Risks of Rapid Application Development Ritu Agarwal, Jayesh Prasad, Mohan Tanniru, and John Lynch Proliferating information systems (IS) application development backlogs and everescalating business demands for applications software have resulted in a constant search by IS organizations for paradigms and tools that accelerate the pace of software development. Along with rapid application development (RAD), technologies such as CASE, object-oriented development, client/server computing, and flexible middleware are being hailed as potential solutions to the software crisis [11]. Although industry experience with these paradigms is mixed, the benefits of such technologies are acknowledged—at least with regard to improvements in development productivity and application delivery times [3]. While no universal definition of RAD exists, it can be characterized in two ways: as a methodology prescribing certain phases in software development (similar in principle to the spiral, iterative models of software construction), and as a class of tools that allow for speedy object development, graphical user interfaces, and reusable code for client/server applications. Indeed, the tools and methodology are inextricably linked: the tools enable the methodology and circumscribe what is accomplished during a development project. In a world dominated by deadlines and irate users, where competitive advantage can depend on how quickly software is developed to support new business, RAD is certainly appealing. The popular press extols its virtues with adjectives like “evolutionary,” “iterative,” “interactive,” and “dynamic,” emphasizing the delivery rate increases facilitated by RAD, which range from 25% to 1,000% [4]. But software development does not conclude with speedy application delivery. For the true benefits of the technology to be realized, the applications must exhibit reusability and maintainability so that total life-cycle costs are reduced. Surprisingly little time has been spent determining if the value of RAD tools to business extends beyond the
Ritu Agarwal (
[email protected]) is an associate professor at the Robert H. Smith School of Business, University of Maryland, College Park, MD. Jayesh Prasad (
[email protected]) is an associate professor in the Department of MIS and Decision Sciences, University of Dayton, Dayton, OH. Mohan Tanniru (
[email protected]) is Professor and Director of the Applied Technology in Business Program at Oakland University, Rochester, MI. John Lynch is a senior consultant with Vertechs Software Solutions, Atlanta, GA. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. © 2000 ACM 0002-0782/00/1100 $5.00 COMMUNICATIONS OF THE ACM
177
next deadline or the next deliverable. RAD methodologies are clearly needed to produce software quickly, and a compressed development life cycle requires powerful RAD tools. But knowledge of how to incorporate RAD tools into an IS shop’s tool kit and the longterm implications of current usage patterns of RAD is woefully limited. Systems developers are assumed to be receptive to the new tools; indeed, the prevailing view amongst IS management appears to be “If we buy it, they will come.” But mandating technology usage has long been recognized as an inappropriate management tactic for obtaining buy-in and commitment from systems developers. Management must identify the types of developers more likely to respond positively to RAD. Another perhaps unsubstantiated claim of RAD technology is its ability to build higher quality systems that are more maintainable, reusable, and extensible [10]. Questions regarding precisely how RAD tools are used remain unanswered. What are the pitfalls of RAD, and will the methodology truly address the software crisis? Will RAD tools provide business value in the long-run, or do caveats exist that managers should be sensitive to? In this article, we focus first on the steps needed to implement RAD tools, including the types of developers more likely to be receptive to the technology and how it should be positioned to maximize acceptance; and second, on usage patterns of RAD tools and their relationship to perceived benefits. The research strategy utilized is one of triangulation: we use a broad-based cross-company survey to determine patterns of diffusion and use, and to generate plausible explanations. We then verify explanations using richer data through interviews with practicing managers and analysts using the technology. One RAD tool that supports all the features commonly associated with RAD was selected to control for possible variance arising from the tool itself. The results are somewhat disturbing in regard to the changes in the development life cycle being wrought by the tools.
