EFFECTS OF USER COMMUNITY SENSING CAPABILITY IN DIGITAL PRODUCT INNOVATION: EVIDENCE FROM THE VIDEO GAME INDUSTRY

Digital technology continues to extend the co-creative role of users and user communities as sources of innovation-conducive knowledge. While the potential of user communities in this context is well established, little is known about the capabilities needed to successfully manage the interface with virtual user communities. The paper investigates User Community Sensing (UCS) capability as a measure of rms’ ability to anticipate changes and opportunities for innovation by interfacing with relevant user communities. Based on existing research and data from 173 product innovation projects, the study employs structural equation modelling to test the hypothesised effects of this capability on product innovation performance and product innovation speed. The results indicate that UCS capability affects performance positively and indirectly by increasing knowledge about users but has no signi cant effect on speed. As well as contributing to the literature on innovation management, the study has a number of implications for practitioners.


Introduction
Advances in digital technology render product innovation processes increasingly open (Chesbrough and Bogers, 2014) and decoupled (Ehls et al., 2020).However, although external users and user communities have long been recognised as knowledge resources for innovation (Chang and Taylor, 2016), understanding how rms can systematically leverage virtual communities in this context is still more art than science.As the ability to match emergent needs to product innovation becomes increasingly pivotal (von Hippel and Kaulartz, 2020), rms are moving away from ad hoc approaches like crowdsourcing and lead-user methods to develop more formal and institutionalised capabilities (Ehls et al., 2020).However, little is known about the capabilities needed to manage the interface with virtual user communities or how this capability might affect the performance and the ef ciency of the innovation process (Chang and Taylor, 2016;Roberts et al., 2021).
To address this issue, the present paper investigates what we characterise as User Community Sensing (UCS) capability: a rm's ability to utilise decoupled virtual user communities for digital product innovation.To conceptualise this capability, we draw on previously separate strands of marketing and strategy research exploring rms' capacity to identify and exploit external market information in innovation (Day, 1994(Day, , 2020;;Teece, 2007).The paper responds to calls for more research on managing users in innovation, especially in virtual community settings (Bogers et al., 2016;Chesbrough et al., 2018), and further conceptualisation of digital innovation management (Lyytinen et al., 2016;Nambisan et al., 2017).In particular, the study explores the capabilities needed to leverage digital information from decoupled actors beyond the rm's direct control (Ehls et al., 2020).
Drawing on survey data from 173 development projects in the video game industry, we used structural equation modelling (SEM) to assess the impact of UCS on product innovation processes.Our results con rm that UCS has a positive effect on rms' knowledge about their users and a positive indirect effect on product innovation performance, measured as nancial performance.However, we found no signi cant relationship between USC and product innovation speed, measured as development speed.The video game industry was selected as a suitable research setting in light of the sector's established capacity to engage and co-create effectively with user communities (Burger-Helmchen and Cohendet, 2011;Dahlander and Magnusson, 2005;Parmentier and Mangematin, 2014).For that reason, our ndings also have important implications for innovation management in other sectors that are similarly fast-paced, digitalised, and technology-intensive.In a landscape placing increasing emphasis on the ability to leverage distributed innovation (Chesbrough and Bogers, 2014: 17) but where theory formation is still ongoing 2250007-3 (Chang and Taylor, 2016;Nambisan et al., 2017;Roberts et al., 2021), we conceptualise UCS as the capacity to manage the user community interface for the purposes of product innovation (Roberts et al., 2021).Rather than focusing solely on consumers and users as a source of information-the most widely studied form of distributed innovation (Ehls et al., 2020;Roberts et al., 2021)-UCS captures how rms can enable, manage, and utilise the interface with users to co-create innovation throughout the product development process.This conceptualisation also enhances the existing understanding of how to manage distributed innovation interactions around a digital artifact (Becker et al., 2021)-in this case, video game development.Additionally, as previous research has generally overlooked the development of structured processes of engagement with virtual user communities (Koch and Bierbamer, 2016), especially as a driver of performance (Bogers et al., 2016;Randhawa et al., 2016), the empirical study illuminates how UCS impact product innovation processes and their outcomes.
Our ndings also have practical relevance for managers seeking to involve user communities in digital product innovation.The observed positive indirect effect on product innovation performance (mediated by increased knowledge about product use and user needs) con rms the need for formal UCS processes.Another nding of interest to managers is that UCS seems to have no signi cant effect on product innovation speed; in other words, while resourcing the user community interface has no negative impact on project deadlines or milestones, there is no net positive effect on the speed of development.
The rest of the paper is structured as follows.The next section outlines the conceptual and theoretical background, including the hypothesised effects of UCS capability, drawing on previous research on rm sensing capabilities and the role of user communities in product innovation.After describing data collection and analysis, we discuss the empirical ndings.Finally, we identify the study's limitations and some implications for future research and practice.

