Discussions on the difficulty of feedback loops and  communications with users.  Both sides  Workshop 2 Can simply indicate “We are airing,” but users want to  know why something is happening.  Markus Holm Steffensen (VELUX) Workshop 2 On user awareness: “This is not on the top of the  head.”  Markus Holm Steffensen (VELUX) Workshop 2 On the greatest problem being: users acknowledge  indoor air quality is important, but can’t feel it.  Markus Holm Steffensen (VELUX) Workshop 3 The value of products is only really delivered if used properly, and we know they’re not.  VELUX side  Workshop 3 Main driver for innovation: air quality turned into value  for the end-user.  Jacques Touillon (Airboxlab) Workshop 3 But during set up is probably the last time that users  engage with the system.  Markus Holm Steffensen (VELUX) Workshop 3 Learnings from VELUX experiments: operability is  important, but also the ability for users to make their own priority settings.  Thorbjørn Færring Asmussen (VELUX)  Workshop 4 Agility with self-learning system desired for indoor climate control; response to users as part of this.  Thorbjørn Færring Asmussen (VELUX) Workshop 4 AI not as ‘magic’ — train based on available  environmental data (utility bills, weather data, performance calculations in real time).  Jacques Touillon (Airboxlab)  4  Table 3: Entries relating to artificial intelligence.  Workshop AI-relation Speaker (if specified) Workshop 1 Employing a digital twin as the aim, as three months is  too short for a full pilot.  Thus the collaboration between the VELUX Group and Airboxlab comes to an ideological impasse: how to negotiate an AI solution that both technologically optimizes and is able to incorporate user data and preferences.  Conclusion While incumbent businesses are asking how they might integrate AI, AI startups are asking how they might scale demand for their products and services. However, if we ask the people what they want from AI, they want ‘faster horses’, i.e. more efficient buildings. Though the energetic and air quality outcomes are dependent on users, the focus of AI building solutions is drawn towards technical modeling. In this way, the bias away from user engagement inherent in the construction industry already starts to transfer to AI. Yet, while this collaborative innovation can certainly bring AI knowledge into the incumbent, it can also open pathways for bringing incumbent knowledge into the AI’s development. We can ask the people what they want, and then deliver what they need.  VELUX side  Workshop 2 AI — agile, self-learning system is what would bring the  added value for VELUX.  Thorbjørn Færring Asmussen (VELUX) Workshop 2 “That’s nothing like a turnkey solution” — have to start  with data and then build the algorithm.  Jacques Touillon (Airboxlab) Workshop 3 The AI is meant for optimization and quantification. Jacques Touillon  (Airboxlab) Workshop 3 “The AI is not delivering a user experience, but rather  adding the experience.”  Jacques Touillon (Airboxlab) Workshop 4 One of AI building simulations as reducing energy  consumption, though not much improving comfort.  Jacques Touillon (Airboxlab) Workshop 4 Machine learning as based on reward system and  reciprocation among agent, environment, and interpreter.  Jacques Touillon (Airboxlab)  Workshop 4 AI as experimental; not fully implemented outside of  simulations yet, and still learning.  Inouk Bourgon (Airboxlab) Workshop 4 Starting at digital twin point to test AI simulation on  residential buildings; then decide how to proceed from there.  VELUX side  5  References  Bughin, J., Seong, J., Manyika, J., Chui, M., and Joshi, R. (2018). Notes from the AI Frontier: Modeling the impact of AI on the world economy (Discussion Paper). The McKinsey Global Institute. Retrieved 16 October 2019 from: https://www.mckinsey.com/featured-insights/artificial- intelligence/notes-from-the-ai-frontier-modeling-the-impact-of-ai-on-the-world-economy  Hockerts, K., & Wüstenhagen, R. (2010). Greening Goliaths versus emerging Davids – Theorizing about the role of incumbents and new entrants in sustainable entrepreneurship. Journal of Business Venturing, 25(5), 481–492. https://doi.org/10.1016/j.jbusvent.2009.07.005  Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’s the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence. Business Horizons, 62(1), 15–25. https://doi.org/10.1016/j.bushor.2018.08.004  KPMG (2018). The road to opportunity: An annual review of the real estate industry’s journey into the digital age (KPMG Global PropTech Survey). Retrieved 16 October 2019 from: https://home.kpmg/uk/en/home/insights/2018/05/kpmg-global-proptech-survey-2018.html  Langley, A. (1999). Strategies for theorizing from process data. Academy of Management Review, 24, 691-710.  Marr, B. (2018). The 4th Industrial Revolution Is Here — Are You Ready? Forbes. Retrieved 16 October 2019 from: https://www.forbes.com/sites/bernardmarr/2018/08/13/the-4th-industrial- revolution-is-here-are-you-ready/#4b8c3047628b  Menezes, A. C., Cripps, A., Bouchlaghem, D., & Buswell, R. (2012). Predicted vs. actual energy performance of non-domestic buildings: Using post-occupancy evaluation data to reduce the performance gap. Applied Energy, 97, 355–364. https://doi.org/10.1016/j.apenergy.2011.11.075  McKay, J., & Marshall, P. (2001). The dual imperatives of action research. Information Technology & People, 14(1), 46–59.  Schwab, K. (2017). The fourth industrial revolution. New York: Crown Business.  Slot, M. (in press). About introvert incumbents and extravert start-ups: An exploration of the dialectics of collaborative innovation in the Dutch journalism field. Journalism, 1–16. https://doi.org/10.1177/1464884918794303  Strauss, A. L., and Corbin, J. M. (1998). Basics of qualitative research: Techniques and procedures for developing grounded theory, 2nd edition. Thousand Oaks: Sage Publications.  te Braak, P., Minnen, J., Fedkenheuer, M., Wegener, B., Decock, F., Descamps, F., Pauquay, S., Feifer, L., Hale, L. A., Asmussen, T. F., Christoffersen, J. (2019). How to transform  6  European housing into healthy and sustainable living spaces? – the RenovActive principles tackle climate and renovation challenges. Manuscript submitted for publication.  Weissbrod, I., & Bocken, N. M. P. (2017). Developing sustainable business experimentation capability – A case study. Journal of Cleaner Production, 142, 1–14. https://doi.org/10.1016/j.jclepro.2016.11.009  7