Concurrent IT Sourcing: Mechanisms and Contingent Advantages

Abstract

A growing trend to simultaneously insource and outsource the same information technology (IT) activities (“concurrent IT sourcing”) has not yet received research attention. Although it is widespread and recent empirical studies have detected that in-house IT can complement IT outsourcing, when and how concurrent IT sourcing pays off is not yet understood. This study introduces the notion of concurrent IT sourcing. It then develops two interrelated ideas: concurrent IT sourcing simultaneously enhances in-house and outsourced IT performance: (a) via distinctive mechanisms, but (b) only when vendors’ IT capabilities complement the client’s. Econometric tests using survey data from 233 firms support these ideas. Our novel contribution is to explain when and how concurrent IT sourcing enhances a client firm’s inhouse and outsourced IT performance. The explanatory mechanisms for outsourced IT performance are socialization and modeling of clients’ in-house IT practices by vendors; for in-house IT performance they are knowledge spillovers and ratcheting. For practice, our study shows that when a firm’s in-house capabilities complement its IT vendors’ capabilities, firms can simultaneously outsource and insource the same IT activities to enhance both in-house and outsourced IT performance.

Key words and phrases:

• concurrent IT sourcing
• insourcing
• IT capabilities
• IT governance
• outsourcing
• plural governance
• seemingly unrelated regression

Notes

1. Concurrent sourcing is theoretically distinct from related notions such as taper integration (wherein an activity is outsourced after some stages are done in-house) [27] and multisourcing (partitioning of outsourced work among multiple vendors) [4]. It is observed in IT services, industrial purchasing, manufacturing, construction, and in restaurants [11, 29, 43, 49].

2. Concurrent IT sourcing differs in two ways from restaurant chains (e.g. [11]) and industrial components (e.g. [29, 49]) predominantly studied in prior concurrent sourcing studies. First, IT activities are often client-specific and less repeatable, thus requiring greater idiosyncratic knowledge of the client firm. Second, IT sourcing involves noncontractible elements that are less readily measurable, unlike the quality of industrial components that can be objectively assessed [46]. This can introduce challenges in outsourcing of IT activities that might not exist to the same extent in non-IT activities.

3. Studies in the reference disciplines exhibit lesser consistency; besides projects [43] and dyads [31], they use the transaction [28], product line, channels [57], even entire agglomerations of franchised and corporate stores [11].

4. Specifically, we first delineate how IT is unlike restaurant chains, sheet metal manufacturing industries, and industrial purchasing in prior concurrent sourcing studies (e.g. [11, 29, 49]).

5. For example, Parmigiani [49] found manufacturing firms’ reliance on suppliers with skills unlike their own; and Mayer and Nickerson [43] observed that firms relied on outside IT vendors for complex technical tasks. Unlike our model, the concurrent sourcing literature emphasizes interfirm complementarity in products rather than firm capabilities (e.g. [29, 49]).

6. This mimics Rustagi, King, and Kirsch’s [56] strategy for measuring the variety of control mechanisms in IT projects.

7. Given high correlations between modeling and socialization in Table 2, we calculated variance inflation factor (VIF) scores for all independent variables including the interaction terms. The VIF scores ranged from 1.11 (volatility) to 2.45 (socialization), all below the conservative cutoff of 5 [42].

8. Structural equation modeling was unsuitable due to interactions, endogenous/ exogenous unspecifiability and model underidentification challenges [18].

9. We found substantial consistency using a more stringent coding scheme counting as concurrent only IT activities scoring in the middle (4) of the scale (37–63 percent).

10. Although we followed the prior concurrent sourcing literature in dummy coding each IT activity for presence or absence of concurrency, we also checked whether the extent of insourcing rather than a dichotomous absence/ presence of concurrency mattered (i.e., a more granular distinction between say 80 percent vis-à-vis 30 percent outsourced). We therefore reanalyzed the data using a summative scale adding the actual scores of the individual items in the scale without dummy coding them. We found no support that degree mattered beyond simple presence or absence of concurrency, consistent with Heide [29].

11. We assessed whether explicitly differentiating outsourcing from insourcing as the two nonconcurrent sourcing modes changed the results by interacting outsourcing and insourcing mode dummies with concurrent sourcing. We found that the signs and significance were negative and identical across the two nonconcurrent sourcing modes, consistent with their conceptual aggregation as sole sourcing in our model.

