Can decision intelligence help organizations retain employees? Serial multiple mediation of job characteristics and meaningful work

Abstract

New artificial intelligence (AI) powered technologies such as OpenAI’s ChatGPT model, intelligent decision support systems, and autonomous robots are transforming decision making leading to the increased prevalence of decision intelligence in organizations. This paper explores the relationship between decision intelligence, job characteristics, meaningful work, and employees’ intentions to leave the organization or turnover intentions. The research model is based on robust theoretical foundations and was tested with data collected from a survey on Prolific. The study utilizes PLS SEM (partial least squares structural equation modeling) method to test the hypotheses. Three categories of model fit indices are used to assess the final model. The results interpreted from direct effects revealed a positive relationship between decision intelligence and intention to leave. Nevertheless, the mediation analysis within the path model demonstrated that this relationship transformed into a negative one when mediated by job characteristics and meaningful work. In its conclusion, the paper discusses research findings, addresses limitations, and underscores contributions, thus paving the path for integrating decision intelligence into academic literature and industry practices.

Keywords:

• decision intelligence
• job characteristics
• meaningful work
• intention to leave
• serial multiple mediation
• employee retention

1. Introduction

Modern organizations grappled with rampant employee turnover are facing significant challenges and disruptive consequences in their wake. Employee turnover impacts an organization’s financial bottom line resulting in expenditures related to recruitment and training, productivity loss, low morale, reduced enterprise reputation, and loss of opportunities (Zhang, 2016, O’Connell & Kung, 2007). As remote and hybrid work arrangements become commonplace, companies may have to strategize for a year-over-year employee turnover rate surpassing their usual numbers by 50% to 75% (Wiles, 2021, Tupper & Ellis, 2022). Thus, there is a pressing need to understand the factors or constructs shaping the dynamics of employee turnover.

Intention to leave or turnover intention is considered the strongest predictor of actual employee turnover behavior (Hom et al. 2017). While previous research has associated intentions to leave with personal and occupational characteristics, work conditions, interpersonal relationships, organizational culture, and internal and external organizational factors (Arnoux-Nicolas et al., 2016; Kim & Kim, 2021; Li & Yao, 2022; Muchinsky & Morrow, 1980); Zimmerman & Darnold, 2009), less focus has been directed towards exploring how the use of technology for decision making (decision intelligence) impacts employees’ intentions to leave. Decision Intelligence, or DI, unifies several existing technologies into a single framework and bridges them into a natural and familiar form for non-technical decision-makers (Pratt, 2019). DI tools come in various forms, ranging from simple internet searches and large language models like ChatGPT (Mehdi, 2023) to advanced intelligent decision support systems (Zikos & DeLellis, 2018, Sutton et al., 2020), specialized artificial intelligence (AI) powered decision augmentation systems tailored for predictive and prescriptive analysis, and autonomous agents such as robots (Duan et al., 2019, Jain et al., 2021, O’Callaghan, 2023).

In the modern landscape, employees and managers spanning all organizational levels and functions are embracing DI tools for more effective and efficient decision making (O’Callaghan, 2023, Stone et al., 2020, Srivani et al., 2023, Chien et al., 2020, Bankins, 2021). Consequently, management by algorithm—delegating critical business decisions to smart algorithms is becoming a common business practice (Schrage, 2017). Based on a prediction by Gartner (2022), more than a third of large organizations might already be using decision intelligence, including decision modeling. Despite gaining such popularity, few academic studies have examined DI’s impact on employees, work, and organizations. Given its increasing pervasiveness and apparent integration into our lives, it becomes crucial to understand how DI influences employees’ intentions to either remain with or leave their organizations. Furthermore, while both job characteristics and meaningful work have been studied individually to have an impact on employee turnover (Hackman & Oldham, 1975, Humphrey et al., 2007, Garg & Rastogi, 2006, Holbeche & Springett, 2004, Wingerden et al., 2018), no research has been discovered that explores how these factors can mediate the relationship between DI and intention to leave.

Previous studies on the impact of DI on employees have been limited to specific contexts, focusing on particular sectors, industries, and occupations, for example, federal and private sectors (Hellman, 2010, Hur & Abner, 2023), information technology (Ladelsky & Lee, 2022, Joseph et al., 2007), manufacturing (Skelton et al., 2019), teachers (Li & Yao, 2022), and healthcare workers (Kim & Kim, 2021, Shen et al., 2020). In order to examine relationships between the four constructs from a broader perspective, this research studies employees irrespective of any sectors, industries, and occupations.

Thus, this research aims to examine if decision intelligence mediated by job characteristics and meaningful work influences employees’ intention to leave their organizations. Research investigating such relationships is needed because it can be highly actionable, for example, by offering implications to employers for how much to expose employees to DI and related technologies. Therefore, drawing from prior research, this study proposes a mediation-based theory represented through a conceptual model (Figure 1).

Figure 1. Caption: conceptual model with serial multiple mediation.

Note: Solid black lines represent direct hypotheses H1, H2, H3, and H4, whereas hypotheses with mediation effects are represented by a dashed grey line for H5 (mediation effect 1), a dotted grey line for H6 (mediation effect 2), and a solid grey line for H7 (mediation effect 3).