The RAD Phenomenon James Martin coined the term RAD in the early 1990s to distinguish the methodology from the traditional waterfall model for systems development. “RAD refers to a development life cycle designed to give much faster development and higher quality results than the traditional life cycle. It is designed to take maximum advantage of powerful development software that has evolved recently.” [9]. According to Martin, four fundamental aspects of fast development exist: tools, methodology, people, and management. The quintessential characteristics of RAD tools include the capability for planning, data and process modeling, code generation, and testing and debugging. RAD methodologies encompass three-stage and four-stage cycles. The four-stage cycle consists of requirements planning, user design, construction, and cutover, while in the three-stage cycle, requirements planning and user design are consolidated into one iterative activity. In a typical RAD life cycle, the requirements specification and design phases consume approximately 30% of the total effort [9]. RAD teams typically consist of individuals with diverse skills and titles, including repository managers, data modeling experts, and construction experts. Effective management of RAD necessitates managing the cultural change and potential resistance associated with the implementation of any new technology. First-generation RAD tools suffered from the limitations inherent in their interpretive nature and inability to scale up to enterprise-level production applications. Current RAD tools have overcome these shortcomings and provide a range of varied functionality, including CASE features, visual programming, object creation, remote data access using SQL, and support for COMMUNICATIONS OF THE ACM
178
multitier client/server architectures.
RAD Implementation RAD technologies incorporate a fundamentally different approach to systems development than the dominant software development approaches of the last three decades [9]. RAD implies successive iteration, refinement, and an accelerated movement from a prototype to a production system. As a tool it uses objects and message-passing metaphors, and emphasizes reusability, visual programming, and graphical user interfaces. Given that the majority of IS professionals in the workforce today did not receive formal training in such tools, issues exist related to acceptance of the new technology. Extensive research on the acceptance of innovations suggests that implementing and managing new technologies poses challenges of considerable magnitude. We undertake an exploration of management interventions that can help facilitate the assimilation of RAD technology using the research model shown in Figure 1. The model identifies three perceptions that have been consistently related to positive usage intentions, which predict actual usage of an innovation [1, 12]: • Relative advantage assesses the extent to which a developer believes the technology represents an improvement over prior methods. • Ease-of-use measures the perceived cognitive effort necessary to effectively utilize the new tool. • Compatibility measures perceived congruence of new technology with preferred methods of accomplishing tasks. The model identifies two additional factors, the study of which can also help formulate management interventions: individual difference, and method-of-implemention variables. Prior studies focused primarily on the internal psychological processes that lead to acceptance of new technologies; few have focused on how management might proactively influence the process of acceptance (as opposed to compelling technology use by edict). The individual difference variables of a developer include his/her: prior mainframe and client/server experience and personal innovativeness with respect to new information technologies. Our selection of these variables was based on extensive research in learning theories and the diffusion of innovations. In the human-associative view of learning, the law of proactive inhibition suggests that prior experiences can affect the ability to learn new behaviors, depending on the extent of similarity between prior experiences and the new behavior to be learned. Similar experiences promote positive attitudes toward the new object while dissimilar experiences influence learning and attitudes negatively. Research examining skills transfer of systems analysts experienced in process-oriented modeling identified a performance deterioration when analysts moved to an OO modeling environment [2]; thus, empirical support exists for a persistent assumption in the IS community—that developers with significant experience in more traditional systems construction methods have more difficulty accepting new development paradigms. Perhaps their experience with the extensive discipline and structure associated with mainframebased projects makes them skeptical of the seemingly ad hoc nature of RAD. “The introduction of a RAD life cycle is alarming and threatening to professionals comfortable with an older and slower methodology,” notes Martin [9]. In addition to the prior experience of a developer, his/her personal innovativeness in the domain of information COMMUNICATIONS OF THE ACM
179
technology, defined as the propensity of an individual to try out any new information technology, has also been shown to significantly influence behavior toward a technology innovation [8]. The method-of-implementation variable includes a measure of the speed of RAD technology adoption, as well as a developer’s assessment of the scope of the change— the “radicalness” of its departure from existing software development and delivery paradigms. Gallivan et al. point out that strategies for the successful assimilation of any new technology necessitate attentiveness to the speed of its introduction into an organization, as well as to the extent to which the change represents a radical as opposed to an incremental modification of current practice. The items used to measure each construct are shown in Table 1.
