AI Transformation: Beyond Tools to True Business Reinvention

The AI Transformation Paradox: Why Tools Alone Won't Save You

Every large enterprise today runs on a sprawling, often undocumented patchwork of software—SAP and Workday monoliths, hundreds of vertical SaaS apps, and countless Excel spreadsheets that quietly run critical departments. This is the messy reality Benedict Evans describes in his recent essay, 'AI, Tools and Transformation.' The promise of AI is intoxicating: instead of building tools one by one, you can simply ask a model to do the task itself. But as Evans and others argue, this vision misunderstands how software actually gets adopted and how companies genuinely change.

The hard part was never writing the code. It's knowing a problem exists in the first place, and then designing a solution that fits into existing workflows. As Evans points out, a great matrimonial lawyer thinks about clients, not legal discovery software. The 'forward-deployed engineer'—someone who can walk into a law firm and spot automation opportunities—is a real role, but it's rare. Most people, even with powerful AI tools, don't instinctively see how their jobs could be done differently.

The Institutionalized vs. Improvised Spectrum

Software adoption follows a spectrum from top-down institutionalization to bottom-up improvisation. SAP, Workday, and Carta represent institutionalized tasks—processes that have been standardized, audited, and secured. Meanwhile, Excel, email, and shared folders form the 'freeform substrate' where users improvise solutions for edge cases and exceptions. This is where AI is already having its most immediate impact, expanding the capabilities of these improvised spaces.

But when an improvised workflow becomes critical, it must be institutionalized. You need audit trails, security, and accountability. This is why companies end up with hundreds of apps—each one represents a paved desire path. AI doesn't eliminate this cycle; it shifts the thresholds. A small company might stick with Google Sheets longer because AI makes it more scalable, but eventually, a new SaaS app will appear that solves the problem more elegantly. The question remains: buy, build, or improvise?

Why Enterprise AI Pilots Often Stall

Three years into the enterprise AI experiment, the pattern is clear. Every big company gave everyone Copilot or ChatGPT, and a small percentage of employees use it heavily, while a larger group uses it occasionally, and many barely use it at all. This mirrors the rollout of PCs in 1983 or web browsers in 1997. Giving everyone a tool doesn't transform the company; it just gives people a new way to do the same things.

The typical response is to run pilots—targeted experiments with new AI capabilities. But as Evans notes, roughly half of these pilots fail, which is normal. The problem is that CEOs and boards look at 500 workflows and see only five successful pilots. That doesn't scale. The real challenge is moving from individual experiments to company-wide decisions, which requires a governance structure that can test, approve, and scale what works.

The Three Questions Every Company Must Ask

Evans proposes three fundamental questions for any transformative technology. First, how do we buy, build, and deploy this? Do we take the bundled product from Microsoft or Google, build internally, or buy from a startup? Second, how does this change our operations? The answer differs radically for an insurance company versus a law firm. Third, does this create new competitive threats or existential risks? These questions can't be answered by giving everyone Claude for X.

This is where professional services firms come in. Accenture, the Big Four, and McKinsey are all positioning themselves as AI transformation partners. Ironically, AI poses as many questions to their own business models as it does to their clients'. The forward-deployed engineer is essentially a systems integrator hired by OpenAI. The market is consolidating around a familiar pattern: vendors build, integrators deploy, and consultants advise.

From Automating Legacy to Reinventing Operations

Telefónica's guidance cuts to the heart of the matter: 'The most common mistake is to automate the legacy process.' The right question isn't 'how can I do what I already do faster?' but 'how would I do it today if the company were founded with AI?' This requires mapping the work—the tasks repeated a thousand times a day—not the organizational chart. You must design the human + AI architecture: where the person decides, where the machine suggests, and where traceability lies.

If the process remains the same and only the tool changes, there's no transformation. The future belongs to those who can reinvent their organization around AI, not just adopt it. This means creating a 'center of excellence' that gives AI adoption a working rhythm, as the CIO article suggests. Each team needs someone close enough to the work to spot where AI is useful and where it's a distraction. This person can also prevent 'shadow AI'—the unauthorized use of tools that leads to data leaks and compliance violations.

Leadership, Culture, and the New Operating Model

ARC Advisory Group emphasizes that technology alone won't determine success. The organizations that create lasting impact will be those that rethink how decisions are made, how people are empowered, and how business processes are designed. Leaders must create environments that encourage experimentation, empower teams closer to the point of action, and establish clear guardrails for responsible decision-making. This is a cultural shift as much as a technical one.

The technology review highlights another critical factor: time to market. AI can accelerate legacy modernization dramatically—one case study showed a 60% reduction in delivery time compared to the pre-AI era. This speed is a competitive advantage, but it also raises the stakes. Companies that move fast can leapfrog competitors, but they also risk automating flawed processes. The key is to combine speed with a willingness to rethink the underlying workflows.

The Bottom Line: AI Is an Operating Model Shift, Not a Software Upgrade

AI will not sweep away the messy reality of enterprise software. It will expand existing apps, create new vertical tools, and add a new freeform space—the chatbot—alongside Excel and email. But the fundamental dynamics of how companies adopt technology remain unchanged. You still need to identify the problem, design the solution, and get hundreds of people to change their behavior. AI makes the tool-building easier, but it doesn't make the change management any simpler.

As Evans concludes, we will use AI to automate broad classes of work inside existing companies, but that will be enormously more trouble than just giving everyone a model. The real opportunity lies in creating things that weren't possible before—new products, new services, new ways of working. The companies that succeed will be those that treat AI as a catalyst for reinvention, not just a productivity tool. They will empower their people, redesign their processes, and ask the fundamental questions that lead to genuine transformation.