Based on the results of extensive empirical testing, model fit assessment, and evaluation of the model’s prediction capability, we can assert that this model can significantly contribute to our understanding, on both theoretical and practical grounds, of how decision intelligence impacts employees’ turnover intentions. The evidence suggests that using DI technologies while focusing on job characteristics and meaningful work may decrease employees’ intention to leave. Consequently, this research provides valuable insights into employee retention in the contemporary AI-driven business landscape.

2. Theoretical foundations

2.1. Decision intelligence—using technologies for decision making

Studies advocating the use of computer technologies in decision-making can be traced back to the 1970s, highlighted by Mintzberg et al. (1976) research on non-routine, unstructured decisions at the strategic level of organizational hierarchy. In 1981, Lucas Jr. conducted an experimental investigation of computer-based graphics in decision making, while in 1987, March’s study contributed additional insights into the use of information systems for decision making when faced with ambiguity, uncertainty, and incomplete data (Parra et al., 2022). Ever since, numerous studies have examined the role of computer technology in decision-making in various contexts, e.g., AI for decision diagnosis and look ahead in decision making (Pomerol, 1997), computerized decision support systems in general practice (Thornett, 2001), clinical decision intelligence supported by visual cluster analysis (Gotz et al., 2011), supporting decision making process with ideal software agents in the context of what business executives want (Duan et al., 2012), organizational decision-making structures in the age of AI (Shrestha et al., 2019), AI for decision making in the era of big data (Duan et al., 2019), the impact of technology on the human decision‐making process (Darioshi & Lahav, 2021), and augmenting organizational decision-making with AI, deep learning algorithms (Shrestha et al., 2019). Nevertheless, the literature shedding light on how DI affects employee turnover intentions is scarce.

DI allows crystalizing technology into solutions, sheds light on the complexity, and provides insights and solutions that were not possible previously (Pratt, 2019). The use of modern technologies, such as AI, facilitates decision-making at all levels of organizations, from strategic (Price et al., 2018, Trunk et al., 2020, Moser et al., 2021) to tactical (Beal et al., 2019) and operational decisions (Xu et al., 2020). In addition, studies have been reviewed investigating the implications of AI on decisions related to various functional areas of businesses, including marketing (Huang & Rust, 2022, Volkmar et al., 2022), finance (Sujith et al., 2022, Kunnathuvalappil Hariharan, 2018, Bussmann et al., 2021), operations (Helo & Hao, 2022, Grover et al., 2022), innovation (Verganti et al., 2020, Pietronudo et al., 2022), and human resource management (Rodgers et al., 2022, Jarrahi, 2018). In recent years, there has been a significant increase in the use of AI for talent management decisions, such as recruitment, performance evaluation, compensation analysis, and employee training and development (Votto et al., 2021). As DI becomes widespread in business, understanding its effect on employee retention is increasingly vital, particularly during the “age of great resignations”—in 2021, the USA’s “quit rate” hit a 20-year high (Parker & Horowitz, 2022). Thoroughly researched models are critically needed to explain and predict DI’s impact on individuals, society, and organizations, considering its widespread integration into various aspects of our lives and work.

2.2. Intention to leave

In their study, Poberznik et al. (2021) present the case of how mobile robots combined with smart care environments minimize risks related to employee turnover, especially during pandemics. Advocating a similar proposition, Kareem et al. (2020) published their empirically tested findings on how using a technology-based customer relationship management system can significantly reduce employees’ intention to quit. Eickemeyer et al. (2021) proposed a software tool, i4.0, to identify the potential negative effects of digitization on employees to ensure higher employee retention. While some studies suggest that DI may reduce turnover intentions, the literature also presents opposing evidence. Certain studies indicate that DI could lead to an increase in intentions to leave due to added stress caused by the use of information and communication technologies, also known as technostress (e.g., Ayyagari et al., 2011).

Techno-stressors can lead to decreased job satisfaction, intentions to stay, and organizational commitment; higher work pressures, perceptions of work overload, information fatigue, frustration, demoralization, loss of motivation, job burnout, poor job performance, increased intentions to quit a job and dissatisfaction at work (Ragu-Nathan et al., 2008, Gaudioso et al., 2017, Vuori et al., 2019). Current dissatisfaction with one’s employment is indicative of intent to leave, which is a strong predictor of an employee’s turnover decision (Johnsrud & Rosser, 2002, Knani & Fournier, 2013).

The discrepancies in the results of these conflicting studies could stem from the latter studies excessively emphasizing the stress aspects of technology while overlooking its other significant benefits. In Pew research studying 1,150 experts, 47% predicted that individuals’ well-being would be more helped than harmed by digital life in the next decade, while only 32% believed it would cause more harm. The remaining 21% predicted there would not be much change (Anderson & Rainie, 2018). The cases against the use of DI mostly center around limited aspects such as technostress, whereas the evidence in support of it seems more compelling. Therefore, the first hypothesis in this study proposes:

H1:

There is a negative relationship between decision intelligence and employees’ intentions to leave.