Understanding RAD Outcomes As noted previously, the goals of RAD use include faster software development, and better, more maintainable systems [9]. In addition, given that many RAD tools now incorporate OO design paradigms, IS shops that adopt RAD clearly expect downstream reusability benefits. But it is not clear exactly how the usage of specific features
Table 1. Constructs and scales. COMMUNICATIONS OF THE ACM
180
Table 2. Operationalizing the benefits of RAD.
of RAD tools relates to benefits, or what features of RAD tools are even being utilized. The model presented in Figure 2 is used to explore the relationship between RAD tool usage and expected outcomes. Outcomes are categorized into two classes based on a review of the literature. The global benefits class includes overall application development productivity, resource consumption, and overall management of software development projects. Specific outcomes represent more micro-level values like greater application development flexibility, reduced coding, and code reuse (Table 2). RAD tool features are organized into three categories, based on their roles in the overall software development process. The project management class includes features such as configuration and version management and team repository management; the development class includes, among others, interactive debugging, screen layouts and navigation, and GUI code generation; and the modeling class consists of data and business process modeling.
A Field Study of RAD In order to explore the key issues of RAD usage in IS organizations, a field study was conducted in 1996. One RAD tool that incorporates all the functionality typically associated with such tools was selected as the target technology. This tool, currently available in Version 8, provides features such as an OO 4GL, visual programming tools, support for multiple operating system platforms, multitier capability, database independence, and SQL support. A list of all registered licensees of the tool (173 licensees in 39 businesses) was obtained from the vendor of the tool. A total of 36 individuals from 19 organizations responded to a survey constructed to elicit the constructs shown in the two research models mentioned previously, representing a response rate of approximately 21% at the individual level and 49% at the company level. Sample demographics are provided in Table 3. The majority of the respondents are systems developers with significant experience (an average of over 11 years) in software development, which was the intended target respondent class for RAD tools. COMMUNICATIONS OF THE ACM
181
Table 3. Sample characteristics.
The organizations using RAD tend to be reasonably large, with an average of approximately 2,800 employees, approximately 10% of whom work in IS. The bulk of the industries represented are service-oriented, and thus their ability to deploy new IT applications in a timely manner is often a major source of competitive advantage. Iivari points out that some of the early studies of IT process innovations such as CASE tools did not sufficiently ensure reliability and validity of the measures [7]. The constructs in the two research models are thus rigorously operationalized using previously validated scales. As Tables 1 and 2 show, multi-item scales were utilized for all research constructs, except for two developer characteristics: prior mainframe experience and prior client/server experience. These two were measured by survey questions that asked respondents the length of time they worked with both sets of technologies. RAD technology usage was assessed by asking respondents to rate the extent to which they utilized each feature of the tool in software development. Usage was scored on a 7point scale, with the end points anchored at “Not at all used” and “Used whenever appropriate.” Responses on all features belonging to a particular class were averaged to obtain a feature usage score for the class. A similar procedure was utilized for operationalizing the global and specific benefits of RAD. Descriptive statistics for all research variables, as well as scale reliabilities are shown in Table 4. The reliability values for all scales are indicative of high internal consistency. Multivariate procedures were utilized to test both research models. COMMUNICATIONS OF THE ACM
182
Table 4. Descriptive statistics.
Findings The data reveals that on average, developers have neither overwhelmingly negative or positive perceptions of the RAD tool. Recall that the first research model identified the systems developers most likely to be most receptive to RAD and to transition from old to new skills, and addressed how managers desiring to implement RAD should position it to developers. A multivariate analysis of variance (MANOVA) procedure was utilized with developer characteristics and implementation characteristics as independent variables, and the three salient perceptions about RAD as dependent variables. The overall F-test for the MANOVA was significant; univariate tests revealed that the personal innovativeness of an individual developer in the domain of information technology was a significant determinant of all three perceptions, while prior client/server experience had a negative effect on perceptions of compatibility (Table 5). Among the implementation characteristics, the scope of the change had a positive effect on all three perceptions. The second research model was also tested using a MANOVA procedure, because of the expectation that global benefits and specific benefits are likely to be correlated. Although the overall relationship between the two benefit categories and the three feature sets was significant at p < 0.1, a significant coefficient was obtained only for the usage of development related features as a predictor of global benefits (ß = 0.379, t-value = 2.21, p < 0.05). Discussion Although not reported here, several studies examining technology acceptance indicate that perceptions about the RAD tool were significant predictors of usage intentions. Clearly, management needs to focus on developing positive perceptions about the new technology to ensure its successful adoption into the IS organization’s tool kit. Two developer characteristics among the determinants of perceptions—prior client/server experience and personal innovativeness—emerged as significant predictors. An encouraging finding was the absence of a relationship between prior mainframe experience and perceptions about RAD—this relationship might have been COMMUNICATIONS OF THE ACM
183
Table 5. Influences on developers’ perceptions about RAD.
