Problem resolution with business analytics: a task-technology fit perspective

2.1 Task-technology fit

Technology has specific characteristics that make it suitable for addressing business problems. At the same time, users who employ a technology and the task through which the technology is applied also possess unique characteristics. The relationship between the characteristics of the technology, task and user is termed TTF (Chen et al., 2021b; Goodhue and Thompson, 1995; Howard and Rose, 2019; Junglas et al., 2008). TTF can be used to evaluate the efficacy of attaining task outcomes. When a technology has too many or too few features, this may undermine the attainment of task outcomes (Howard and Rose, 2019). It is only when there is a perfect fit between technology features and task characteristics that users obtain the best outcomes from using the technology in the task (Junglas et al., 2008). Therefore, the utilization of a poorly designed technology may not lead to the best outcomes (Goodhue and Thompson, 1995). Here, poorly designed technology refers to having either too many or too few features than necessary to accomplish a task effectively. For instance, BA can lead to an excessive number of insights through visualizations, which are a feature of BA (Saggi and Jain, 2018).

Technology is often designed with specific features, irrespective of how those features can be applied to particular tasks (Gebauer et al., 2010). Typically, technology is designed to accommodate a range of tasks rather than being tailored to a single task (Sein et al., 2011). This design approach results in technology being equipped with numerous features, some of which may be irrelevant to certain tasks. As a result, when technology is applied to various tasks, it can exhibit an excess, shortage or the ideal number of features for each task. This is the essence of TTF. These scenarios can also arise when technology is employed for tasks it was not originally designed for (Gregor and Hevner, 2013).

When technology includes more features than necessary for a specific task, users may choose to utilize the relevant features for that task (Fuller and Dennis, 2009). Previous research has thus contended that, to a certain extent, an excess of features may not detrimentally impact outcomes (Yang et al., 2013). This implies that an excessive number of features in technology only starts to negatively affect the achievement of outcomes once a particular threshold is surpassed. Earlier examinations of TTF have primarily concentrated on the use of IT systems for tasks such as scheduling and document approval (Yang et al., 2013), as well as using the Internet to locate specific information on webpages (Aljukhadar et al., 2014). In contrast, this study focuses on BA as a technology employed to address a variety of business problems. Since BA only brings value when its use is attached to tasks (Kitchens et al., 2018), TTF is well positioned to provide an understanding of the business value of BA and how the value can be achieved through tasks (Furneaux, 2012).

2.2 Business analytics and problem resolution

Firms have realized the potential of BA in different applications. BA refers to “the techniques utilized in order to examine, process, discover and expose hidden underlying patterns, interesting relations and other insights concerning the application context under investigation” (Iqbal et al., 2020, p. 1). What insights are considered relevant and how quickly they must be generated is determined by the usage context (Conboy et al., 2020). Nonetheless, the success of organizations is determined by BA’s ability to analyse data intelligently (Shiau et al., 2023). Among the plethora of BA features is the ability to make predictions and prescriptions (Lepenioti et al., 2020). Other features relate to specific types of BA such as information extraction, text summarization and sentiment analysis within text analytics (Gandomi and Haider, 2015). Additionally, certain features are associated with how BA is applied such as decision modelling, decision-making and decision execution (Saggi and Jain, 2018). The application contexts of BA include manufacturing processes (Popovic et al., 2018), supply chain management (Chen et al., 2015, 2021a, b), customer preference discovery (Kitchens et al., 2018; Zhou et al., 2018) and healthcare (Wang et al., 2018). In each of these contexts, the application of BA may entail starting with what and moving toward how or starting with how and then defining what (Fernandez-Vidal et al., 2022). In the former case, an organization identifies inefficiencies or opportunities (what) and then applies BA (how) in addressing them. In the latter case, an organization identifies and understands the capabilities of BA (how) and seeks to identify which aspects of the organization can be improved through those capabilities (what).