Firm sensing capabilities
Previous studies of distributed innovation have highlighted the role of users as external knowledge sources in the product development process.These studies (and related research) as a rule explored diverse efforts to engage users through innovation contests, crowdsourcing (Malhotra et al., 2017;Gatzweiler et al., 2017), or versions of the lead-user method (von Hippel, 1986).Increasingly, however, rms are developing dedicated and specialised capabilities in preference to these ad hoc approaches (Ehls et al., 2020).

2250007-4
In general terms, rm capabilities are bundles of skills and knowledge that facilitate adaptation to market changes (Moorman and Day, 2006); rather than ad hoc solutions to a problem (Day, 2014), these processes and abilities are institutionalised and repeatable.As they are developed and emergent, they are not easily acquired or transferred and are therefore somewhat idiosyncratic, although with generalisable elements (Eisenhardt and Martin, 2000).Effective capabilities are critical for performance in dynamic environments (Makadok, 2001;Teece et al., 1997;Zollo and Winter, 2002) such as the digital industries.In such contexts, sensing capabilities play a critical role in product innovation (Mu, 2015).
Two distinct literatures-marketing (e.g., Day, 1994) and strategy research on dynamic capabilities (e.g., Teece, 2007)-inform our existing understanding of sensing capabilities.In both areas, sensing is understood as the ability to search for external information and to internalise that information.Both research streams emphasise the critical impact of sensing on performance, as the processes that underpin sensing enable a rm to anticipate change in markets and technologies and to identify opportunities for innovation (Day, 1994;Foley and Fahy, 2009;Teece, 2007).While the two perspectives overlap in terms of how they conceptualise sensing, they differ in focus (Day, 2020).Table 1 lists some key studies of sensing capabilities.
Strategy researchers view sensing as a dynamic capability that underpins the " rm's ability to integrate, build, and recon gure internal and external competencies to address rapidly changing environments" (Teece et al., 1997: 516).This dynamic capability perspective is commonly seen as an extension of the resourcebased view (RBV) (Schilke et al., 2018), which addresses how internal resources are used to gain competitive advantage.While the dynamic capability perspective also addresses how external knowledge is used to enhance internal processes (Teece, 2007), its inside-out focus is rooted in its RBV heritage (Day, 2014(Day, , 2020)).In general, the dynamic capability perspective assumes that the external search implicit in sensing is relatively well-de ned, mandated, and structured (Day, 2014).However, marketing studies place greater emphasis on sensitivity to weak signals and acting on partial information beyond the well-de ned search or "scouting" in the dynamic capability literature (Day, 2014: 28).
In focusing on the market rather than internal resources as the starting point for sensing processes, marketing studies emphasise the role of sensing in fostering proximity to users and markets and supporting innovation by creating a permeable net of knowledge ows between employees and users (Day, 2011).In this view, sensing is both responsive and proactive, addressing existing and explicit as well as latent and emerging needs (Narver et al., 2004).By developing effective sensing processes, a rm can more effectively generate, internalise, and disseminate market information in order to develop better product offerings (Kohli and Jaworski, 2250007-

2250007-7
1990).To that extent, effective sensing capabilities can be said to enable an outside-in approach to innovation that is strongly market-oriented.

Community sensing capabilities
User communities have known potential to contribute to innovation processes, but there is also evidence that rms require dedicated capabilities if they are to leverage that potential effectively (Dahlander and Magnusson, 2005;Parmentier and Mangematin, 2014).In light of the limited existing efforts to conceptualise these capabilities, we differentiate here between community sensing and sensing in general, and we contend that the former should be treated as a conceptually distinct special case of the latter.This distinction is warranted by the complexity of user communities and the need for dedicated processes to harness this type of user involvement to enhance performance (Arvanitis et al., 2015).Previous research con rms that digital user communities differ from the wider customer base (Burger-Helmchen and Cohendet, 2011) and a market sensing capability cannot be readily deployed in community settings.The transformative shift in product innovation toward digitally enhanced openness and user co-creation requires a corresponding conceptual evolution (Ehls et al., 2020;Nambisan et al., 2017;Roberts et al., 2021).To be effective, we argue that rms need both the structured and well-de ned external search processes described in the strategy literature and the ability to anticipate and act on weak signals advocated by marketing studies.To that extent, any conceptualisation of UCS capabilities must encompass both inside-out and outside-in dimensions (Day, 2020).
As an active user community can provide more complex and up-to-date information about user needs than a rm itself possesses at any given time (Cui and Wu, 2017;Jeppesen, 2004), interfacing with users can enhance innovation by supplying knowledge resources beyond those of the rm or its network (Balau et al., 2020).However, developing the requisite capabilities to leverage this information for innovation presents signi cant challenges in terms of understanding, navigating, and interacting in community contexts.Based on previous sensing research, we de ne UCS as the rm's ability to anticipate market change and emerge opportunities for innovation by interfacing with user communities.Figure 1 visualises the main processes and key practices and routines associated with this capability, which are outlined in more detail below.
Two sensing processes are critical for UCS capability.The rst of these is developing and maintaining user proximity and peripheral vision (Day and Schoemaker, 2016), enabling the rm to detect weak early signs of an opportunity for innovation.This determines search breadth and depth and the extent of passive or active scanning of outside sources (Day, 2020;Teece, 2007).While rms have collected 2250007-8 information from users for many years (Chang and Taylor, 2016), managers still tend to underestimate the user's potential contribution to innovation (Bradonjic et al., 2019).In this regard, one fundamental requirement is to develop ties with user communities in order to accommodate information ows at points of interaction.To foster the necessary awareness, alertness, and open-mindedness to identify and make sense of both articulated and unarticulated needs, the rm must be immersed in the user context.Empathetic interaction with users enables the sensing rm to elicit weak signals and information from user-generated content (UGC) and community discussions.
Most of the activity in user communities occurs in online forums maintained by a rm, a third party, or user groups, and these commonly contain salient information for all aspects of the product development process (Jeppesen, 2004).As users engage in internal discussion of product development, this UGC can help to identify latent and emergent needs that would be missed by traditional information-gathering methods (Ho-Dac, 2020).To elicit this innovation-conducive information, rms must "absorb" users' online conversations (rather than actively seeking them out) by leveraging unstructured and rich information that is not intended for the rm (Ho-Dac, 2020).
Much of this information is stored as digital content and is therefore relatively enduring.However, sifting, translating, and retrieving this content in an effective and timely way is not a straightforward task, as it may be codi ed in