2.3. Job characteristics

One of the most influential attempts to design effective jobs is the job characteristics model proposed by Hackman and Oldham (1975), which suggests job characteristics such as skill variety, specialization, task significance, and meaningfulness can result in higher employee satisfaction and retention. While their study remains a significant contribution to understanding different aspects of job design, it lacks consideration of one of the most crucial aspects—technology, which plays a vital role in today’s job landscape.

According to Gibbs and Bazylik’s (2022) study, new technologies such as AI and automation are positively associated with job characteristics as they create new non-routine cognitive and social tasks and make work in those tasks more productive. In their twelve-month study of 2,794 employees in a telecommunications firm, Morris and Venkatesh (2010) found that a technology-based enterprise resource planning system implementation moderated the relationships between three job characteristics (skill variety, autonomy, and feedback) and job satisfaction. A study from Bayo-Moriones et al. (2010) shows that integrated manufacturing, which is often facilitated through advanced technologies and automated systems, has a positive effect on job characteristics such as job variety, autonomy, and interdependence for production workers. In line with the findings of these studies, the next hypothesis posits:

H2:

There is a positive relationship between decision intelligence and job characteristics.

2.4. Meaningful work

Meaningfulness, as Chalofsky (2003) argued, refers to an inclusive state of being and is a significant contributor to the individual’s sense that they have achieved their purpose in life. Therefore, meaningful work can be explained as work that is important, worthwhile, or valuable to self and or others (Pratt & Ashforth, 2003); the work that helps people make sense of their world, facilitates personal growth and contributes to the greater good (Steger et al., 2012, Allan, 2017). Previous studies have found that meaningful work is associated with work engagement, commitment, job satisfaction, life satisfaction, life meaning, general health, and withdrawal intentions, organizational citizenship behaviors, and job performance (e.g., Allan et al., 2019, Johnson & Jiang, 2017, Bailey et al., 2019). Hence, organizations that foster meaningful workplaces have a higher chance of attracting, retaining, and motivating employees, which is essential for the sustainable growth of the business (Bailey & Madden, 2016).

Hackman and Oldham’s (1975, 1976) are most extensively referenced in relation to understanding experienced meaningfulness of work as one of the psychological states arising from job characteristics. As Wrzesniewski and Dutton (2001) theorized, job crafting has important implications for employees’ sense of meaning and identity in their work. The relationship between different job characteristics and meaningful work has been studied multiple times. For example, Allan (2017), in his longitudinal study, found that task significance, a critical job characteristic, significantly predicted meaningful work overtime. Similarly, Schnell et al. (2013) identified that meaning in work can be predicted by work-role fit, the significance of work tasks, socio-moral climate, and organizational self-transcendent orientation. In addition to other job resources, job variety, and autonomy have been found to have a significant and positive direct association with meaningful work (Albrecht et al., 2021). Thus, the following hypothesis proposes:

H3:

There is a positive relationship between job characteristics and meaningful work.

A substantial body of empirical studies suggest that an individual’s psychological and subjective well-being is negatively related to their turnover intentions (e.g., Wright & Bonett, 2007, Yuniasanti et al., 2019, Shi et al., 2021). Employees who perceive their work as meaningful are more committed to the organization and therefore are less likely to leave the organization (Holbeche & Springett, 2004, Wingerden et al., 2018). Studies that investigated relationships between meaningful work and employee turnover directly or indirectly (e.g., Fairlie, 2011, Vermooten et al., 2019, Sun et al., 2019, Arnoux-Nicolas et al., 2016, Humphrey et al., 2007, Garg & Rastogi, 2006, Bailey et al., 2019, Janik & Rothmann, 2015) found that meaningful work was associated with low levels of turnover intentions. For example, in their study, Charles-Leija et al. (2023) showed that meaningful work contributes to people’s life purpose, feelings of being appreciated, enjoyment of daily tasks, and reduces turnover intention. The next hypothesis, building upon robust theoretical foundations in prior literature, postulates:

H4:

There is a negative relationship between meaningful work and employees’ intentions to leave.

Job characteristics can mediate relationships between several constructs related to different organizational outcomes. For example, in their investigation, Judge et al. (2000) found evidence supporting the proposition—job characteristics and job complexity mediate the relationship between core self-evaluations and job satisfaction. Also, mediation analyses in Parker’s (2003) study showed that the negative effects of lean production were partly attributable to declines in perceived work characteristics, including job autonomy, skill utilization, and participation in decision-making. Through the findings in their study, Piccolo and Colquitt (2006) suggest that core job characteristics can mediate the relationship between transformational leadership and job behaviors. Based on the evidence in previous studies, we can propose that job characteristics can also mediate the relationship between decision intelligence and meaningful work. Thus:

H5:

Job characteristics mediate the relationship between decision intelligence and meaningful work.

