Leading with AI-first – Rethinking and redesigning processes and organizations

By Elisabeth Goos and Michael Schich

Despite all the hype and a proliferation of AI pilots, most organizations never seem to get out of the "pilot trap," with 95% of AI investments failing to deliver the expected ROI according to MIT NANDA's The GenAI Divide: State of AI in Business 2025 report. The root cause? AI is too often treated as an add-on, optimizing only within silos and constrained by legacy processes and structures that were never designed for intelligent automation.

Reimagining process design

Instead of tinkering with historically grown, human-centric processes and trying to make minor adjustments here or there, the AI-first approach starts from zero. It identifies the outcomes that matter most and, working backward from there, builds completely new workflows with AI at the core. This zero-based, outcome-driven methodology breaks free of legacy constraints, enabling organizations to unlock radical efficiency, agility and future-proof operations.

Shifting the paradigm –
From AI add-ons to AI-first

We have developed a framework that guides organizations through this transformation in three steps:

1. It inventories a company's ambitions for AI and its direction of travel.
2. It orchestrates business, people and technologies to translate this strategy into business value.
3. It scales the impact of AI across functions, business units and regions.

The figure below shows the general path that sees AI placed rigorously at the center of every link in every process chain. The ultimate goal is to deliver end-to-end transparency, seamless automation and significant, measurable results.

The transformation to AI-first truly begins at level 3 in the figure above, where processes are derived from clear outcome statements and critical success factors: What needs to be improved? Speed? Quality? Cost? Some combination with other targets?

Separating the "what" from the "how"

This step essentially isolates what needs to be done (the outcome, which is stable) from how it is to be done (the process by which this outcome is to be achieved, which should be adaptive). With AI technology advancing at a very rapid pace, the best way to reach an outcome today will not still be the best way tomorrow. An adaptable process design allows the "how" to be updated and improved repeatedly in line with new developments.

Addressing the "how" question more closely, the AI-first approach allows neither traditional process structures nor individual AI use cases to limit the role of automation. Instead, a zero-based design policy is adopted: Targeted outcomes are translated into business capabilities and jobs to be done, and each step is aligned with the organization's strategic business goals.

In this way, existing AI capabilities can be inventoried and embedded in process flows and decision logic from the outset. Similarly, areas where human expertise remains critical can also be defined: Explicit human-in-the-loop governance too must be built in from the start, balancing optimal automation with the need for human oversight and accountability.

While it is important to remember that "AI-first" does not mean "AI-only," the trend toward agentic AI will ultimately see human intervention increasingly called in by integrated, automated systems, not the other way round. Along the way to this goal, processes must be tested incrementally and refined – allowing the organization to learn and adapt – before they are rolled out at scale. This step-by-step transition will also require strong change management to mitigate cultural resistance and drive adoption, ensuring that teams understand the measurable benefits and can genuinely embrace new ways of working.

A simple example from corporate life illustrates the astonishing potential of applying AI-first design principles to familiar, everyday processes: Booking business travel involves a raft of individual steps that, while not complex, consume a great deal of time and effort – from checking policy guidelines to submitting requests, from researching travel options to booking and payment, from delivering travel documents to dealing with disruptions and documenting expenses.

By comparison, integrated AI-led travel management looks rather different: AI can instantly turn a spoken or typed request into a clean, policy-compliant itinerary proposal complete with options ranked by time, cost, risk and sustainability . Human selection and approval of the best options prompts AI to execute all bookings and payments. After automatically synchronizing team calendars, AI then keeps a constant lookout for delays, providing rerouting options and any support needed while also cataloguing expense records cleanly and in real time.

Excluding travel time, the traditional option typically takes five hours to complete. The AI-driven option involves only 30 minutes of human involvement, again excluding travel time. From a process perspective, this equates to a productivity gain of a factor of ten. And this is only one example of many.

Deep dive: How does AI-first design work in practice?

This figure visualizes the technical implementation of our AI-first framework – and clearly shows where the future is heading. As organizations increasingly shift from AI use case–based approaches to AI-first strategies, AI agents gradually move from executing individual tasks to ownership of end-to-end workflows. Conversely, humans shift from execution tasks to supervisory roles, the handling of exceptional situations and governance issues. Moreover, as this transformation spreads throughout the entire organization, a centralized platform enables the scalable and reusable deployment of agentic capabilities across multiple processes.

Getting started with AI-first process design

To break out of the "pilot traps" and point-solution approaches that are currently widespread (and the source of much frustration, for all their undisputed benefits), organizations need to begin with the following practical steps:

• Start with clearly defined outcomes and draw up an inventory of critical success factors
• Translate these outcomes into business capabilities and jobs to be done (separating the "what" from the "how")
• Build a detailed and sequenced inventory of AI activities, including human-in-the-loop governance wherever this is appropriate
• Use incremental pilots and iterative design to manage complexity and retain focus
• Accompany the transition with strong change management to mitigate resistance and cultivate an understanding of the benefits
• Leverage cross-project synergies and expertise.

Sign up now to receive regular insights on digital transformation and value creation via e-mail.