2250007-9
context-dependent discussions and forum threads deep inside community settings.Its utility may also be temporary, as users' reactions and preoccupations shift over time.Although this freer and more native format presents challenges, it also tends to enhance the information's quality, richness, and value for product innovation (Ho-Dac, 2020;Nambisan, 2002).This depends in part on the rm's commitment to interaction with the user community during the development process rather than merely listening in when a problem or question arises.This approach also limits reliance on AI and machine learning methods (Shrestha et al., 2019), as creating and sustaining proximity to users depends largely on the efforts of employees and management.
The second key sensing process is ef cient interpretation and comprehension of codi ed information, weak signals, and the opportunities these afford (Day, 2020;Day and Schoemaker, 2016), along with the willingness and ability to act on these cues.The capacity to understand and learn is critical for interfacing more actively with the user community; if executed ef ciently, this can provide invaluable solution-related information (Nambisan, 2002) and can enhance problem solving (Poetz and Shreier, 2012).As complex networks of dispersed but virtually connected users and groups (Cova and Pace, 2006;Dahlander and Frederiksen, 2012;Dahlander and Wallin, 2006;Ganley and Lampe, 2009;Malinen, 2015), user communities often comprise distinct subcommunities within a larger structure (Burger-Helmchen and Cohendet, 2011;Plant, 2004;Schulz and Wagner, 2008).Within these networks, users with differing interests form subgroups around testing, user innovation, and content creation (Burger-Helmchen and Cohendet, 2011).To understand these groups and to detect changing user needs and concrete opportunities for product innovation, rms must develop appropriate learning and sense-making processes (Day and Schoemaker, 2016;Teece, 2007).
The subgroups within user communities are heterogeneous in terms of characteristics, skills, and motivations, and rms must understand these differences in order to access the information and knowledge embedded in these structures.For example, users who are interested in developing additional product content (Koch and Bierbamer, 2016) can provide valuable need-related information about relevant product architectures.By understanding the drivers of UGC and user innovation, rms can also identify user needs that might be overlooked by traditional information gathering methods (Timoshenko and Hauser, 2019).These co-creative processes enable users to contribute nonredundant information to the product offering (Cui and Wu, 2017) if the rm is sensitised to detect and act on it (Balau et al., 2020).
These closely linked and mutually reinforcing UCS processes are sustained by effective rm practices for interfacing with user communities.In order to match innovation-conducive information with extant problems and so discover new opportunities, rms must know how to incentivise interactions between users, share ongoing development efforts, and invite feedback.If effectively utilised, UCS capability enhances a rm's understanding of its existing and potential user base and enables it to anticipate changing user preferences, needs, demands, and opportunities for product innovation.The hypothesised effects of deploying UCS capability for product innovation are summarised in Fig. 2 and are outlined in more detail below.

UCS and product innovation performance
In dynamic and fast-moving digital markets, the performance of individual products can contribute signi cantly to a rm's overall competitive performance (Teece, 2007), and it is therefore an important outcome variable for UCS capability.Previous research has reported a positive relationship between performance and openness in innovation (Oduro et al., 2021), and sensing capabilities have been linked to rm-level performance (e.g.Bharadwaj and Dong, 2014;Mu, 2015;Ngo et al., 2019;Olavarietta and Friedman, 2008;Wilden and Gudergan, 2015).However, few if any studies have explored the link between these capabilities and nancial performance, and to our knowledge, no study to date has examined this issue in relation to digital product innovation or user communities.While there are indications that leveraging user communities may increase overhead costs (Jepppesen, 2005), several studies point to positive performance effects (e.g.Arakji and Lang, 2007;Parmentier and Gandia, 2013;Parmentier and Mangematin, 2014), and we would therefore expect UCS processes to have a positive effect on nancial performance at the product level.On that basis, we formulated the following hypothesis.