Oprea et al. (2020) study suggest that meaningful work and work engagement can mediate the negative relationship between job crafting and the intention to leave. Meaning in work can also mediate the relationship between some job crafting dimensions and performance (Junça Silva et al. 2022). As Simonet and Castille’s (2020) study shows, meaningful work acts as a key mediator linking job characteristics to important organizational outcomes; it is increasingly recognized as a source of personal fulfillment and a protective factor against daily stress and adversity. Arnoux-Nicolas et al. (2016) examined 336 French workers from different job contexts and found that meaning of work mediated the relationship between perceived working conditions and turnover intentions and was negatively related to both these constructs. In a similar study, Sun et al. (2019) explored how the meaning of work functions as a mediator between work context perception and turnover intention; their results confirmed that social mission could be entirely mediated by the work meaning to predict low turnover intentions. Despite making significant contributions to understanding meaningful work as a mediator, these studies do not directly investigate if meaningful work can mediate the relationship between job characteristics and intention to leave. Therefore, addressing the gap in the existing literature, the following hypothesis posits:

H6:

Meaningful work mediates the relationship between job characteristics and intention to leave.

Panda et al. (2021) studied 219 executives and their 38 supervisors in a large paint manufacturing industry and found that if an employee finds their job meaningful, they are likely to be more engaged emotionally, psychologically, and cognitively to deliver better job performance. Employee performance is directly related to turnover intentions. High performers are more satisfied with their work and therefore are less likely to leave their jobs compared to low or poor performers (Willyerd, 2014, Zimmerman & Darnold, 2009). Through their study, May et al. (2010) explained how the relationships between job enrichment, work role fit, and engagement can be fully mediated by the psychological condition of meaningfulness. As we approach the final hypothesis, it is essential to recognize that specific job resources, such as job characteristics, significantly and positively influence employee engagement through meaningful work (Albrecht et al., 2021). Additionally, we should acknowledge that technology-based job autonomy, along with job engagement, satisfaction, and organizational commitment, may contribute to reducing employee turnover intention (Carlson et al., 2017). Taken together, the evidence indicates that job characteristics and meaningful work may serve as mediators between decision intelligence and intention to leave. Thus, the final hypothesis in this study postulates:

H7:

Job characteristics and meaningful work mediate the relationship between decision intelligence and intention to leave.

By incorporating arguments in the seven hypotheses, this study proposes a serial mediation model linking the four constructs: decision intelligence, job characteristics, meaningful work, and intention to leave (Figure 1). While focusing on both direct and indirect relationships between the four constructs, the model specifically theorizes that decision intelligence via job characteristics increases employees’ perception of meaning in their work which, in turn, lowers their intentions to leave their organizations.

3. Methods

The primary methodology employed in this study is PLS SEM (partial least squares structural equation modeling, also called PLS path modeling). PLS SEM was chosen due to its efficacy in handling complex models, superior statistical power, and ability to examine mediation effects (Hair et al. 2021). Two software packages, Smart PLS 4 and Amos, are employed for building the model, testing hypotheses, and assessing model fit, ensuring a comprehensive and reliable approach to model development and evaluation. Although both packages provide comparable results, they each offer unique features that this study leverages. For instance, unlike Amos, Smart PLS includes Q² values in its standard output, which plays a vital role in assessing the model’s predictive relevance. On the other hand, Amos excels in providing robust model fit indices, while Smart PLS indices continue to evolve (Dash & Paul, 2021).

3.1. Data collection and participants

The A-priori sample size calculation method (Memon et al., 2020, Kuvaas et al., 2020) was used to determine the adequate sample size. The recommended minimum sample size for this study considering the number of observed (11) and latent (4) variables in the model, the size of the expected effect (0.3), as well as the anticipated probability (0.05) and the level of statistical power (0.8), was 241 participants. In order to ensure the reliability of the results while accommodating resource limitations, a total of 448 participants residing in the United States were recruited through Prolific. Before participating in the survey, all study participants were given a clear understanding of the study’s context and purpose. Their explicit and informed consent was obtained through an agreement to a detailed consent form prior to participation in the study. Nine responses from speeder participants were removed to ensure data quality. Speeders were defined as participants who completed the study in less than 50% of the median time, in accordance with previous research by Greszki et al. (2015).

The final sample consisted of responses from a total of 439 participants and comprised 49.20% females and 50.56% males. One participant (0.22%) did not disclose their gender. In total, 73.12% of the participants reported their ethnicity as white, 6.60% as black, 10.70% as Asian, 5.46% as mixed, and 4.10% as other. Based on age, the participants were divided into two categories: Digital natives and digital immigrants. Digital natives are individuals born after 1980. They were raised in environments surrounded by technology and therefore possessed technological skills different from digital immigrants, a generation born before 1980 (Palfrey & Gasser, 2013, Jarrahi and Eshraghi, 2019). In this sample, 33.71% of participants were digital natives, and 66.28% were digital immigrants. Since the study focuses on work-related aspects, only participants who were either part-time or full-time employees were included, with 72.44% being full-time and 27.56% being part-time employees.