The resolution of a problem or capturing of an opportunity, through tasks, represents a benefit to an organization that we categorize as business value. Thus, the extent to which a user takes advantage of the fit between the task and BA determines how much business value is realized. Notwithstanding, tasks can be either well-defined or unclear (Avital and Te’eni, 2009). Well-defined tasks have low ambiguity and are finite. Unclear tasks have high ambiguity and may not have an obvious end. Given the desirability of ideal fit (Junglas et al., 2008), we argue that when the task to be addressed with BA is not clear, achieving an ideal fit in such a scenario can be difficult (Avital and Te’eni, 2009). The fit between BA and a task can only be estimated accurately when the features of BA and the characteristics of the task are clear (Mathieson and Keil, 1998). Therefore, while the material features of BA, such as the capability to conduct predictive analytics are present, there is a requirement for clarity regarding the specific task and its objectives to which BA can be applied (Lehrer et al., 2018). Without this clarity, it becomes challenging to effectively fit BA's known features with the task at hand. Arguably, the ability of BA to fit a task remains ambiguous if the task itself is not clear. Therefore, the greater the task's ambiguity, the more challenging it becomes to achieve an ideal fit between BA and the task. In other words, we propose that:

Ideal fit is associated with the best possible outcomes (Junglas et al., 2008). In this case, BA provides neither more nor fewer features than are required to adequately perform the task (Howard and Rose, 2019). There is a perfect match between the features of BA and the requirements of the task. Indicatively, the ideal fit will result in the best possible problem resolution. Expressed differently, when BA matches the task at hand, the best possible business value will be realized. For instance, BA use has been associated with a reduction in waste in manufacturing processes (Popovic et al., 2018). When the task is to optimize the manufacturing processes, ideal fit means BA manages to identify ways to achieve this in a way that fulfils all criteria that must be met. In other words, meeting all optimization criteria with BA signifies successful task resolution. Therefore, we assert that:

When there is an under-fit, technology does not solve the problem in an ideal manner. Notably, when there are too few features in the technology, it may be hard to perform essential functions of the task (Howard and Rose, 2019). In other words, the features of BA that should coincide or intersect with those of the task are missing or not available in the right amounts (Smith and Mentzer, 2010; Zigurs and Khazanchi, 2008). A salient aspect of realizing business value with BA is to first identify tasks that can be carried out (Kitchens et al., 2018). This means the task precedes the application of BA. Under-fit occurs when it is discovered that BA cannot meet all the requirements of the task. One possible reason for this under-fit is the technical limitations of BA algorithms that may simply not be capable of meeting the complexity of task requirements. Thus, under-fit may result in failed or partial success in task resolution. Therefore, we propose that:

When technology offers an excessive number of features for a task, creating an over-fit, it can detrimentally impact the achievement of outcomes, indicating a failure to address the task adequately (Howard and Hair, 2023; Howard and Rose, 2019). There is a decrease in task productivity due to the excess number of features in the technology (Yang et al., 2013). Hence, we contend that over-fit between BA and the task negatively influences task resolution. BA can be used in the task of improving a product feature based on analysing large volumes of online customer reviews (Zhou et al., 2018). However, BA often generates not only heterogeneous but also conflicting insights that far exceed the requirements for improving the product feature. This makes it difficult to complete the task. Specifically, not all insights can be incorporated into the product feature development. Any attempts to do so may be detrimental. Succinctly, we propose that:

The propositions advanced are summarized in Table 1.