2250007-11
H1: User community sensing is positively related to product innovation performance.

UCS and increased knowledge of product users
User communities possess innovation-conducive information about products, who uses them, and in what ways (Hau and Kim, 2011;Jeppesen, 2004).Sensing capabilities contribute to product innovation by tapping into this information (Mu, 2015).For that reason, more extensive and effective use of UCS processes during development is likely to augment a rm's knowledge about user needs, demand, and product use (referred to here as user knowledge).On that basis, we formulated the following hypothesis.

H2: User community sensing is positively related to the extent of a rm's user knowledge.
Based on existing evidence from several research streams, we also anticipate that more extensive user knowledge will result in better-performing products.Marketing capability research suggests that this type of user knowledge relates positively to performance (e.g., Day, 1994Day, , 2011;;Kohli and Jaworski, 1990;Narver and Slater, 1990;Slater and Narver, 2000).The central premise here is that the ability to identify and respond to market shifts will enhance performance by creating greater value for users.By anticipating trends and responding to emergent demand, a rm will be better able to develop higher-performing products (Day, 1990(Day, , 2011)), especially in dynamic markets (Teece, 2007).Additionally, product development studies have typically reported a positive link between the integration of market information and new product performance (Evanschitzky et al., 2012;Henard and Szymanski, 2001).In short, the greater a rm's user knowledge, the better its products can be expected to perform.On that basis, we formulated the following hypothesis.

H3:
The extent of a rm's user knowledge is positively related to product innovation performance.

UCS and product innovation speed
Ef ciency and speed are important factors in product development (Evanschitzky et al., 2012;Henard and Szymanski, 2001;Menon et al., 2002), but there is little empirical consensus of how the ability to leverage user communities affects 2250007-12 the development speed in product innovation.Instead, the reported effects of user involvement at different stages of the development process are mixed.During the early and late stages of development, user involvement can have a positive effect on speed, but the effects are less clear (and possibly detrimental) in the mid-stages (Chang and Taylor, 2016).While these studies have little to say about dedicated capabilities that provide useful and timely market information (Day, 1994;Olavarietta and Friedmann, 2008), it seems likely that some such capabilities contribute to the speed of innovation.There is some evidence that focusing on demands and needs can improve the development speed (Feng et al., 2012); for example, involving users in problem solving can help to accelerate software development (Mallapragada et al., 2012).We therefore anticipate that UCS capability will have a net positive effect on the product innovation speed, and on that basis, we formulated the following hypothesis.
H4: User community sensing is positively related to the product innovation speed.
Both innovation management and marketing researchers have noted the importance of the product development speed (Evanschitzky et al., 2012;Henard and Szymanski, 2001;Menon et al., 2002) in securing competitive advantage by being rst, or early, to the market (Kerin et al., 1992;Kessler and Chakrabarti, 1996).Conversely, an overextended development trajectory may hinder rm performance.As digital markets and technologies change rapidly, product life cycles shorten, and conditions may have changed by the time a product is launched (Bohlmann et al., 2013;Buganza and Verganti, 2006).A longer interval between ideation and launch may undermine the t between product offering and user needs while faster development is likely to mean better innovation performance.On that basis, we formulated the following hypothesis.
H5: Product innovation speed is positively related to product innovation performance.

Data collection and sample
To test the hypothesised relationships (see Fig. 2), this study draws on survey data from the Swedish video game industry.This empirical context was chosen for three main reasons.First, previous research has identi ed user communities as potent external sources of innovation-conducive knowledge and information in this context (Arakji and Lang, 2007;Jeppesen, 2004;Jeppesen and Molin, 2003;Parmentier and Gandia, 2013;Prügl and Schreier, 2006).Second, as in other sectors that depend on digital technologies, the video game industry incentivises the development of effective sensing capabilities.As competition is global, market conditions and technologies change rapidly (Li et al., 2010;Readman and Grantham, 2006).Product life cycles are therefore typically short, user preferences are dynamic, and product performance is often dif cult to predict-all conditions that reward effective marketing and innovation capabilities (Parmentier and Mangematin, 2014;Takata, 2016;Wilden and Gudergan, 2014).Third, con ning the study to a national context enabled us to conduct a census survey rather than sampling representative rms and served as a soft control for environmental and institutional variations.
In recent years, the Swedish video game industry has experienced signi cant growth and international success.Between 2017 and 2020, turnover grew by more than 260%, with revenue increasing by more than 175% during the same period (Nylander, 2021).At the time of data collection, about 370 rms were active in the sector; of these, 246 were deemed suitable for inclusion, as their main business activity was game development.This does not include the suppliers, consultancies, industry organisations, and investors that form part of the gaming industry ecosystem but were beyond the scope of this study.
The qualifying rms were contacted by email and/or phone and were invited to participate in the survey, targeting key informants in middle-management positions or higher.The main inclusion criterion was that respondents should have a good overall sense of their rm's product development processes.To improve face validity and response rates, the data were collected by means of structured interviews, either on-site or by video call.In total, 173 interviews were deemed suitable for inclusion in the nal analysis.