3.2. Measures

All the items in the four constructs of this study are based on existing research. Decision intelligence (DI), which measures the extent of technology’s involvement in the decision-making process at the employee’s organization, is a single-factor variable measured by the item scale from Darioshi and Lahav’s (2021) study. Respondents rate this item on a five-point Likert scale ranging from 1 (extremely low) to 5 (extremely high). The other three constructs are also measured on a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

For Job characteristics (JC), the two highest-loading items were selected from the job characteristics model developed by Hackman and Oldham (1980). The questions for these two items are: “The job itself is very significant and important in the broader scheme of things” and “The job requires me to use a number of complex or high-level skills.” Another item focused on the question, “The job requires a depth of knowledge and expertise,” was added from The WDQ (Work Design Questionnaire) developed by Morgeson and Humphrey (2006).

The WAMI (Work and Meaning Inventory) developed by Steger, Dik, and Duffy (2012) was used to measure meaningful work (MFW). The five highest-loading items were retained from the original study. The questions for the five selected items are: “I view my work as contributing to my personal growth,” “I understand how my work contributes to my life’s meaning,” “I have a good sense of what makes my job meaningful,” “I have discovered work that has a satisfying purpose,” and “The work I do serves a greater purpose.”

Two items are adapted from the study performed by Landau and Hammer (1986) to assess respondents’ intentions to leave (ITL) their organizations. The questions for these items are: “I am seriously thinking about quitting my job” (thoughts) and “I am actively looking for a job outside of my organization” (action).

4. Results

The statistical model in Figure 2 displays the serial mediation (Quintana et al., 2020) in which the variable decision intelligence is modeled as affecting intention to leave through four pathways (i.e., a1b1, a2b2, a1d21b2, c′). Arrows in the figure represent the paths of the tested model, and a1, a2, b1, b2, d21, c, and c′, indicate the standardized path coefficients.

Figure 2. Caption: statistical model with serial mediation.

Note: All reported values are standardized path coefficients with p values in parentheses. Direct effects are reported for paths c’ (H1), a1 (H2), d21 (H3), b2 (H4), and indirect effects are reported for mediation effect 1 (H5), mediation effect 2 (H6), mediation effect 3 (H7), a2, and b1. The values for c are total effects. For all paths, p<0.05.

4.1. The measurement model

The validity and reliability of the measurement model were confirmed through CFA (confirmatory factor analysis) as the primary analysis tool using Smart PLS 4. The convergent validity of all constructs was supported by loadings greater than 0.70 for all indicators. Additionally, all constructs exceeded the minimum threshold of 0.70 for both Cronbach’s alpha values and composite reliability values. As depicted in Table 1, the AVE (average variance extracted) values above the threshold of 0.50 indicate that the construct explains at least 50 percent of the variance of its items (Hair et al., 2019).

Table 1. Construct reliability and validity

	Cronbach’s alpha	Composite reliability (rho_c)	AVE
ITL	.875	.941	0.888
JC	.789	.871	0.693
MFW	.927	.945	0.774

ITL intention to leave; JD job characteristics; MFW meaningful work

The discriminant validity is assessed by the HTMT (Heterotrait-Monotrait) ratio of the correlations. Since the constructs in this study are conceptually distinct from each other, the more conservative threshold of 0.85 is selected. Values above the threshold suggest that discriminant validity is not present (Henseler & Sarstedt, 2013). As shown in Table 2, all the HTMT values are lower than 0.85, confirming the discriminant validity of this model.

Table 2. Discriminant validity heterotrait-monotrait ratio matrix

	DI	ITL	JD	MFW
DI
ITL	0.059
JC	0.206	0.397
MFW	0.089	0.533	0.703

DI decision intelligence; ITL intention to leave; JC job characteristics; MFW meaningful work

4.2. The structural model and hypothesis testing

The bootstrapping method with a consistent PLS algorithm in Smart PLS 4 was used to assess the structural model and test hypotheses. The VIF (variance inflation factor) values of all items were found to be lower than 3.5, well below the maximum threshold of 5, indicating no collinearity issues that could bias the regression results. Hypotheses were then tested using 5000 resampling iterations with the bootstrapping method, as recommended by Hair et al. (2019). The first four hypotheses (H1, H2, H3, and H4) were validated with direct effects (path coefficients), while mediation-based hypotheses (H5, H6, and H7) were validated with specific indirect effects. All hypotheses were tested at a 95% confidence level with an alpha value of 0.05, and CIs (Confidence Intervals) were used as assessment matrices. CI values were considered significant if the ULC (Upper-Level Confidence) and LLC (Lower-Level Confidence) did not contain zero (Hayes, 2013). The results of all direct effect hypotheses are presented in Table 3.

Table 3. Hypothesis testing for direct effect hypothesis

Hypothesis	Relationship	Beta	SE	T value	p value	LLC	ULC	VIF
H1	DI -> ITL	0.108	0.047	2.279	.023	0.016	0.200	1.008
H2	DI -> JC	0.203	0.053	3.789	.000	0.095	0.306	1.000
H3	JC -> MFW	0.728	0.032	22.795	.000	0.662	0.786	1.000
H4	MFW -> ITL	−0.544	0.043	12.738	.000	−0.623	−0.452	1.008

DI decision intelligence; ITL intention to leave; JD job design; MFW meaningful work

The first hypothesis (H1) postulated a negative association between DI and ITL. However, the results revealed a significant positive relationship between these variables (β = 0.108, t = 2.279, LLC = 0.016, ULC = 0.200, p < 0.05), leading to the rejection of H1.

