Digital transformation is an operations problem, not a software problem

Updated 5 June 2024 — added 2023 data on employee change fatigue and what it implies for sequencing.

Digital transformation programs seldom fail because the wrong platform was chosen. They fail because nobody owns the process the software is meant to run, and because the target was never written as a number. BCG studied 895 transformations and found that 30% met their objectives. The rest fell short.

Key takeaways

• The "70% of transformations fail" figure traces to a May–June 2000 Harvard Business Review article that asserts it with no citation, sample or method.
• A 2011 review in the Journal of Change Management examined five published instances of that number and found no valid and reliable empirical evidence behind it.
• BCG's October 2020 study of 895 digital transformations placed 30% in the win zone, 44% in the worry zone and 26% in the woe zone.
• Five of BCG's six success factors describe operations; only the sixth is a platform decision, and BCG frames even that one as business-led.

Where the 70% failure figure comes from

Every transformation deck cites it; almost none cite a study. The trail ends at "Cracking the Code of Change", Harvard Business Review, May–June 2000. There, Michael Beer and Nitin Nohria assert that about 70% of change initiatives fail. No dataset, sample or method is attached.

Michael Beer and Nitin Nohria state that "the brutal fact is that about 70% of all change initiatives fail" in Cracking the Code of Change, Harvard Business Review, May–June 2000. The article gives no sample, no method and no citation for that figure.

Mark Hughes tested the claim instead of repeating it, and found nothing underneath it.

Mark Hughes reviewed five published sources for the 70% change failure rate. He concluded there is "no valid and reliable empirical evidence to support such a narrative" (Journal of Change Management, 11(4), 451–464, 2011).

A failure rate with no method behind it cannot tell you what to do differently. It only tells a steering committee to be afraid.

What the measured studies actually found

In October 2020 BCG published results from 895 digital transformations: 70 programs it had supported directly and 825 executives surveyed. It placed 30% in what it called the win zone, 44% in the worry zone and 26% in the woe zone.

The headline lands near the folklore number, but the shape underneath differs. Most programs did not collapse; they produced some value and missed the target they were approved against. That is a governance outcome, not a software one.

BCG also named six factors and reported that companies addressing all six moved their odds to roughly 80%, while addressing three or four did not move them at all. Five are operational: quantified strategy, leadership commitment through middle management, talent, agile governance and monitoring against defined outcomes. The sixth, a modular data and technology platform, BCG describes as business-led.

Ownership is the variable, not the vendor

In the May–June 1995 issue of Harvard Business Review, John Kotter reported on more than 100 companies attempting major transformation. Well over half failed in the first phase—establishing urgency—long before a system was chosen. Twenty-six years on, the questions that decide whether a project fails still get asked after the contract is signed.

The pattern repeats in the briefs we receive. A company buys a CRM to fix a sales process nobody has written down, or a CMS to fix an approval chain with no approver. The tool encodes that ambiguity and makes it permanent.

The question that predicts the outcome is not which platform, but who decides when the process and the tool disagree. If the answer is a committee, the old process wins by default and the new platform becomes a reporting layer nobody trusts.

Write the target as a number before the shortlist

Four lines should exist on one page before any vendor conversation: the process being changed, the person accountable for it, today's baseline metric, and the target with its measurement date. This is also what makes a brief usable, and replacing a wish list with decision criteria changes what a supplier can propose and what a board can approve. Selection then becomes a constraint problem rather than a preference problem, which is the honest way to choose between Jamstack and WordPress.

What changed in this update

The operational side of this argument now has a number. Gartner research reported in Harvard Business Review in May 2023 found that employees' willingness to support enterprise change fell to 43% in 2022, from 74% in 2016. The average employee absorbed 10 planned changes that year, against two in 2016. Sequencing is therefore part of the design. Four programs at once stack the fatigue rather than divide the risk, a constraint that also shaped the first production deployments of generative AI.

FAQ

The order that matters

Write the target before the RFP, name one accountable owner before the kickoff, and run one program at a time. Software selection is the last decision in the sequence, not the first. Six months in you will know whether the baseline metric moved, and if nobody can say what the baseline was, that is already the finding.