Measures
To operationalise the theoretical constructs, we adapted previously validated self-report items based on seven-point Likert scales (Table 2).
To develop a measure of UCS capability, we drew on previous sensing capability research (Day, 1994(Day, , 2020;;Teece, 2007) as a conceptual starting point.We adapted Mu's (2015) market sensing scale to user communities rather than markets and single development projects rather than rm-level processes.This latter adjustment was designed to ensure that we measured the deployment of UCS capability rather than its mere presence in a given organisation.The UCS capability construct was operationalised as scanning and identi cation of trends and changing conditions, responsiveness to interactions in user communities, and triangulation of community information with other sources.The user knowledge construct was grounded
2. User communities made us aware of changing market conditions.
3. The team was sensitised to listen to opportunities in user communities.
4. We anticipated trends in user communities before they became fully apparent.

User knowledge
Hau and Kim (2011); Jeppesen (2004); Parmentier and Gandia (2013) In our markets, we have good knowledge of the following.
1. Customer demand 2. Customer needs 3. How our products are used

Product innovation speed (development speed)
Rind eisch and Moorman (2001) In retrospect, our development outcomes can be described as follows.
1. Far behind our time goals/Far ahead of our time goals 2. Signi cantly slower than the industry norm/Signi cantly faster than the industry norm 3. Much slower than expected/Much faster than expected

Product innovation performance ( nancial performance)
Cheng and Huizingh (2014); Grif n and Page (1993); Salomo et al. (2007) Relative to the original objectives 1.The game is very successful in terms of pro tability.
2. The game is very successful in terms of sales volume.
3. The game is very successful in terms of sales pro tability (relative to existing titles).

Firm Age
Year of rst product launch

Project uncertainty
During development

2250007-15
conceptually in previous research identifying user communities as a potent source (e.g.Hau and Kim, 2011;Jeppesen, 2004;Parmentier and Gandia, 2013); these items measured knowledge of user demand, user needs, and product use in a rm's markets.
To measure the product innovation speed, we adapted Rind eisch and Moorman's (2001) time-to-market scale to evaluate development project outcomes relative to time goals, industry norms, and internal expectations.The use of multiple reference points helped to avoid any bias caused by unduly optimistic project planning.To measure product innovation performance, we adapted three items from scales measuring nancial product performance (Cheng and Huizingh, 2014;Grif n and Page, 1993;Salomo et al., 2007) to assess sales volume and pro tability relative to project goals, along with one item assessing pro tability when compared to competing products.Correlations between these constructs are detailed in Table 3.
We also included a number of control variables.To account for differences between rms in terms of maturity and organisation, we controlled for rm size (measured as a number of employees) and rm age (measured as years since founding).To take account of differences between product innovation projects, we included a two-item construct to capture perceived project uncertainty.Finally, to control for differences in the market environment, we included a market dynamism construct based on four items adapted from Cheng and Huizingh (2014).We also used a marker variable to test for common method bias.This construct comprised three items measuring improvisation in external business relationships (Moorman and Miner, 1998) and was based on the criteria for appropriate markers as theoretically unrelated and uncorrelated but comparable in terms of format and measurement (Simmering et al., 2014).

Data analysis
Using LISREL 10.3, we performed SEM to test the conceptual model in Fig. 2. SEM is an appropriate method of exploring relationships between multiple variables (Hair et al., 2014: 546).It can be understood as an extension of factor analysis and multiple regression, with some important differences.First, SEM facilitates the examination of dependence relationships within a single model, where one construct can act as both independent and dependent variables (Hair et al., 2014: 552).Unlike multiple regression, this enables simultaneous testing of complex sets of relationships between theoretical concepts of interest.SEM also assesses relationships between latent constructs; as in factor analysis, each latent construct is based on the consistency of several measured variables that together represent the construct.Using several indicators in this way serves to explicate theoretical concepts that cannot be measured directly.
In SEM, the use of latent constructs also improves the estimation of dependence relationships by taking account of measurement error (Hair et al., 2014: 547), so ensuring that coef cients will be closer to their true value and that the identi ed relationships will be more accurate (Hair et al., 2014: 549).In contrast, measurement error can cause regression techniques to understate the true relationships between variables, and this is to some extent unavoidable for a survey and self-reported data.The SEM analysis followed a two-step approach (Anderson and Gerbing, 1988), where the measurement model was assessed prior to estimation of the structural model, enabling us to assess convergent and discriminant validity in the rst step (Campbell and Fiske, 1959) and the theoretical model in the second step (Campbell, 1960;and Meehl, 1955).

