70% of Digital Transformations Fail and We Keep Studying Adoption

McKinsey research puts the failure rate for major transformations at roughly 70%, and only 16% of digital transformations successfully improve performance and sustain the improvement over time. I keep thinking about this while reading IS research on technology adoption.

The 70% figure has circulated long enough that it has become a piece of business folklore, repeated in consultant decks and board presentations without much examination of what it actually claims. But the 16% number is more specific and more useful. Not just "does the transformation succeed or fail" but "does it improve performance AND sustain the improvement." That is a higher bar, and it is the right bar. A transformation that improves performance for one quarter and then reverts is not a transformation. It is a project with a good launch quarter.

The IS field's contribution to understanding these failures is substantial but unevenly distributed. Technology Acceptance Model, developed by Davis (1989), and its many extensions give us precise predictive power over whether individuals will intend to use a new system. Perceived usefulness and perceived ease of use explain adoption intention well across a wide range of systems and populations. Decades of research have refined the model, added external variables, and improved measurement. TAM is genuinely good at what it does.

DeLone and McLean (1992, 2003) pushed further. Their IS success model connects system quality, information quality, and service quality through use and satisfaction to net benefits. The model traces the whole lifecycle from system characteristics to organizational outcomes. It is more comprehensive than TAM and it explicitly includes the outcome side of the equation: net benefits, which can be individual, organizational, or societal. In principle, this should let us study transformation success, not just adoption.

And yet the 70% failure rate persists alongside a research literature that is much richer in adoption findings than in transformation outcomes. I think the reason is structural, not accidental. Both TAM and the DeLone and McLean model are designed to study system-level phenomena. They ask questions about a specific IS artifact: is this system used? Is its information quality sufficient? Does it produce net benefits? Transformation is not a system-level phenomenon. It is an organizational change process in which multiple systems, processes, roles, and structures are simultaneously altered. The unit of analysis shifts from the system to the organization, and that shift is more significant than it appears.

The IS field has frameworks that operate at the organizational level: structuration theory (Giddens 1984, applied to IS by Orlikowski 1992), dynamic capabilities (Teece et al. 1997), absorptive capacity (Cohen and Levinthal 1990). These frameworks can in principle say something about transformation. Orlikowski's enactment perspective, for instance, argues that technology does not have fixed properties but is enacted differently depending on the structures people bring to it and the structures they create through use. That is actually a very useful lens for understanding why the same ERP system transforms one organization and fails in another. But these frameworks are harder to operationalize for survey research, harder to generalize across contexts, and produce fewer testable hypotheses than TAM. The incentive structure of IS research pushes toward the operationalizable.

I wrote earlier about how the IS success model struggles with modern systems where use is ambient, information is AI-generated, and net benefits are distributed across misaligned stakeholders. Transformation is a harder version of the same problem. You cannot measure transformation success by aggregating individual adoption decisions, because transformation requires changes to routines, incentive structures, power distributions, and business model logic that individual adoption decisions do not capture. People can adopt a CRM system without the sales organization transforming how it sells. They can use an analytics platform without the company becoming data-driven. The adoption happens. The transformation does not.

This is the gap the McKinsey data is pointing at. The 84% of organizations that do not sustain performance improvement after transformation have presumably adopted the technologies involved. Adoption was not the constraint. Something else failed, and my read is that the something else is the organizational change problem that IS research is equipped to name but not always to measure.

One candidate explanation comes from institutional theory. The same mimetic pressure that pushes organizations to announce AI strategies and zero-trust roadmaps also pushes them to announce transformations. A transformation is a legitimacy signal in the current organizational environment. McKinsey's own transformation consulting practice is part of the ecosystem that makes "transformation" the vocabulary organizations reach for when they want to signal strategic ambition. The result is that some organizations announce transformations and pursue technology adoption without the harder work of changing the routines, roles, and power structures that determine whether the technology produces different business outcomes. The adoption was real. The transformation was rhetorical.

I want to be fair to TAM and to adoption research here. Predicting whether people will use a system is not a trivial question and it matters for practice. Systems that are not used do not produce benefits regardless of their quality. The adoption prediction problem is a prerequisite for the transformation problem. But the IS research community has treated the prerequisite as if it were the full problem, and the McKinsey failure rate data is one of the clearest signals that it is not.

What would IS transformation research look like if it were designed to address the 70% figure directly? My guess is that it would focus less on system-level constructs and more on organizational routines, specifically on what changes in how decisions are made, how information flows, and how accountability is assigned after a new system is implemented. It would probably look more like organizational behavior research than traditional IS research in its methods. And it would require longer time horizons than most IS studies use, because the McKinsey finding is specifically about sustaining improvement, which cannot be measured in the 6-month or 12-month windows that most IS field studies use.

The Verizon 2024 Data Breach Investigations Report found that 90% of breaches are financially motivated and 68% involve the human element. The McKinsey data finds that 84% of digital transformations fail to sustain improvement. These are different domains but the structural resemblance is hard to ignore. In both cases, the technology alone is not the constraint. The organizational factors, how people use systems, how decisions are made about security, how transformation is governed and sustained, are where the outcomes are actually determined. IS research is good at studying the technology side. The organizational side is where the explanatory gap is widest.