There is a phrase that has become increasingly common in boardrooms: “We need an AI strategy.” It sounds sensible. But it may be the wrong starting point. A company doesn’t really need an AI strategy in isolation. It needs a strategy for how intelligence will change the way the company creates value. That distinction matters. If AI is treated as another technology program, organizations may deploy assistants, automate isolated tasks and create impressive demonstrations without materially changing performance. If AI is treated as a transformation of the operating model, the questions become much more fundamental: Which decisions should happen faster? Which work should disappear? Which activities should be redesigned? Where should machines augment people? Where should humans remain firmly in control? And what capabilities does the organization need to make these changes sustainable? McKinsey’s 2025 State of AI research found that high-performing organizations are more likely to pursue growth and innovation alongside efficiency, while workflow redesign emerges as a key factor in capturing value. This is the difference between using AI and transforming with AI.
Automating a Bad Process Just Makes It Faster
Consider a finance department that spends thousands of hours processing invoices. The obvious AI opportunity is invoice automation. But suppose the underlying process contains unnecessary approvals, duplicated data entry and poorly defined exception rules. Automating the existing workflow may reduce manual work. Redesigning the workflow could eliminate entire stages. This distinction is subtle but important. The first approach asks: “How can AI perform this task?” The second asks: “Why does this task exist, and what would the process look like if we redesigned it around what AI can now do?” BCG’s research on agentic AI increasingly emphasizes this shift toward AI-first workflow redesign rather than simply inserting AI into existing processes. Its work reports that agent-enabled workflows can accelerate processes by 30% to 50% in some applications, but also stresses that organizations need to redesign platforms and workflows to capture that value. The technology creates the opportunity. The redesign creates the value.
Think About the Factory, Not the Machine
There is a useful historical analogy. The introduction of electricity didn’t transform factories simply because electric motors were better versions of steam engines. The larger transformation came when factories were redesigned around electricity. Machines could be positioned differently. Production lines could become more flexible. Factories no longer had to be organized around a central mechanical power source. The technology changed the architecture of work. AI presents a similar opportunity. A company can add an AI assistant to an existing workflow. Or it can reconsider the workflow itself. The second is harder. It is also where the larger strategic advantage may exist.
AI Changes the Economics of Knowledge Work
Consider a legal or consulting organization. A significant amount of highly skilled employee time may be spent searching for prior work, comparing documents, summarizing research and preparing first drafts. These activities require expertise, but not all of them require the same level of human judgment. AI can potentially reduce the time spent on the mechanical parts of the process. That doesn’t necessarily mean fewer professionals. It can mean professionals spend more time on the parts of the work where judgment, relationships and creativity matter. BCG’s recent research similarly frames AI’s workforce impact around redesigning the division of labour between humans and intelligent systems rather than simply automating jobs wholesale. The important question is therefore not: “Which jobs can AI replace?” It is: “Which parts of the workflow should humans and machines perform?” That is a much more useful management question.
The Data Foundation Determines How Far Transformation Can Go
There is a practical limit to workflow redesign. AI cannot automate decisions that the organization cannot represent digitally. If customer information is fragmented, the AI system cannot easily reason about the customer. If policies exist only in inaccessible documents, an agent cannot reliably apply them. If operational events aren’t captured, predictive systems cannot learn from them. If data definitions differ across departments, automated decision-making becomes risky. This brings the transformation discussion back to data engineering. McKinsey’s latest AI data-readiness research argues that organizations need governed, reusable data foundations precisely because AI repeatedly reconstructs information across systems and workflows. More than two-thirds of high-performing companies in its research identify data as a major obstacle to enabling AI. AI transformation therefore rests on an infrastructure transformation.
The Operating Model Has to Change Too
Technology teams cannot carry AI transformation alone. Imagine a business where the technology department builds an excellent AI forecasting system. Operations doesn’t trust it. Finance measures success differently. Sales doesn’t use the predictions. Nobody owns the resulting decisions. The model may be technically successful. The transformation has failed. This is why leading organizations increasingly treat AI initiatives as business products rather than technology projects. There should be a business owner. There should be a measurable outcome. There should be feedback. There should be a process for improving the system. And there should be a clear understanding of where human accountability remains. BCG’s work on enterprise agents makes a similar point: successful deployment requires business leaders, technology teams and platform teams to work together, with agents treated as part of operating processes rather than isolated software features.
Transformation Should Be Measured in Business Terms
AI adoption metrics are easy: number of employees using Copilot, number of prompts, number of AI pilots, number of agents deployed. These numbers can be useful. They are not transformation metrics. The more meaningful measures are: revenue per employee, cost per transaction, time to decision, customer resolution rate, inventory turns, forecast accuracy, conversion, cycle time, and risk exposure. Those metrics tell you whether AI has changed the economics of the business. BCG’s recent “Targets Over Tools” perspective is particularly relevant here: AI transformation should be tied to explicit growth, cost and productivity outcomes, with progress measured against enterprise value rather than technology adoption. That is the standard AIGebra should hold itself to as well.
The AIGebra Perspective
We don’t see AI transformation as an exercise in adding intelligence to existing systems. We see it as an opportunity to rethink the equation by which a business creates value. Data is the input. AI is the reasoning capability. Engineering is the mechanism that makes it operational. Workflow is where it creates impact. Business outcomes are the measure. That means some transformation programs will require sophisticated AI. Others will require better data engineering. Some will require workflow redesign. Some may require all three. The technology should be selected only after the problem is understood. Because the ultimate objective isn’t to make a company more “AI-enabled.” It is to make the company faster, more intelligent, more adaptable and better at creating value. And that is a transformation worth engineering.