Scrum Leads the Charts, but Evidence Lags: What 54 Studies Reveal About Software Models in Developing-Country Universities

Choosing a software development model for a university digitization project is never a purely technical decision, but in developing countries the stakes are unusually high. A new systematic literature review, published open access in Discover Education on 27 October 2026, maps which software development models have actually been reported in academic digitization projects across higher education in developing countries between 2013 and 2025, and the picture that emerges is one of strong reporting preferences paired with strikingly weak outcome evidence.

The review, led by Janvier Twizerimana Sindambiwe with colleagues, screened 239 records retrieved from six primary databases—IEEE Xplore, Scopus, ScienceDirect, SpringerLink, DOAJ and ERIC—plus Google Scholar as an additional search engine. After screening and a structured quality assessment, 54 primary studies met the inclusion threshold of at least 2.0 on a four-point quality scale. The authors applied a PRISMA procedure: 88 records were removed before screening (40 duplicates and 48 by automation), 151 reports were sought, 52 could not be obtained, and of 99 full-text reports assessed for eligibility, 45 were judged irrelevant. The final corpus is heavily weighted toward recent work, with 42 of the 54 articles—77.8 percent—published between 2020 and 2025, and 10 documents appearing in 2023 alone.

What did those 54 studies report? Agile approaches dominate the frequency counts, accounting for 33 occurrences, or 61.1 percent of the twelve software development models identified. Within the agile family, Scrum is the most frequently reported model, appearing in 13 studies (24.1 percent), followed by DevOps with 7 occurrences (13.0 percent) and Kanban with 6 (11.1 percent). Lean (3), Incremental (2), XP (1) and RAD (1) trail well behind. Traditional models appear in 9 cases (16.7 percent), with the V-Model leading at 4 occurrences, Waterfall at 3, and the Spiral and W-Models at one each. Hybrid methods account for 12 occurrences, or 22.2 percent. The authors supplement these counts with a VOSviewer co-occurrence analysis of keywords from the primary studies, in which ten keywords met a minimum occurrence threshold of five and formed two clusters: one centered on the academic domain (higher education, digital transformation, universities, students) and one on the software engineering domain (software development, software engineering, countries).

The most consequential negative finding concerns artificial intelligence. Not a single study in the corpus documented AI-assisted or AI-based software development as an observed model in academic digitization projects in developing countries. The review’s authors are careful about what this means: AI-based development is treated as a research gap and an emerging strategy, not as an empirical discovery. They explicitly separate what the examined literature actually contains from broader technology trends, noting that the absence signals a continued lack of research on AI-supported development in this specific context rather than any evidence about its performance.

Equally important is what the reviewed studies failed to measure. The authors state plainly that reporting frequency does not indicate superiority or efficacy. The primary studies did not record cost, delay, success or abandonment data, so the prominence of Scrum in the literature reflects study focus and adaptability of iterative delivery, not demonstrated performance. As the discussion puts it, a frequency-based approach does not capture project success rates, customer satisfaction, scalability or implementation difficulties, and differences in project size, context and resource availability further limit direct comparisons between models. The results, the authors insist, should be read as indicators of adoption patterns, not proof that any model outperforms another.

To move beyond frequency counting, the review proposes a Context-Sensitive Software Development Model Selection Framework, built through cross-case thematic synthesis of the corpus. The framework links six suitability criteria—flexibility, traceability, cost, risk, stakeholder participation and sustainability—to four contextual dimensions: infrastructure (hardware, connectivity, electricity availability), human capability (developer skills, user digital literacy, stakeholder availability), resources (financial, technical and human) and the institutional environment (bureaucracy, legal requirements, documentation requirements). The framework does not prescribe a single model. Instead, it suggests that agile approaches may be considered where adaptability and stakeholder participation matter and sufficient human and infrastructure capacity exists; traditional approaches may be relevant where formalization, traceability and documentation are paramount; and hybrid approaches may offer a balance when both flexibility and formal structure are needed.