Measurement model
For the measurement model in Table 4, all composite reliability measures exceeded 0.7, indicating acceptable construct reliability (Nunnally, 1978).Item reliability was further assessed by examining standardised loadings between latent constructs and individual items.All but two of the items retained in the measurement model loaded above the 0.50 cut-off and exceeded the recommended value of 0.70 (Hair et al., 2014); one indicator of user knowledge and one of product innovation speed loaded at 0.50 and 0.058, respectively.All measurement paths in the model were signi cant at p < 0.05.The average variance extracted (AVE) was calculated to assess the items' convergent validity, and all constructs met or exceeded the cutoff value of 0.50 (Fornell and Larcker, 1981;Hair et al., 2014).User knowledge returned the lowest AVE; removing one item would raise the AVE for this construct above 0.60, but as this would undermine construct validity, the item was retained.This follows the rule of thumb of at least three items per latent construct (Hair et al., 2014).The measurement model, which included 13 items and four latent constructs, yielded good t (χ 2  (94) 103.16 (p = 0.24); root mean square error of approximation (RMSEA) = 0.058; comparative t index (CFI) = 0.99).Overall, these results indicate testing of the structural model to be appropriate.
Harman one-factor testing for common method bias (Podsakoff and Organ, 1986) showed that this was not a concern, as one factor explained 29.7% of the variance while the cumulative variance explained by four factors was 72.3%.Additionally, a marker variable (Podsakoff et al., 2012) (CR = 0.88, AVE = 0.71) was also used to detect any effects of common method bias in the data.As a rst step, changes in the AVE of focal constructs were assessed when specifying paths 2250007-18 between the marker variable and indicators of the other constructs.The change in AVE (less than 0.1) was not signi cant, falling below the common threshold of 0.3.A zero-constrained test was then performed to determine whether any potential bias had an impact signi cantly different from 0. A measurement model with all paths between the marker variable and indicators constrained to 0 was compared to an unconstrained model with the same relationships.χ 2 change of 13.09 with 13 degrees of freedom was lower than the critical χ 2 value of 22.36 with 13 degrees of freedom, indicating that common method bias had no signi cant impact on variance in the measurement model.The marker variable was retained as a control when testing the structural model.

Structural model
In the structural model (see Fig. 3), the t indices were χ 2 (97) 106.55 (p = 0.24); RMSEA = 0.058; and CFI = 0.99.Our rst hypothesis (H1) predicted a positive relationship between UCS and product innovation performance.Contrary to our expectation, however, we found a signi cant negative relation between UCS and product innovation performance (Γ = −0.20 t = −2.06).There was support for H2, which predicted a positive effect of UCS on user knowledge (Γ = 0.46; t = 5.03).H3 anticipated a positive effect of user knowledge on product innovation performance, and the model supported this hypothesis as well (β = 0.57; t = 5.68).H4 predicted a positive path between UCS and the product innovation speed, but this was not supported, as the path was not signi cant (Γ = 0.08; t = 0.88).Finally, H5 posited a positive relationship between the product innovation speed and product innovation performance.In line with our expectations, this effect was found to be positive and signi cant (Γ = 0.22; t = 2.63).

2250007-19
Taken together, these results from the structural model indicate good t with the data, providing overall support for the theoretical argument but not for all of the posited relationships.This suggests that UCS has a relatively weak direct negative effect on product innovation performance but exerts an indirect positive effect through increased user knowledge.This inconsistent partial mediation (MacKinnon et al., 2007) implies that UCS capability has a positive effect on performance only at higher levels of user knowledge.To seek support for this partial mediation effect, we followed the established recommendations of Baron and Kenny (1986).A Sobel test to assess the signi cance of this effect yielded a signi cant Z-value of 3.64 for p < 0.01 (two-tailed), con rming partial mediation.The observed positive relationship between the product innovation speed and product innovation performance aligns with previous ndings.The effects of the control variables were assessed by applying a jack-knife procedure (Andersson et al., 2002) to the structural model.First, we identi ed the quartiles for each control variable to identify subsamples (such as high or low levels of market dynamism).Each subsample was then removed one at a time, re-running the model to identify any changes in structural path loadings.This procedure was repeated for all subsamples for each of the four control variables size, age, market dynamism, and project uncertainty (16 models in total).This assessment of potential control variable effects on the results from the structural model yielded no signi cant changes to the paths of the model at the 0.05 level.