12. To assess whether firms that do not use concurrent IT sourcing at all might confound the results, we reestimated the models after dropping cases of firms that did only sole sourcing (i.e., retaining only cases with concurrent IT sourcing construct scores of one or higher). All the results were consistent.

13. We dropped three different controls from each equation in Tables 4 and 5 so they met the order condition for identification.

14. The null for the Basmann’s [8] test is that the model’s overidentifying restrictions are valid. Basmann χ² was nonsignificant for modeling (6.06; p = .20), socialization (2.94; p = .57), ratcheting (7.20; p = .13), and spillovers (4.13; p = .39), suggesting that the model is overidentified as appropriate. This was also reconfirmed by the Hansen J-test (χ² = 4.49, p = .34 for modeling; χ² = 2.68, p = .61 for socialization; χ² = 8.17, p = .09 for ratcheting; χ² = 3.81, and p = .43 for spillovers; all nonsignificant).

15. In Harman’s one-factor test, the emergence of a single factor that accounts for a large proportion of the variance suggests a common methods bias. No such single factor emerged, and the first factor accounted for 22.6 percent of the 80.7 percent variance.

16. Marker variable tests use a theoretically unrelated variable to adjust the correlations among the principal constructs in the model. As a marker variable is unrelated to the study’s principal constructs, the correlations should be close to zero. We used firm employment count as the marker variable, for which we have little theoretical basis to expect a relationship with the study’s principal constructs. The average correlation between the model’s theoretical variables was 0.06 (t = .55; ns), providing no evidence of common methods bias.

17. We statistically triangulated IT performance responses of internal IT managers with those from line function managers. Since our matched-pair sample from line managers yielded only a 31.3 percent response rate (N = 73), the line function respondent could not be used directly in the analysis. This limited subsample sufficed to assess line function managers’ and IT department managers’ interrater agreement in three complementary ways: correlations in responses, intraclass correlation coefficient (ICC), and paired tests of differences in raters’ assessments. We found positive and statistically significant correlations in their assessments of in-house IT performance (γ = .42; p < .01) and outsourced IT performance (γ = .29; p < .05). The ICCs indicating agreement were also significant for outsourced IT performance (.45, p = .06) and in-house performance (.28, p = .09). Paired t-tests for differences in the two sets of raters for each firm were nonsignificant for in-house performance (t-value = –.65, ns) and outsourced IT performance (t-value = –.90, ns), further confirming agreement.

18. T-tests revealed no significant differences between the model variables or in the characteristics of the firms that did and did not provide matched-pair responses (IT experience of respondents t = –.13; IT capability complementarity t = .95; outsourced IT performance t = 1.64; in-house IT performance t = –.16; concurrent IT sourcing t = .47; volatility t = .11; concurrent IT sourcing experience t = –.84; vendor density t = .49; outsourcing intensity t = –1.42; all nonsignificant).

Additional information

Notes on contributors

Amrit Tiwana

Amrit Tiwana ([email protected]; corresponding author) is the George Benson Professor at the Terry College of Business at the University of Georgia. He is a senior editor at Information Systems Research and on the editorial boards of Journal of Management Information Systems and Strategic Management Journal. He previously served on the editorial boards of MIS Quarterly and IEEE Transactions on Engineering Management. His research focuses on information technology governance at the firm, interfirm, and ecosystem levels. His work has appeared in Information Systems Research, Journal of Management Information Systems, MIS Quarterly, Strategic Management Journal, California Management Review, and other journals. His research has been supported by various firms, industry consortia, and government agencies in the United States, Europe, and Japan.

Stephen K. Kim

Stephen K. Kim ([email protected]) is Dean’s Professor of Marketing at the College of Business, Iowa State University. He holds a Ph.D. from the University of Southern California. His research interests include new forms of interfirm governance, interfirm control systems, and incentive issues in marketing and sales. His research has been published in Information Systems Research, Journal of Marketing, Journal of Marketing Research, Journal of the Academy of Marketing Science, and other journals. He serves as senior associate editor for Marketing Channels Research Journal (South Korea) and on the editorial review board of the Journal of the Academy of Marketing Science. He has consulted for organizations including Coors Brewing Company, Global Technology Distribution Council, IBM, Hyundai Motors, and SPC Corporation.