The Business Imagination Problem: Why Companies Can't Think Outside the AI Box

There is a pattern in how most organizations adopt AI that limits the value they get from it dramatically. They take their existing processes — the way work has always been done — and ask: where can we add AI to this? The question sounds sensible. It is, in most cases, the wrong question. The right question is: if we were designing this process from scratch today, knowing what AI can do, what would it look like? The organizations that ask this second question and do the hard work of answering it are seeing genuine transformation. The organizations that ask the first question — which is most organizations — are spending money on AI that underperforms because it was never given the chance to do what it actually does well. **The Spreadsheet Analogy** When spreadsheet software arrived in the early 1980s, the first generation of business adopters used it to produce the same reports that had previously been produced by hand — faster. That was real value. But the organizations that realized the biggest gains were those that recognized that spreadsheet software made possible entirely new analyses that had been impractical by hand: scenario modeling, sensitivity analysis, dynamic forecasting. They did not just speed up existing work. They did work that had never been done before because it had been too expensive. AI is in the same moment. The organizations deploying AI to generate first drafts of the documents that humans have always written, or to speed up the manual review processes that humans have always performed, are gaining real efficiency. The organizations that ask "what analyses, what processes, what products could we create if cognitive work were dramatically cheaper?" are finding the larger opportunity. Most organizations are in the first group. Research from MIT Sloan Management Review and McKinsey consistently finds that the most common AI deployment is what practitioners call "augmentation" — assisting humans in existing tasks — rather than "transformation" — redesigning how work gets done. Augmentation is valuable. Transformation is where the larger gains are. And transformation requires the harder question. **Why Process Imagination Is So Difficult** The inability to imagine different processes is not a failure of individual intelligence. It is a structural feature of how expertise develops and how organizations operate. Domain expertise is largely built on knowing how things work in a particular context. A senior manager in insurance has deep knowledge of how claims are processed, what the regulatory requirements are, what the exception cases look like, and what happens when things go wrong. That knowledge is genuinely valuable — and it is also a set of mental models built on how the process works now. Asking that person to imagine a fundamentally different process requires them to hold their expertise lightly enough to question its underlying assumptions, which is cognitively and emotionally difficult. Organizations also have existing investments — technology systems, trained staff, established vendor relationships, documented procedures — that create lock-in to current approaches. Proposing a process redesign that makes those investments obsolete is politically difficult regardless of its technical merits. The people whose roles depend on the current process are not enthusiastic advocates for redesigning it. The result is that AI deployments in most organizations are constrained by what existing stakeholders can imagine and approve, which is largely the current process with AI inserted at specific points. **What Shallow AI Implementation Looks Like** Shallow AI implementation is recognizable in practice. A law firm adds an AI tool that summarizes case documents — but the workflow for reviewing those summaries is the same workflow that existed before, requiring the same number of attorney hours. A hospital deploys an AI that flags abnormal lab results — but the downstream process for acting on those flags is unchanged, so the alert adds to physician workload rather than reducing it. A retailer implements an AI chatbot for customer service — but the chatbot is trained only on the existing FAQ, which was written for humans browsing a website, not for AI to synthesize and explain conversationally. In each case, the AI is real, capable, and deployed. In each case, the value it delivers is a fraction of what it could deliver if the surrounding process had been redesigned to take advantage of what AI does well. The Harvard Business Review has published extensive research on digital transformation outcomes. One consistent finding: organizations that treat technology adoption as a process transformation initiative, with dedicated resources for workflow redesign and change management, consistently outperform those that treat it as a technology deployment. The technology is necessary but not sufficient. The imagination to redesign the process is the scarce ingredient. **What Genuine AI Integration Looks Like** The organizations that are getting substantial value from AI share a common approach: they work backward from the AI capability to redesign the process, rather than working forward from the current process to find where AI can fit. A legal technology company that redesigned its contract review workflow around AI capability — asking "if AI can read and summarize documents in seconds, what does contract review look like when we accept that premise?" — found that the roles that had previously been focused on reading and summarizing shifted to verification, judgment, and client communication. The same amount of attorney time produced significantly more throughput. The key insight was that this required changing the process, not just adding a tool. A financial services firm that redesigned its fraud investigation workflow around AI-generated case summaries and risk scores found that investigators spent dramatically less time on data gathering and dramatically more time on judgment calls about borderline cases. But this required the firm to redesign what investigators did, how their performance was measured, and how cases were routed — not just to give them an AI summary at the top of their existing workflow. **The Opportunity for Technologists** The business imagination problem creates a specific and valuable opportunity for technologists who understand both the AI capability and the business process. The ability to look at a business process, identify where AI could genuinely change its economics and outputs rather than just speed up one step, and then help the organization design and implement that change — is a combination of skills that very few people currently have. It requires technical knowledge: what AI can actually do reliably, what it cannot, what the cost structure looks like, how to build a system that works in production. It requires business knowledge: how the current process works, who the stakeholders are, what the actual objectives are that the process is serving. And it requires change management knowledge: how to propose and implement a process change in a way that brings people along rather than creating resistance. This combination is rare. It is also what organizations that want to move from pilot to transformation need most. The technologists who develop it — by deliberately engaging with the business context of AI deployment, not just the technical context — are positioning themselves for some of the highest-value work in the current transition.