predicted as negative because of the potential difficulties associated with moving to a new software construction paradigm. But experienced mainframe programmers do not appear skeptical with regard to the value of RAD. A negative relationship was found to exist between prior client/server experience and perceptions about RAD, perhaps because the prior client/server experience of these developers was primarily with non-OO technologies (that dominated early client/server implementations), and not with the OO paradigm. Finally, the positive relationship between personal innovativeness toward information technology and RAD perceptions is encouraging. Consistent with Martin’s conceptualization of the mix of IT professionals in any organization, the “experimenters” and “early adapters” are more likely to have positive perceptions about RAD technology [9]. Results for the speed of change implementation and scope of change suggest that it does not appear to matter how quickly RAD technology is brought into the organization—what is critical is how the technology is presented and positioned for the target developers. More radical changes were found to engender more positive perceptions, a finding that supports the commonly held view that software developers are motivated by work-related outcomes. Perhaps they are so frustrated with the inability of extant development environments to allow them to meet their work-related demands (to develop and deliver systems, and to meet critical deadlines), that the more radical a change, the better its perceived ability to fill the gaps in the current technologies. The relationship between usage of RAD tool features and perceived benefits was examined to identify the source for the gains from RAD. The fact that only development-related features were significantly associated with global benefits was disturbing. If, in addition to speed, RAD tools are used to produce high quality, maintainable, reusable software, shouldn’t the usage of features such as project management and modCOMMUNICATIONS OF THE ACM
184
eling emerge as highly correlated with perceived benefits? One explanation is that these results suggest a subversion of development methodology, and perhaps systems developers are ignoring or side-stepping crucial phases in the development life cycle. Further support for this explanation can be found in the fact that the difference between usage of development features compared with modeling features was statistically significant.1 Systems developers appear to be utilizing the features of the tool that allow them to deliver systems in a speedy manner, but are not fully utilizing the features that contribute to longer-term quality and maintainability of systems. This confirms the supposition articulated in a recent debate on RAD that “most RAD projects skip design and rigorous methodology altogether.”[10] To gain further insight into the underlying issues our findings point to, we gathered follow-up data, with the intent of gaining a rich, qualitative understanding of the RAD phenomenon as observed by systems developers. Particular emphasis was placed on any changes to the systems development process wrought by the technology. Four organizations, Alpha, Beta, Gamma, and Delta were targeted for interviews: two of which utilized the same tool that formed the basis for the survey. In order to ensure the results obtained were not idiosyncratic to the particular tool being studied, two organizations using a different RAD tool for software development were also included. Alpha is a large “information provider” that operates in the financial and insurance sectors. Its primary line of business is providing information about consumers to retail companies. For instance, Alpha provides insurance companies with information about all previous claims made by a new applicant for an insurance policy. Beta is a mid-sized utility company that provides electricity and gas services to approximately one million customers. Gamma is a large investment bank that provides personal and institutional client brokerage services; it is in the top 10 of its kind, ranked internationally. Delta is a large manufacturing organization that develops highly engineered products for a global marketplace. In each case, interviewees were IS managers with direct responsibility for projects utilizing RAD tools. The interviews were structured around their experiences with RAD and their overall impressions of what changes the technology was instrumental in bringing about.