To further illustrate these propositions, we focus on how BA helps in addressing tasks. The features of BA constituting the analytics goal include objectives for description, prediction and prescription (Nalchigar and Yu, 2020). The task’s features include the question goal, referring to what the business wants to achieve by applying BA. When the question to be addressed is not clear, this makes the task ambiguous thus making it difficult to achieve the fit between the analytics goal and question goal (P1). When there is a fit between BA and the task, it means the analytics goal provides the right insights to address the question goal. In other words, the insights generated from BA are sufficient to address the problem (P2). Under-fit can occur when the question goal demands specific insights from BA that cannot be availed through the analytics goal. As such, this hurts the resolution of the task (P3). Even though the task’s question goal may demand the description goal (i.e. descriptive insights), the analytics goal may also include prediction and prescription goals that are not required for the task. The inclusion of these other goals in the insights generated from BA means there is more information provided than is required for the question goal. Consequently, the greater the amount of information provided beyond the task’s requirements, the more difficult it can be to address the task (P4).

3.1 Pattern matching

This study aims to develop theory using pattern matching, which entails comparing predicted theoretical patterns with observed empirical patterns (Sinkovics, 2018). The theoretical patterns are derived through explanatory theory construction, involving the formulation of preliminary explanations for what might be observed empirically. These preliminary explanations can be articulated as propositions or hypotheses. In the context of this study, we propose the preliminary explanation that the fit between BA and tasks influences task resolution, subsequently impacting the business value derived from BA. As part of constructing the explanatory theory in the theoretical realm of pattern matching, we present four propositions (refer to Table 1).

The observation realm of pattern matching consists of data construction theory and observation theory. Data construction theory describes the relevant raw data and how it should be collected. This data construction theory is highlighted in the subsequent section on data collection. Observation theory includes the construction of a framework that details how convincing inferences can be advanced from the data. The data analysis section, highlighting the coding of data, corresponds to observation theory. An illustration of the pattern matching process is provided in Figure 1.

We chose to employ pattern matching for two primary reasons. Firstly, it is well-suited for exploratory research, and secondly, the patterns presented in explanatory theory construction (i.e. propositions) serve as guiding principles for exploration and may evolve during the study (Sinkovics, 2018). The use of pattern matching enables this study to benefit partially from deductive and inductive approaches without being affected by the challenges each approach entails. Deduction involves deriving observations based on established rules and explanations as premises, while induction moves from specific observations and explanations to infer general rules (Mantere and Ketokivi, 2013). The inductive part of the observation realm, as depicted in Figure 1, benefits from “the ability of qualitative data to offer insight into complex social processes that quantitative data cannot easily reveal” (Eisenhardt and Graebner, 2007, p. 26). The theoretical realm, as shown in Figure 1, includes aspects of deduction in the formulation of propositions. Even though premises in deduction can be empirically falsified, they are not revised. This is not the case with pattern matching in which the initial arguments can be revised (Sinkovics, 2018). The reconciliation of theoretical patterns (i.e. propositions) and observed patterns leads to a theory of the fit between BA and tasks.

We chose to use semi-structured interviews for collecting data within the observation realm of pattern matching because the utilization of BA varies considerably across different domains and business contexts (Grover et al., 2018). Consequently, semi-structured interviews offer the opportunity to delve into the intricate relationship between business problems and BA, and how their fit influences the resolution of those business problems. Essentially, this study aims to gain insight into how the fit between tasks and BA manifests in various situations. Of particular interest are the diverse business problems that BA addresses, which can differ significantly between individuals and organizations.

Since data collection in the observation realm is informed by the theoretical realm, the second rationale for using semi-structured interviews comes from the TTF theory, which is fundamentally based on the effective usage of BA by users to accomplish their objectives (Burton-Jones and Grange, 2013). In the TTF literature, the assessment of the fit between a task and technology has typically been evaluated from the user's perspective (Howard and Hair, 2023; Howard and Rose, 2019; Junglas et al., 2008). Consequently, a qualitative research methodology employing semi-structured interviews, highlighting users’ perspectives of BA within specific contexts, was deemed apt for assessing the fit between BA and tasks.

