What the research actually shows.

Why large IT programmes fail — an analysis of the research.

Core thesis

With large IT programmes, the norm is not the plan but the deviation from the plan. And where the studies ask about causes, they point predominantly to steering, coordination and decision paths — not to the technical implementation. Three methodologically very different studies arrive at the same qualitative finding across twelve years, even though their figures cannot be aggregated into a single rate.

Source base

The analysis draws on three published studies that are to be contextualised differently by survey year and methodology. The most recent is the BCG study "Most Large-Scale Tech Programs Fail" from 2024, a survey of more than 1,000 companies in 59 countries. Considerably older, but still influential today, is the study by Bloch, Blumberg and Laartz (McKinsey & Company with the University of Oxford, 2012) based on 5,400 IT projects: large IT programmes exceed their budget by 45 per cent on average and deliver 56 per cent less value than planned. The Standish CHAOS report 2020 ("Beyond Infinity") classifies 31 per cent of the projects studied as successful, 50 per cent as "challenged" and 19 per cent as failed. That two of the three sources are older than ten and more than five years respectively is part of the honest contextualisation — the 2024 BCG survey shows, however, that the underlying pattern has not dissolved since then.

Analysis

Together, the three studies show two things. First: the norm for large IT programmes is not the plan but the deviation from the plan. Whether one draws on the Standish classification (barely a third successful) or the McKinsey/Oxford figures (systematic budget and value deviations) — the expectation that a large programme stays within time, budget and scope without active countersteering is not empirically supported. Second: where studies ask about causes, steering topics dominate. In the BCG survey, more than 60 per cent of programmes cite the absence of an end-to-end master plan with a critical path and clear dependencies as a central cause; 60 per cent point to the absence of an active PMO that monitors value delivery and identifies emerging risks.

The differences are as revealing as the commonalities. The three sources do not measure the same thing: Standish classifies project outcomes according to its own, repeatedly criticised concept of success (time, budget, satisfaction), McKinsey/Oxford quantifies deviations against planned values, BCG collects companies' self-reported accounts of causes. A "failed" project at Standish is not the same as a "failed program" at BCG. The figures therefore cannot be aggregated into a single failure rate. What is robust is not the individual percentage but the convergence of direction: three methodologically independent approaches arrive at the same qualitative finding across twelve years.

Contextualisation by loumiTECH

Beyond the pure analysis, three recurring patterns can be synthesised: a missing master plan with explicit dependencies, a missing active PMO with a mandate to countersteer, and — as a cross-cutting topic that Gartner puts at an average of USD 12.9 million in costs per organisation per year for the reporting context — inadequate data quality as the foundation of every steering decision. All three patterns concern the same question: does the programme have the steering building blocks that its actual complexity demands? This is precisely the question the Modular Project Management Blueprint addresses, selecting governance, reporting, decision, coordination and delivery building blocks based on the specific project situation instead of presupposing a universal method.

Limits of this analysis

Three limits are to be stated explicitly. First: this review is based exclusively on published research by third parties; loumiTECH has conducted no independent empirical study. Second: the selection of sources is not systematic in the sense of an academic literature review — publication bias is possible, since spectacular failure figures are published and cited more frequently than unremarkable successes. Third: two of the three sources come from consultancies (BCG, McKinsey), which have a commercial self-interest in the diagnosis "programmes fail on steering" because they sell steering services; the Standish Group also markets its CHAOS data commercially. The findings are therefore to be read as convergent indications, not as a neutral measurement.