The second hypothesis (H2) proposed a positive association between DI and JC. The analysis demonstrated a significant positive relationship between these constructs (β = 0.203, t = 3.789, LLC = 0.095, ULC = 0.306, p < 0.05), providing evidence for the validation of H2.

Hypothesis 3

(H3) suggested a positive relationship between JC and MFW. The results revealed a significant positive relationship between these constructs (β = 0.728, t = 22.795, LLC = 0.662, ULC = 0.786, p < 0.05), thus validating H3.

Finally, the fourth hypothesis (H4) hypothesized a negative relationship between MFW and ITL. The results demonstrated a significant negative relationship between these constructs (β = −0.544, t = 12.738, LLC = −0.623, ULC = −0.452, p < 0.05), providing support for the validation of H4.

4.3. Mediation analysis

Hypotheses 5, 6, and 7 of this study involve mediation effects that have been tested using bootstrapping, as recommended by Hair et al. (2017) and Preacher and Hayes (2008). The results of the mediation analysis are presented in Table 4. Hypothesis 5 proposes that JC mediate the relationship between DI and MFW, and the results demonstrate a significant mediation effect (β = 0.148, t = 3.686, LLC = 0.068, ULC = 0.226, p < 0.05) in support of Hypothesis 5, which is validated. It should be noted that mediation is indirect only mediation that occurs due to the significant indirect effect between DI and MFW in the absence of a direct effect (Meule, 2019, Hair et al. 2021).

Table 4. Mediation analysis

Hypothesis	Relationship	Beta	SE	T value	p value	LLC	ULC	Mediation type
H5	DI -> JC -> MFW	0.148	0.040	3.686	.000	0.068	0.226	Indirect only
H6	JC -> MFW -> ITL	−0.396	0.037	10.825	.000	−0.466	−0.323	Indirect only
H7	DI -> JC -> MFW -> ITL	−0.080	0.023	3.516	.000	−0.127	−0.038	Competitive

DI decision intelligence; ITL intention to leave; JD job characteristics; MFW meaningful work

Next, H6 posits that MFW mediates the relationship between JC and ITL; the results confirm a significant effect of MFW as a mediator between JC and ITL. With β = −0.396, t = 10.825, LLC = −0.466, ULC = −0.323, and p < 0.05, H6 is validated. Regarding mediation type, H6 also shows an indirect-only mediation since the indirect effect between JC and ITL is significant in the absence of any direct effect.

Finally, H7 is the most important hypothesis in this study, proposing that JC and MFW mediate the relationship between DI and ITL. The results reveal significant mediation effects of JC and MFW on DI and ITL (β = −0.080, t = 3.516, LLC = −0.127, ULC = −0.038, p < 0.05), and thus, H7 is validated. Notably, this hypothesis represents competitive mediation where the indirect and direct effects are significant but point in opposite directions (Hair et al., 2021).

The predictive capacity of the model was evaluated using the Q² predict measure, which was computed via the blindfolding technique employing the PLSpredict algorithm in Smart PLS 4. To determine the predictive accuracy of the endogenous construct’s structural model, the Q² values for each construct should be greater than zero. According to the established benchmarks, Q² values of 0.25, 0.50, and above indicate medium, large, and substantial predictive relevance, respectively (Hair et al., 2017). In this study, the endogenous constructs demonstrated a small predictive relevance (ITL = −0.002, JC = 0.029, and MFW = 0.004). However, as the values were greater than zero, it can be argued that the model’s predictive relevance is confirmed, albeit with a small effect. Table 5 summarizes the predictive relevance (Q²) and effect size based on f² values.

Table 5. Predictive relevance (Q²) and effect size (f²)

Relationship	Q^2 predict	f^2	Effect size (f^2)
DI -> ITL	−.002	0.016	Small
DI -> JC	.029	0.043	Small
JC -> MFW	.004	1.13	Large
MFW -> ITL		0.418	Large

DI decision intelligence; ITL intention to leave; JD job characteristics; MFW meaningful work

The analysis based on R² values indicates that DI accounts for 4.1% of the variance in JC, while JC explains 53.1% of the variance in MFW, and MFW accounts for 29.7% of the variance in ITL. The benchmarks for R² values are substantial (0.75), moderate (0.50), and weak (0.25) (Henseler et al., 2009, Hair et al., 2011). Based on these benchmarks, JC has weak explanatory power (R² = 0.041), MFW is above moderate (R² = 0.531), and ITL is between weak and moderate (R² = 0.297). It is recommended we interpret the R² values considering the context of the study. The R² value is influenced by the number of predictor constructs in the model, which means that a larger number of predictor constructs may lead to a higher R² value. Thus, an R² value as low as 0.10 may be considered satisfactory in some cases. As the number of predictor constructs in this study is not large, the possibility of obtaining lower or weak R² values for each construct cannot be ruled out (Hair et al., 2017).