3.2 Data collection

We collected data through semi-structured interviews. Interviews are regarded as one way to generate data that relies on the interviewees’ valid knowledge and their ability to express such knowledge appropriately (Goldkuhl, 2019). We employed purposeful sampling, involving the use of pre-determined criteria, to select interviewees (Matavire and Brown, 2013). These criteria included the appropriateness of the interviewees' work roles and the extent of BA application within their respective companies. Specifically, we sought interviewees with roles such as data scientists who actively applied BA to address business problems, acknowledging that work roles may vary across organizations. We conveyed the study's nature in the consent forms provided to interview partners and, in the initial stages of the interviews, gathered information about their work roles and the business problems they addressed with BA. These steps facilitated the assessment of the interviewee's suitability for the study.

In addition to IT service providers, interviews were also conducted with representatives from other business sectors. We evaluated the extent of BA application by referencing information available on the companies' websites, which detailed the nature of problems addressed with BA. For IT service providers, we were particularly interested in their provision of BA solutions across a range of industries, contributing to the diversity and depth of this study. Stated differently, these IT service providers were selected because they were using BA to address different business problems in diverse industries like manufacturing, mining and retail. BA solutions encompass the generation of insights through data analysis and their application to tackle specific problems (Nalchigar and Yu, 2020). A guideline for the interview questions is included in Appendix. The interviews, conducted in English, took place from March to June 2023, with an average duration of approximately 50 min. The interview participants were all based in Finland. This made engagement easier with the authors who are also based in the same country. Consent forms were signed by the interviewees, granting permission to record the interviews, which were then transcribed using Microsoft Teams. Table 2 below provides further details regarding the interviews. It is important to note that this study considers the time interviewees had spent in their present roles within their current companies, regardless of their overall professional experience.

3.3 Data analysis process

The data analysis relates to the observation realm, as illustrated in Figure 1. We were guided in conducting the data analysis by the analytical move of asking questions (Grodal et al., 2021). The question we kept in mind in the analysis of each interviewee’s responses is: “How is BA being applied to address problems in this context?”. The analysis of the data yielded first-order categories. Second-order categories were generated using the data analysis move of merging categories (Grodal et al., 2021) and ensuring category viability (Lo et al., 2020). Category viability is achieved by grouping similar entities and differentiating those that are dissimilar. The following questions, following the analytical move of asking questions, further helped to generate second-order categories from the first-order categories. Is BA addressing the business problem successfully? How is the nature of the problem affecting how it is addressed with BA? These questions broadened the analytical scope, providing room for the identification of new emerging categories that extended beyond the initial categorization. Subsequently, the authors engaged in discussions and reached a consensus on framing the emerging themes derived from the data analysis.

4.1 Coding process and selected interview quotes

The coding process was sequential, progressing from first-order themes to second-order themes and finally the core themes. This is shown in Figure 2. The first-order themes are descriptive and emerged directly from the analysis of interview quotes.

The second-order themes indicate a merging of first-order themes into superordinate themes (Grodal et al., 2021). Further, the core themes are an aggregation of the second-order themes. For the second-order themes, the number in brackets (e.g. 6/12) is the number of interviewees reflected in this theme as a fraction of the total interviewees. In Table 3, we show the supporting first-order interview quotes for each specific theme. The BA context column indicates to what purpose BA was being applied.

4.2 Description of four core themes

4.3 Matching theoretical and observed patterns

In this section, we discuss the match between theoretical patterns and observed patterns. This is shown in Table 4. The term “no pattern” means either the theoretical or observed pattern did not have a corresponding pattern. Pattern correspondence reflects the match between patterns. For each theoretical pattern, a corresponding observed pattern that aligns with it was found. The observed pattern of uncertainty in problem resolution suggests that the lack of problem clarity can make it hard to match specific problems and BA. The ability of BA to solve problems improves when the problems are clear. Thus, proposition P1 is supported.