Discussion
Quantitative empirical evidence of structured utilisation of user communities remains limited, both in the video game sector (Koch and Bierbamer, 2016) and other settings, especially in relation to performance outcomes (Bogers et al., 2016;Randhawa et al., 2016).In this study, we hypothesised that UCS capability would impact positively product innovation performance, rm user knowledge, and product innovation speed.Our results indicate that while UCS capability impacts positively the performance, the effect is indirect, helping rms to develop better-performing products by providing more accurate and up-to-date information about user needs and demand.As further con rmation of this nomological structure, the direct relationship between UCS and product innovation performance was negative while the indirect path indicated a positive effect.
This partial mediation relationship is what MacKinnon et al. (2007) characterised as inconsistent mediation, where the signs of the mediated and direct effects differ.The likely explanation is that UCS processes have a positive impact on innovation performance if they deliver more information about users but may otherwise 2250007-20 have a detrimental effect.Indeed, previous evidence suggests that involving users in product innovation is not always bene cial, as rms differ in their ability to effectively leverage these inputs (Arakji and Lang, 2007;Burger-Helmchen and Cohendet, 2011;Koch and Bierbamer, 2016;Parmentier and Gandia, 2013).This form of user engagement may also increase development costs (Jeppesen, 2005;Tranekjer and Søndergaard 2013), and as sensing is a company-wide and potentially resource-intensive endeavor (Day, 1994), any negative impact on performance may be attributable to how increased costs reduce the speed and/or the time and energy available for development work.
These negative effects may be offset if the rm acquires and internalises valuable knowledge from user communities.Notably, the nonsigni cant relationship between UCS and the product innovation speed suggests that deploying UCS for product innovation does not signi cantly affect a rm's ability to bring products to market in a timely fashion.This in turn implies that sensing processes have no net impact on the planned completion of the product development process.

Theoretical implications
As digital technologies and pervasive digitalisation drive more open (Chesbrough and Bogers, 2014) and decoupled (Ehls et al., 2020) innovation, the role of users and user communities as a source of use-and need-related knowledge becomes increasingly important (von Hippel and Kaulartz, 2020).In dynamic and technology-intensive sectors like the video game industry, improving awareness and utilisation of user communities can enhance the nancial performance of products.However, theoretical understanding of the capabilities needed to leverage distributed innovation remains limited (Chang and Taylor, 2016;Nambisan et al., 2017;Roberts et al., 2021), as most previous studies have focused on ad hoc approaches or the user perspective (Ehls et al., 2020).
This study contributes to this area through the concept of UCS capability.Building on existing marketing and strategy research on sensing capabilities, this study introduces and de nes UCS capability as a rm's ability to anticipate market change and identify emerging opportunities for innovation by interfacing with user communities.We argue that rather than prioritising either structured external search (Teece, 2007) or constant monitoring of weak market signals (Day, 2011), UCS capabilities are oriented both inside-out and outside-in (Day, 2020).By maintaining user community proximity and peripheral vision, a rm can interact natively with users, actively and passively scanning the user context for codi ed information and knowledge.These processes of interpretation and opportunity identi cation enable the sensing rm to understand and exploit user knowledge as a resource for product innovation and enhanced performance.

2250007-21
This study also responds to a recognised need for a fuller understanding of the capabilities required to manage distributed innovation in user community settings (Bogers et al., 2016;Chesbrough et al., 2018) and a stronger grounding in established marketing and capability theory (Randhawa et al., 2016).To that end, we explored the management of user communities for distributed innovation from a capability perspective (Bogers et al., 2016).The study also contributes to theory development in the area of distributed digital innovation (Lyytinen et al., 2016;Nambisan et al., 2017) by clarifying how rms can manage interactions centered on a digital artifact (Becker et al., 2021).Video games and virtual user communities (and their interactions with rms) are intrinsically linked to and enabled by digital technologies.In the empirical study, each video game under development, and its underlying software, constitutes digital artifacts around which interaction with user communities is based.Digital artifacts impact the rm's knowledge management and innovation capabilities (Lyytinen et al., 2016), and for UCS, provide the basis for interfacing with user communities.As a means of interfacing with user communities, UCS enables rms to anticipate changing needs and emerging opportunities for innovation.By identifying the processes that underpin this capability and assessing its impact on product innovation projects, the present study clari es how rms can manage the "interacted actor" (Ramaswamy and Ozcan, 2020) within the co-creative process.
Finally, the paper clari es how structured utilisation of user communities affects product innovation performance outcomes-an area where quantitative evidence is scarce (Bogers et al., 2016;Koch and Bierbamer, 2016;Randhawa et al., 2016).The potential of user communities as external information sources has been reported in contexts that include software (Flowers, 2008;Jeppesen and Laursen, 2009); open-source software (Dahlander andMagnusson, 2005, von Krogh et al., 2003); crowdsourcing (Ebner et al., 2009;Piller and Walcher, 2006); and sporting and hobbyist goods (Baldwin et al., 2006;Lüthje et al., 2005).However, these studies typically utilise qualitative methods (Koch and Bierbamer, 2016) and emphasise ad hoc approaches and the user perspective (Ehls et al., 2020).The present SEM analysis complements these earlier works by advancing a quantitative account of how UCS mechanisms impact performance.

Managerial implications
This study also has some practical implications for developing or re ning UCS processes.Building on previous marketing and innovation research, our ndings highlight the need for sensitisation to user communities if rms seek to leverage external information for product innovation.By developing their UCS capability, rms can interface more effectively with user communities to identify emergent market changes and opportunities for product innovation.