Comments from the Field All four organizations are developing mission-critical applications using RAD tools. Alpha’s major application under development is a system that will support about 400 field offices of insurance companies. Beta is developing a system using a multitier client/server architecture that will support all calls made to the utility during inclement weather. At Gamma, a broker’s workstation that will provide up-to-date information on all client holdings is under construction. Delta is utilizing RAD for applications ranging from support systems for product engineers to billing systems. Managers were consistent in their rationale for introducing RAD: “There is a lot of time pressure to deploy applications quickly,” said Beta managers. The project leader from Alpha noted there was a lot of industry pressure to deploy these tools. The IS manager at Gamma succinctly summarized why they elected to use RAD: “My job is to satisfy client needs in a timely manner. Whatever I can do to meet these needs in a more efficient way, I will.” At Delta, management’s expectations about RAD are 1A t-test of the difference between usage of development features and usage of modeling features was statistically significant (t value = 8.81; p-value < 0.001). COMMUNICATIONS OF THE ACM
185
“unreasonably high.” For example, a project that required about 35 screens, an interface to a Sybase database, and a mainframe Cobol database was estimated to be of medium size by the project team—this estimate was halved by management. Regarding assimilation and implementation of RAD, respondents noted that people with linear C experience had the most difficulty in making the transition. Even developers with client/server experience did not adapt to the tool very quickly, simply because of the paradigm shift from procedural to OO development. The Beta manager explained that he did not feel he could mandate use of the tool, but he was attempting to influence his staffs’ attitude by emphasizing benefits such as its multitier capability, potential impacts on reusability, and database independence. Perhaps the most disturbing aspect of the tools revealed by the interviews, consistent with what the survey data suggested, was the profound impact that RAD tool use was having on development methodology. The Gamma manager was particularly concerned; in his view, the most alarming change he witnessed since RAD’s adoption into his IS organization was the declining emphasis on requirements planning and modeling. “We are already under the gun to deliver applications quickly. Now the tools let you show users something very quick and very dirty—there’s immediately an expectation that the system is ready. There’s very little incentive for the developer now to say let’s go back to the drawing board and see if we have the domain model nailed down accurately. The attitude is that the tools make changes and multiple iterations so easy that errors and mis-specifications can always be caught down the road and fixed. To me this poses a very great danger—how are we going to realize the benefits of reusability if we keep shrinking the analysis phase?” This sentiment was echoed by the manager at Beta, who noted that the rapid prototyping capability of these tools had resulted in the use of a “seat of the pants” methodology. The key concerns about RAD that surfaced through the survey and field interviews are: • Development-related features of RAD tools used more extensively than modeling features; • Declining emphasis on requirements planning and modeling; • Potential subversion of development methodology; major emphasis on system construction, not enough on domain analysis; • Developer tendency to not seek out errors and misspecifications early in the development cycle; • Benefits of reusability cannot be obtained with a shrinking analysis phase; • Unrealistic management expectations about how quickly systems can be constructed using RAD approaches; and • Developers with linear C experience have difficulty making the transition to object-based RAD tools.
Implications and Recommendations Growing evidence supports RAD as an “order of magnitude” improvement in the speed of software construction. Several pragmatic implications emerge from the results of this study and allow us to make recommendations to managers desirous of exploiting RAD as a methodology and as a class of tools (see the sidebar “Recommendations for IS Managers”). First, it is important to recognize that RAD represents a radical change to the process of software construction for most IS organizations. As substantiated by extensive prior work, mandating technology usage is unlikely to produce beneficial results in the long run. Accordingly, careful attention COMMUNICATIONS OF THE ACM
186
must be paid to the developers who are first targeted for RAD and the manner in which the technology is brought into the IS organization. Developers who are personally more innovative with respect to information technology and have greater exposure to OO technology are inclined to respond to RAD with more alacrity. These individuals can serve as key change agents in diffusing the technology more widely. Because such developers are likely to be more recent entrants into the software development world, managers might wish to staff RAD projects with a mix of old and new skills, where the business knowledge and methodology commitment of experienced staff combines and diffuses synergistically with the receptivity to new technology of the newer developers. Managers also need to emphasize the magnitude of the change represented by RAD over previous approaches to software construction, perhaps through the use of information dissemination and training sessions. Training should not focus only on imparting technical knowledge, but on underscoring the radical improvement RAD represents. IS organizations adopt RAD methods and tools in an attempt to realize both productivity and quality benefits. Our results show that managers need to try to avoid the pitfall of allowing RAD capabilities to subvert good development practices. Martin notes that “…higher quality, lower cost, and rapid development, thus, go hand in hand if an appropriate development methodology is used” [9]. In a study of the factors affecting reuse, Frakes and Fox found that a common software process and reuse education were important determinants of reuse, while the specific programming language or CASE tool were not [5]. The dominant driver of an individual developer’s behavior, only exacerbated by the increasing pressure from users, tends to be the next short-term deliverable. Convincing such developers of the importance of up-front domain analysis and modeling represents a critical area for managerial emphasis if RAD is truly going to help alleviate the software crisis. At the very least, the decision by managers to yield to short-term business pressures by sacrificing analysis and using RAD for speed alone (to deliver systems quickly that are intended to be fixed in later iterations)—should be an informed decision, made after consideration of the implications on cost and quality in the longer term. Clearly, this is an issue that merits widespread debate.