BA is an adequate tool for solving problems because it is often developed according to the nature of the problem. Empirical observations reveal that BA often offers more features than necessary for problem resolution (refer to Figure 2). In these instances, the application of BA demonstrated a positive track record in successfully resolving problems, thus reinforcing the support for proposition P2. Nevertheless, given the contextual nature of BA use, some interviewees also noted instances where BA falls short in addressing business problems within their specific domains. More precisely, interviewees acknowledged that not all problems have perfect, foolproof solutions and therefore, the application of BA may not always result in a comprehensive problem resolution. Consequently, these observations provide further support for proposition P3.

While interviewees acknowledged the presence of more features in BA than is strictly necessary to address problems, this stance does not support proposition P4. Notably, interviewees did not emphasize any adverse consequences associated with over-fit. On the contrary, they perceived over-fit as affording flexibility in problem-solving, allowing for diverse approaches or the resolution of different problems. Essentially, over-fit results in the thorough resolution of problems, ensuring that all the necessary features for problem-solving are present in BA, even if some extra features remain unused (Soda and Furlotti, 2017). Despite BA being a tool designed for problem-solving, the observations highlight instances where problems remained unresolved. The concept of “fit” is inherently linked to technology utilization (Goodhue and Thompson, 1995). However, these observations suggest that the utilization (or non-utilization) of BA may be influenced by factors extending beyond fit.

5.1 Theoretical contributions

This study presents three notable contributions. Firstly, this study enriches the literature on the realization of business value using BA (Dong and Yang, 2020; Kitchens et al., 2018; Nalchigar and Yu, 2020). This prior literature focuses primarily on how BA can address business problems without paying attention to the extent to which BA is suitable for addressing those business problems. In other words, there is an implicit assumption within the literature that BA is suitable for addressing business problems. Using the theoretical lens of TTF, this study suggests fit as a dimension that should be considered in the application of BA. Specifically, this study suggests that BA can also experience under-fit when applied to task resolution. Consequently, TTF offers a valuable approach for assessing the effectiveness of BA application, highlighting that the realization of business value from BA does not necessarily imply a perfect fit for the task.

Secondly, this study makes a valuable contribution to the existing literature on TTF. While TTF is commonly perceived as a static point or a state of equilibrium, this study sheds light on the potential dynamics in the state of fit (Howard and Rose, 2019; Rai and Selnes, 2019; Yang et al., 2013). The empirical findings from this study reveal that the understanding of tasks tends to improve as they are resolved, indicating a progressive enhancement in the fit between BA and tasks, including a reduction in the degree of under-fit, as tasks are addressed. This suggests that it may not be a prerequisite for a business problem to be initially well-defined before applying BA. The requirement of problem clarity has been emphasized as crucial for attaining business value from BA (Kitchens et al., 2018). Previous research on how TTF evolves has primarily focused on how technology's appropriation improves for recurring tasks (Fuller and Dennis, 2009). Consequently, this study introduces an alternative perspective, highlighting how the task itself evolves, resulting in changes in fit over time, even though the technology remains unchanged.

Prior research has pointed out that BA is malleable (Benbya et al., 2020; Yoo et al., 2012). Saliently, this would indicate that the state of under-fit should not be persistent. In essence, changes to BA should lead to the addition or alteration of features such that under-fit is eliminated. However, this study notes that under-fit can be persistent because users tend to accept that technology may not meet the entire demands of a task. Interpreted otherwise, ideal fit is not always attainable.

Thirdly, qualitative research can be guided by the principle of theoretical engagement, which suggests that the researcher should “offer theoretical abstractions resulting from the analysis/interpretation of the data” (Sarker et al., 2013, p. xiii). Grounding on the empirical data, we offer the theoretical abstraction that the existence of over-fit between BA and tasks can be difficult to observe. This can be attributed to users of BA appropriating only those aspects of the technology that are relevant to the task. In other words, features of BA that are extraneous to task requirements may not necessarily be considered a liability. Contrary to indications from prior literature that over-fit may be detrimental to the resolution of business problems (Junglas et al., 2008), the observations made in this study suggest that over-fit may not hurt problem resolution. Rather, over-fit was perceived as increasing the variety of problems that can be addressed by BA. In the case of IT consulting companies, the existence of more features in BA than those required to solve problems meant the same BA solution could be applied to different problems.