2250007-22
Crucially, it is not enough to monitor user communities or collect data only when problems or questions arise.Effective sensing depends on a company-wide capacity to scan the external environment (Day, 1994).As some managers may be skeptical about the utility of user involvement in innovation processes (Bradonjic et al., 2019), it is important to develop and maintain user proximity by introducing a range of opportunities for participation in the development process.This user integration can be facilitated by identifying subgroups or individuals with leaduser characteristics and by open or semiopen testing.These activities can serve as channels for information in ow through points of interaction between community and rm.Users can also be encouraged to contribute, co-create, and innovate partly nished or completed products by creating a dedicated platform, again bringing users closer to the development process.
In order to detect and act on weak signals and potential innovation opportunities, the rm must also hone its ability to interpret and understand codi ed and context-dependent information from within the user community (Day, 2020;Day and Schoemaker, 2016).Here again, proximity-and, ideally, multiple points of interaction-can help to build rm awareness, alertness, and openness to the user community over time.As well as outsourcing problem-solving, UCS involves interfacing with user communities to capture weak signals and innovation opportunities through co-creation with selected users, online discussion forums, or user-generated content and innovations.To assess the most promising sources of relevant information and knowledge, the sensing rm must maintain a presence in the user context, based on effective information channels, empathy with users, and a willingness to act on partial information.
We envisage that UCS capabilities can usefully be deployed in a number of settings.For example, the platform economy heightens the need for more effective use of external knowledge sources like user communities (Nambisan et al., 2019).For multisided platforms like steam.com(computer gaming) or amazon.com,there is signi cant potential to gather information about diverse and global users, enabling these rms to introduce their offerings almost instantly to international markets.Conversely, rms can leverage knowledge about user needs and demand (Nambisan et al., 2019) to overcome knowledge barriers to product innovation, especially in the case of SMEs (Shaheer and Li, 2020).
This study focused on the video gaming industry as a highly digitalised and technology-intensive setting where UCS capability is paramount.As product offerings become fully or partly digital, they acquire the malleable characteristics of digital artifacts (Zittrain, 2006), enabling ongoing post-launch product development (Tschang, 2007).In response to emergent needs and innovation opportunities, the software can be updated, new operating systems installed, and new functions added to products already in use.This exibility of digital offerings and products

2250007-23
throughout their life cycle (Buganza and Verganti, 2006) contrasts sharply with the discrete phases of traditional product innovation (Vargo et al., 2020), offering great potential for effective UCS deployment.In the video game industry, for example, post-launch development of additions and updates ensures product longevity and performance (Jeppesen, 2004) far beyond the original release (Parmentier and Mangematin, 2014).This is true of most software products and, increasingly, of physical products with digital and connected components.

Study limitations and future research
This study and its limitations highlight a number of avenues for future research.First, while our empirical results indicate that UCS has no signi cant effect on the product innovation speed, the chosen measurement instrument captured only the effects of UCS on the overall development speed rather than on the different parts of the process.Given the mixed evidence regarding the impact of user involvement and co-creation on speci c phases of the innovation process (Chang and Taylor, 2016), future studies should investigate the effects of UCS in this regard.While sensing capabilities can enhance both exploration and exploitation (Day, 2020), UCS may prove more useful in the early or late stages of digital innovation, and process frameworks such as the stage-gate model (Cooper, 2009) or its more recent incarnations incorporating decoupled and open innovation (Grönlund et al., 2010) could be used to investigate this issue.As a corollary, it seems worthwhile to investigate the role of UCS processes in post-launch development, as few studies to date have explored the structured and institutionalised processes used in the video game industry and other digital sectors to interface with user communities.
Second, while the novel concept of UCS capability elaborated here is grounded in past research, more work is needed to re ne this concept.To that end, future research should continue to explore the effects of UCS and its relationship to other relevant concepts.More research is also needed to develop a ner-grained understanding of this capability that complements our quantitative approach by elucidating the mechanisms, practices, and processes that underpin UCS.In particular, it is important to explore the observed inconsistent partial mediation of product innovation performance by UCS.Researchers should also investigate the factors and conditions that make UCS processes more or less likely to generate innovation-conducive information, including rm-related variables and user community characteristics.
Third, our focus on the video game industry may limit the generalisability of these ndings to other industries and sectors; for example, existing evidence suggests that the effects of marketing capabilities may be industry-speci c (Foley and Fahy, 2009).Finally, as this study relies on cross-sectional data, inferences about 2250007-24 causal effects and the direction of observed relationships, although theoretically anchored, should be substantiated by longitudinal data.In addition, the possibility of respondent bias cannot be completely ruled out in light of the subjective measures used to assess performance outcomes and effects.For that reason, future studies should utilise objective performance data to complement and corroborate these ndings.

5 Table 1 .
Key empirical and conceptual studies of sensing capabilities.