Enterprise Design Thinking was developed at IBM from 2012 by Phil Gilbert, then General Manager of IBM Design. The problem it was created to solve was specific and real. IBM’s product development teams were building things at scale that missed the market. The complexity of large, geographically distributed teams meant that user needs got lost in the development process long before a product reached the people it was supposed to serve. EDT’s tools — Hills to align teams around user outcomes, Playbacks to maintain continuous feedback, Sponsor Users to keep real users embedded in the process — were designed to prevent that drift. It is a legitimate methodology for that problem, and IBM’s design transformation produced measurable results in the product development context it was designed for.
The problem is not with EDT. The problem is with what it has been applied to. Over time, EDT migrated from an internal product development methodology into a client-facing discovery tool, used to help organisations identify opportunities and set strategy. In the AI context specifically, it is now routinely applied to one of the most consequential strategic questions an enterprise faces: where should we invest in AI and what should we expect it to deliver. For that purpose, EDT is structurally the wrong tool. Not because it is poorly designed, but because it was designed for a different problem.
What EDT was built to answer and what AI strategy actually requires
EDT is built around a central question: what do users need, and is what we are building serving those needs? Every practice in the framework serves that question. Empathy maps capture how users think and feel. Hills express goals in terms of user outcomes. Sponsor Users bring real users into the development process. The methodology is optimised for understanding and responding to human experience, which is exactly the right thing to optimise for when you are building a product or service that people have to find worth using.
AI strategy requires answering a different question entirely. Not what do users need, but which decisions in this organisation are being made badly, at what cost, and whether AI intervention at the right point in the decision process would change the outcome. These are questions about business performance, decision quality, and economic value. They are not questions about user experience. The people who need to answer them are not looking at how users feel about a product. They are looking at where outcomes are falling short of targets, what that shortfall costs, and whether the root cause is a decision that AI can improve.
Applying EDT to that problem produces output optimised for the first question while the second question goes unanswered. The organisation learns something about stakeholder sentiment and user experience. It learns nothing systematic about decision quality or economic opportunity.
The peer group problem is structural
Even setting aside the question mismatch, the workshop format at the centre of EDT creates a specific and unavoidable problem for AI strategy discovery. EDT sessions surface what is on participants’ minds in a group setting. What is on participants’ minds in a group setting is shaped by social dynamics that have nothing to do with analytical rigour.
The person whose area is responsible for a significant operational failure will not volunteer that information in a room with peers and senior stakeholders present. The person who has been pursuing a particular technology agenda will frame their contributions to advance that agenda. The person with no strong prior view will find a position that feels safe relative to the room. Hopes exercises — asking people what they hope this initiative will achieve — are particularly prone to this dynamic because hope is a social performance as much as a genuine expression of priority. People express hopes that are acceptable, that reflect well on them, and that do not expose vulnerabilities they are not prepared to discuss in a group setting.
Real discovery of AI opportunity requires access to information that people are often reluctant to share in groups: where performance is falling short, where decisions are being made badly, and what those failures actually cost the organisation. That information exists in operational data and in honest one-to-one conversations with people who understand their area well enough to identify its specific failures. It does not typically surface in a facilitated group session, however skillfully the session is run.
This is not a criticism of EDT facilitation. IDEO’s own founders, when consulting with IBM during the development of EDT, acknowledged that design thinking “works great when you’re doing it in these small workshops at Stanford, but it starts to break apart and fail at scale.” The group workshop format has inherent limitations that scale compounds rather than resolves. IBM’s response was to develop practices that partially mitigate those limitations in the product development context. For AI strategy discovery, where the most important information is precisely the information people are least likely to share in groups, the mitigation is insufficient.
Nothing comes out quantified
The most direct test of whether a discovery process is adequate for AI strategy is simple. Does the output contain numbers. Specifically, does it contain a quantified cost of the current decision quality problem, a quantified estimate of what improvement would be worth, and a basis for prioritising one opportunity over another in financial terms.
EDT output does not typically meet this test. The Hills that emerge from an EDT session express user outcomes in qualitative terms. A Hill describes what a user should be able to do, not what it costs the business when they currently cannot do it. The themes and insights that come out of empathy exercises are observations about experience, not estimates of economic value. And because the methodology is not designed to produce quantified output, the facilitation practices do not drive toward it. The session ends with a richer understanding of stakeholder perspectives and a set of opportunity directions. It does not end with an investment-ready analysis of where AI will move the needle and by how much.
For AI strategy, that gap is not a minor limitation. An AI investment decision made without an economic baseline has no reliable way to measure its own success. The programme that follows cannot be defended in financial terms because the financial case was never established. And the prioritisation of which opportunities to pursue first is driven by enthusiasm and visibility rather than by value, which is exactly the dynamic described in the previous piece in this series on use cases.
EDT is a rigorous methodology for the problem it was designed to solve. Applied to AI strategy discovery, it is the wrong tool, and the organisations that rely on it to set their AI strategy are substituting a structured conversation about user experience for the analytical work that investment decisions of this scale actually require.