4.4. Model fit assessment

The model fit is evaluated using three categories of fit indices: absolute, incremental, and parsimonious, as recommended by Dash and Paul (2021). The analysis is performed utilizing Amos to generate results for all three categories of fit indices. The model values are then compared against established threshold values from previous studies. The relevant indices for each category, ideal threshold values, and model values are presented in Table 6.

Table 6. Model fit assessment based on three categories of model fit indices

Category	Index	Ideal Threshold Value	Model Value
Absolute Fit Measures	CMIN/df (Chi-square minimum discrepancy/degrees of freedom)	<3 (even <5) (Dash & Paul, 2021)	3.81
	GFI (Goodness-of-fit index)	>0.90 (Hair et al., 2017; Miles & Shevlin, 2007)	0.94
	AGFI (Adjusted goodness-of-fit Index)	>0.90 (Shi & Maydeu-Olivares, 2020)	0.90
	RMSEA (Root mean square error of approximation)	<0.08 (marginal values between 0.08 and 0.1) (Kim et al., 2016)	0.08
	SRMR (Standardized root mean square residuals)	<0.08 (Hu & Bentler, 2009)	0.06
Incremental Fit Measures	TLI (Tucker-Lewis index)	>0.90 (Hair et al. 2020)	0.95
	NFI (Normed fit index)	>0.90 (Dash & Paul, 2021)	0.95
	CFI (Comparative fit index)	>0.95 (Hair et al. 2020)	0.96
Parsimonious Fit Measures	PNFI (Parsimonious normed fit index)	>0.50 (Hooper et al., 2008)	0.64
	PCFI (Parsimonious comparative fit index)	>0.50 (Hooper et al., 2008)	0.65

Concerning the absolute fit measures, the CMIN/df (Chi-square minimum discrepancy/degrees of freedom) value of 3.81 falls between the thresholds of 3 and 5. The GFI (goodness-of-fit index) value of 0.94 exceeds the minimum threshold of 0.90, while the AGFI (adjusted goodness-of-fit Index) value of 0.90 matches the minimum threshold of 0.90. Furthermore, the SRMR (standardized root mean square residuals) value of 0.06 is below the maximum threshold of 0.08. Although the RMSEA (root mean square error of approximation) value of 0.08 may be considered a marginal fit, it remains within the upper level of the acceptable fit threshold. Questioning model’s fit solely on RMSEA results would not be suitable, as the model demonstrates exceptional performance on other absolute fit measures (Kim et al., 2016).

As for the incremental fit measures, the TLI (Tucker-Lewis index) value of 0.95 exceeds the minimum threshold of 0.90, as does the NFI (normed fit index) value of 0.95. Additionally, the CFI (comparative fit index) value of 0.96 is above the minimum threshold of 0.95, indicating a good fit for the model.

Furthermore, for the parsimonious fit measures, both PNFI (parsimonious normed fit index) and PCFI (parsimonious comparative fit index) values of 0.64 and 0.65, respectively, surpass the minimum threshold of 0.50, signifying that model aligns with the principle of parsimony.

Based on the results of the model fit evaluation on absolute, incremental, and parsimonious fit measures, it can be concluded that the model exhibits relatively strong performance across all three indices, indicating that it is a well-fitting model.

5. Discussion

The direct effect path coefficients revealed a positive association between decision intelligence and intention to leave, contradicting the initial hypothesis (H1) proposing the opposite. Previous studies did present conflicting evidence, but this study opted to align with research that demonstrated a negative association between DI and intention to leave (e.g., Poberznik et al., 2021, Kareem et al., 2020, Eickemeyer et al., 2021). However, empirical testing ultimately confirmed that DI might increase employees’ intentions to leave, in line with findings from some studies quoted earlier (e.g., Knani and Fournier, 2013, Ayyagari et al., 2011, Ragu-Nathan et al., 2008, Gaudioso et al., 2017, Vuori et al., 2019).

The validation of H2 provides evidence of a positive association between DI and job characteristics. These findings align with earlier studies cited in this paper (e.g., Gibbs & Bazylik, 2022, Morris & Venkatesh, 2010, Bayo-Moriones et al., 2010). Similarly, H3 was confirmed as anticipated, supported by existing literature (e.g., Hackman & Oldham, 1975, 1976, Wrzesniewski & Dutton, 2001, Allan, 2017, Schnell et al., 2013), which consistently indicated, although in different contexts, a positive association between job characteristics and meaningful work.

The theoretical support for H4 was robust, with numerous studies cited in favor of the proposition that meaningful work is inversely related to employees’ intentions to leave (e.g., Humphrey et al., 2007, Garg & Rastogi, 2006, Fairlie, 2011, Vermooten et al., 2019, Sun et al., 2019, Arnoux-Nicolas et al., 2016, Bailey et al., 2019, Janik & Rothmann, 2015, Charles-Leija et al., 2023). Hypothesis testing yielded results that supported H4, validating the claims made in the existing literature and reinforcing the notion that meaningful work significantly reduces employees’ intentions to leave.