5.2 Implications for practice

This study has four key practical implications. Firstly, solving business problems through BA is often an iterative process, where the understanding of the problem improves as it is addressed. This implies that the initial stages of implementing a BA solution offer a learning opportunity. Consequently, these early stages might exhibit characteristics of under-fit. This initial under-fit can serve as a tool for assessing whether the current solution should be continued or if an entirely new one is needed. By emphasizing these aspects, we aim to raise awareness among managers that the problems suitable for BA application may not always be clear from the outset. The lack of problem clarity should not be seen as a deterrent. Moreover, BA can be adapted to address problems differently, allowing for flexibility in problem-solving approaches.

Secondly, the study highlights some factors that contribute to the lack of task clarity. For instance, the interviewees indicated that there is often no baseline to measure the success of problem resolution, and the expected outcome from solving the problem is often unclear. They also highlighted that problems typically lack pre-defined technological solutions, with BA being perceived as merely one potential technological solution. Addressing these concerns about problem clarity can enable managers to enhance the fit between BA and tasks, thereby improving the effectiveness of problem-solving with BA.

Thirdly, the extent to which organizations promote BA is a crucial factor in determining its effectiveness in problem-solving. Key users play a pivotal role in fostering the use of BA. Managers should recognize that a distinction may exist between those who create BA solutions and those who apply these solutions to tackle problems. Therefore, the effective use of BA is contingent upon users being aware of its potential as a solution. Furthermore, it is imperative to avoid addressing symptoms rather than the underlying issues. What is perceived as a problem may just be a symptom of the actual problem. Additionally, business problems evolve, meaning that a BA solution which was previously an ideal fit for a task may become an under-fit.

Finally, a deep understanding of TTF can empower managers to enhance the magnitude of business value derived from BA. This study offers valuable insights in this context. The pursuit of improved business value necessitates the mitigation of both under-fit and over-fit scenarios. Without consideration of TTF, managers may remain unaware of whether they are maximizing the potential business value from their BA applications. Suboptimal applications of BA can still yield positive outcomes. As TTF encompasses both the task and technology dimensions (Jeyaraj, 2022), its comprehension urges managers to address both aspects, rather than focusing solely on one at the expense of the other. To be more precise, enhancing business value may entail the elimination of inefficiencies within both BA and the associated tasks.

5.3 Limitations and future research directions

This study has some limitations. First, purposeful sampling for interviewees creates an obvious limitation that needs to be recognized. However, this explorative study benefited from well-informed interviewees, and we reached saturation with this number of interviews. In addition, the emergent core themes are sufficiently broad and cover an extensive range of BA applications. However, the interviewees are all based in Finland. This may have limited the variety of perspectives on how BA can be applied.

Second, we focused on applications of BA in different organizations. The context-specific nature of BA (Grover et al., 2018) means the match between BA and tasks may not be replicable across different applications. However, the core themes were generated from the merging of specific categories and may be useful in explicating the overall nature of the match between BA and tasks. Further, the core themes demonstrate category viability through distinctiveness among categories (Lo et al., 2020).

While we have not found evidence of the negative impact of over-fit on task outcomes, previous studies have indicated its existence (Junglas et al., 2008). We suggest that future empirical research tackle the specific contexts under which the application of BA to tasks does not generate positive outcomes. While business value is often perceived as a positive value, the focus on over-fit may highlight novel insights on the possibility of negative business value. This study has highlighted the idea of unaddressed problems as a core theme. One reason why BA may not be applied is the costs associated with the execution of specific tasks. Another reason may be that tasks that are not time critical may be ignored. Despite the existence of a potentially ideal fit, future research could tackle the question: When is the application of BA to solve problems not desirable?