The thematic synthesis behind the framework surfaced recurring tensions in the reviewed studies. One theme concerns documentation versus agility: although agile models encourage flexibility, their limited emphasis on formal documentation creates difficulties in academic settings where traceability and compliance are required. Another concerns development velocity versus infrastructure: the reviewed studies identify infrastructure as a significant contextual factor shaping how development methodologies play out in resource-limited environments. A third links digital literacy to stakeholder engagement: low digital literacy may hamper effective participation in iterative development activities, even though meaningful stakeholder participation is a prerequisite for agile methods—suggesting that users’ capacity to engage may matter as much as the chosen model itself.

Notably, the authors flag which claims rest on direct evidence and which do not. They state that the corpus does not offer enough evidence to prove that sprint interruptions in academic software projects in Goma are specifically caused by erratic connectivity and inconsistent power supply; the possible impact of such constraints on iterative development is presented as a contextual implication for the Goma environment that requires particular empirical confirmation, rather than as a direct result of the cross-case synthesis. Similarly, the digital literacy and stakeholder engagement association is treated as a contextual implication rather than a direct finding for academic institutions in Goma, pending empirical validation. This level of epistemic caution is unusual in review literature and worth noting for anyone citing the work.

The study carries substantial limitations, which the authors document in detail. Only one researcher performed the screening, full-text evaluation, extraction and quality assessment, so there was no inter-reviewer disagreement process and no Cohen’s kappa; intra-rater consistency was instead checked by re-evaluating all studies after two weeks. Database coverage may have missed pertinent unindexed research, language bias excluded non-English papers, and publication bias may have favored studies with successful outcomes. The diversity of circumstances, approaches, organizations and projects hampers direct comparison, classification of models as agile, traditional or hybrid may be subject to classification bias, and geographic coverage was uneven with little data from some regions. The authors also note that restricting the window to 2013–2025 limited the available literature, and that the six core databases plus Google Scholar may not be exhaustive. They conclude that the findings should be seen as a synthesis of available research from 2013 to 2025 rather than broadly applicable conclusions.

The practical upshot is a call for context-driven rather than popularity-based model selection. Because agile and DevOps approaches rely on frequent communication, iterative development, testing, integration and deployment, unstable electricity, poor hardware and restricted connectivity can disrupt these operations and erode the benefits of continuous delivery. Resource availability constrains whether institutions can sustain development teams, tools, training and long-term system support, while Scrum demands effective coordination and ongoing stakeholder involvement and DevOps requires technical expertise and appropriate infrastructure. Conventional methods remain useful where formal documentation and regulatory constraints dominate, and hybrid techniques may serve institutions that must reconcile flexibility with formal control. The proposed framework is explicitly a conceptual decision-support model derived from the synthesis, not empirical proof of any model’s superiority, and its effectiveness requires validation in specific resource-constrained academic settings. The authors direct future research toward four areas: developing context-sensitive success metrics, examining organizational and cultural factors, empirically validating hybrid models in African contexts, and assessing the long-term sustainability of digitization in resource-limited environments. For universities in developing countries weighing a digitization roadmap, the message is that the most-reported model is not automatically the best fit—and that the evidence base for choosing one remains thinner than the literature’s volume suggests.

Subject of Research: Software development models used in academic digitization projects in higher education in developing countries

Article Title: A systematic literature review on software development models for academic digitization in developing countries

Article References: Sindambiwe, J. T., Mushage, O. B., Wanguwabo, D. B., Ntayagabiri, J. P., & Ndikumagenge, J. (2026). A systematic literature review on software development models for academic digitization in developing countries. Discover Education, 5(1), Article 1233. https://doi.org/10.1007/s44217-026-02315-x

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02315-x

Keywords: systematic literature review, software development models, Scrum, DevOps, academic digitization, developing countries, higher education, agile methods, hybrid approaches, digital literacy, research gap, PRISMA