References
1. Ajzen, I. and Fishbein, M. Understanding Attitudes and Predicting Social Behavior. Prentice-Hall, Englewood Cliffs, NJ, 1980. 2. Agarwal, R., Sinha, A.P., and Tanniru, M. The role of prior experience and task characteristics in object-oriented modeling: An empirical study. International Journal of Human-Computer Studies 46 (1996), 639–637. 3. Banker, R. and Kauffman, R.J. Reuse and productivity in computer-aided software engineering: An empirical study. MIS Quarterly 15, 3 (1991), 375–401. 4. Creegan, R.W. RAD may be the answer. Computing Canada 20, 13 (June, 1994), 43. 5. Frakes, W.B. and Fox, C.J. Sixteen questions about software reuse. Commun. ACM 38, 6 (June 1995), 75–87. 6. Gallivan, M. J., Hofman, J.D. and Orlikowski, W. Implementing radical change: Gradual versus rapid pace. In Proceedings of the 15th International Conference on Information Systems (Dec. 14–17, Vancouver, BC) ACM/SIGBIT, New York, 1994, 325–339. 7. Iivari, J. Why are CASE tools not used? Commun. ACM 39, 10 (Oct. 1996), 94–103.
COMMUNICATIONS OF THE ACM
187
8. Leonard-Barton, D. and Deschamps, I. Managerial influence in the implementation of new technology. Management Science 34, 10 (Oct. 1988) 1252–1265. 9. Martin, J. Rapid Application Development. Macmillan, New York, 1991. 10. Reilly, J.P. and Carmel, E. Does RAD live up to the hype? IEEE Software (Sept. 1995), 24–26. 11. Price Waterhouse. Technology Forecast: 1996. Price Waterhouse Technology Center, Menlo Park, CA, 1995. 12. Tornatzky, L.G., and Klein, K.J. Innovation characteristics and innovation adoption-implementation: A meta-analysis of findings. IEEE Transactions on Engineering Management EM-29 (Feb. 1982), 28–45
Recommendations for IS Managers Recognize that RAD represents a radical change to the process of software construction for most IS organizations. The development methodology supported by RAD is significantly different from traditional waterfall methods. Emphasize the magnitude of the change represented by RAD through information dissemination and training sessions. Developers appear to respond more positively to RAD if it is positioned as a fundamental change, possibly because they are frustrated by existing development approaches. Allow developers to accept RAD approaches voluntarily and focus on developing positive perceptions about tools. Mandating their use will result in negative attitudes. Select developers for initial RAD usage carefully. Choose those with higher personal innovativeness with respect to information technology or prior OO experience. Staff RAD projects with a mix of experienced and relatively new developers. Experience with methodology resident in the senior staff can be leveraged with the more contemporary technology experiences of the new staff. Avoid the pitfall of permitting the capabilities of RAD tools to subvert good development practices. Carefully weigh the long-term implications to cost and quality.
COMMUNICATIONS OF THE ACM
188
