Master Thesis: The conditional case for customer digital twins
What a master's thesis found when it asked whether customer digital twins actually change how companies decide
We spent part of this spring supporting a master's thesis at Copenhagen Business School, written in the MSc program in Strategy, Organization & Leadership and supervised out of the Department of Digitalization by Arisa Shollo and Panagiotis Keramidis. The thesis was submitted in May 2026. The research question was narrower and more useful than most of what gets written about this technology: how do customer digital twins impact strategic agility in organizations?
That framing matters. Almost everything published on customer digital twins so far argues from capability. The system can simulate a segment, so it must be valuable. This thesis argued from consequence instead, using Doz and Kosonen's three dimensions of strategic agility as the lens: strategic sensitivity, leadership unity, and resource fluidity. The empirical base was 18 semi-structured interviews with developers, organizational users, and AI experts across industries, out of roughly 250 people contacted. We were one of the organizations that participated.
Three mechanisms
The findings identify three distinct paths by which a customer digital twin appears to affect how an organization responds to change.
The first is insight generation. Twins broaden and accelerate customer understanding, which surfaces relevant developments earlier and allows more iterative exploration of options. This is the mechanism everyone expects.
The second is the one we found most interesting, because it is organizational rather than analytical. A twin functions as a shared reference point. It moves customer knowledge across functions that would otherwise hold it separately, and it objectifies debate. Disagreements that were previously contests between competing intuitions become arguments about a common artifact. The thesis links this to faster alignment in decision processes, which is a claim about internal politics as much as about data.
The third is experimentation. Faster and lower-risk testing lets teams prioritize earlier and redirect resources more deliberately, before commitments become expensive to reverse.
The conditional finding
The part of the thesis worth reading closely is the argument that none of this is automatic. The authors identify trust and continued human oversight as cross-cutting conditions. If decision-makers do not regard the outputs as credible, comprehensible, and relevant, those outputs simply do not enter the decision process, regardless of how good the underlying system is. Weak data foundations, bias, hallucinations, opacity, and privacy concerns all undermine that trust, and adoption does not build it. Repeated bounded use, validation against real customer data, and explicit governance do.
The practical implication follows directly. Organizations that treat a twin as a departmental analytics tool get departmental analytics value. The agility effects show up when the twin is positioned as cross-functional decision-support infrastructure and used early, for exploration and option comparison, rather than late, as a validation stamp on a decision already made.
Where the limits are
One finding deserves more attention than it will probably get. The interviews consistently suggested that customer digital twins are good at evaluating plausible options and refining evolving ideas, and considerably weaker at independently generating radically novel ones. That is a real boundary and worth stating plainly. A twin sharpens the option set you bring to it. It does not replace the strategic reflection that produces the option set in the first place.
The study is exploratory and cross-sectional, based on perceptions rather than observed outcomes, and drawn from a sample tilted toward people already engaged with the technology. The authors say all of this themselves. It is still the most structurally honest treatment of the category we have read, largely because it declined to argue that the technology is valuable and asked instead under what conditions it becomes valuable.
That is the right question, and the answer is not flattering to anyone selling certainty.