Hypothesis H5 proposed an indirect mediation, suggesting that job characteristics act as a mediator between DI and meaningful work. This hypothesis lacked direct evidence in the literature but received sufficient indirect support from related studies (e.g., Judge et al., 2000, Parker, 2003, Piccolo & Colquitt, 2006). Hypothesis testing validated this proposition, offering a distinctive contribution to the existing literature by establishing the role of job characteristics as a mediator between decision intelligence and meaningful work.

Oprea et al. (2020) provided the closest support to H6, suggesting that meaningful work could mediate the negative link between job crafting and the intention to leave. Hypothesis testing confirmed this hypothesis, aligning with other cited studies, even though they were based on different constructs and contexts (e.g., Junça Silva et al. 2022, Simonet & Castille, 2020, Arnoux-Nicolas et al., 2016, Sun et al., 2019).

The last hypothesis, H7, posited job characteristics and meaningful work mediate the relationship between DI and intention to leave. Though some studies supported this hypothesis, the use of different constructs in prior relationships only provided partial validation (e.g., Panda et al., 2021, Willyerd, 2014, Zimmerman and Darnold, 2009, May et al., 2010, Albrecht et al., 2021, Carlson et al., 2017). Yet, hypothesis testing confirmed H7, introducing a novel empirically tested proposition to the existing literature, enriching our understanding of the intricate relationship between decision intelligence, job characteristics, meaningful work, and intention to leave.

Overall, the findings suggest that DI can effectively reduce employees’ intentions to leave when mediated by job characteristics and meaningful work. However, without this mediation, DI may lead to a significant increase in turnover intentions. The rapid adoption of DI technologies makes it clear that organizations must incorporate them in some form to remain sustainable. However, increased technology use may also contribute to employee turnover, as supported by this study’s findings and consistent with existing literature. To counteract the negative relationship between technology use and employee turnover, organizations should focus on designing jobs with job characteristics that provide meaning and satisfaction for employees and are negatively associated with turnover intentions. These characteristics include significance, specialization, and skill variety, as evidenced in this study.

5.1. Limitations

It is important to note that the proportion of digital natives in this study was 33.71%, with the remaining 66.28% identifying as digital immigrants. The adoption of new technologies can pose significant challenges for digital immigrants, and eventually, their intention to leave may increase with higher usage of DI technologies. The broad scope of this study, encompassing both full-time and part-time employee participants from diverse sectors, job roles, industries, and occupations, might raise concerns about the validity of findings in situations where these specific factors strongly influence employees’ intentions and actions. Therefore, it is likely that the findings of this study may change and become more useful with a larger, controlled sample and more evenly distributed data.

Due to the structural, resource, and process-based constraints, the model in this study overlooks a few factors, for example, interpersonal relationships in the workplace, development opportunities, compensation, work-life balance, management and leadership, work environment, social support, autonomy, training, and development that are identified as the determining factors of employee retention (Kossivi et al., 2016).

6. Conclusion

The rigorously tested model in this study reinforced the theoretical and empirical validity of the theory, suggesting that decision intelligence’s impact on employees’ intention to leave is dependent on job characteristics and meaningful work as mediating factors. Despite its growing popularity, academic studies on decision intelligence remain scarce. Valuable practical insights emerge, guiding organizations in implementing appropriate job design interventions that incorporate the meaning aspect of work into jobs, ultimately enhancing employee retention rates. Furthermore, the study underscores the necessity for further interdisciplinary academic exploration of decision intelligence and its ramifications for organizations and employees. Subsequent research could expand on these findings, investigating similar or distinct constructs to deepen our understanding of this crucial discipline.

Despite the persistent fears surrounding the implementation of AI technologies for decision making, DI has firmly established its presence and is likely to remain an integral part of our lives. It will significantly impact the nature of work, raising the need for different skills and competencies required for employee success. It will be necessary for organizations to prioritize employee training, development, and engagement in the context of DI adoption to ensure that the technology is leveraged in a way that benefits both the organization and its most important resource—people.

Open Scholarship

This article has earned the Center for Open Science badges for Open Data, Open Materials and Preregistered. The data and materials are openly accessible at Turnover Intentions Dataset

Ethics statement

This research was approved by William Woods University’s Institutional Review Board (IRB), protocol number 442, and followed all required procedures to ensure the safety and informed consent of the participants.

Disclosure statement

The data used in this paper is also used for another study investigating the relationship between procedural justice and intention to leave in presence of two mediators. Informed consent was obtained from the participants prior to the survey, informing them their responses may be used for more than one study. This research followed all required procedures to ensure the safety and consent of the participants.

Data availability statement

The data described in this article are openly available in the Open Science Framework at https://10.1080/23311908.2023.2258475.

Additional information

Notes on contributors

Miriam O’Callaghan

Miriam O’Callaghan, PhD, is an Associate Professor of Management and the Dean of the School of Business and Technology at William Woods University. She also holds the Associate Dean of Research and Scholarship position at William Woods University. Her research interests span organizational psychology and behavior, decision-making, emotional intelligence, and the impact of technology on human behavior.