K3 Insights

The prospect of large-scale workforce reduction (often framed in terms of 30-40 percent) has become a recurring feature of conversations about AI, automation, robotics and digitisation. It is a compelling headline. It also makes for a very efficient board discussion.

It is, however, frequently misunderstood.

In practice, outcomes of that scale are neither inevitable nor easily achieved. They require a very specific set of conditions to be in place across technology, operating model, and demand. Without those conditions, the conversation tends to remain exactly that, a conversation. Or, at best, a slightly more efficient version of the status quo.

A more useful way to understand this scenario is to think of it as a set of cumulative layers, each of which must be addressed in sequence. Most organisations focus heavily on the first one, make some progress on the second, and then quietly avoid the third and fourth, which is where the real impact sits.

The first layer is the automation of transactional work. This includes high-volume, rules-based activities such as data processing, administrative workflows, and structured customer interactions. Most organisations are already making progress here, and the benefits are real. However, based on experiences to date the impact is often overestimated.

Processes are rarely as clean as they appear in PowerPoint. Exceptions have a habit of multiplying. And automation tends to be applied unevenly, creating pockets of efficiency rather than system-wide change. The result is that this layer delivers incremental improvement, but rarely anything that would be described as “dramatic.”

The second layer is AI-enabled decision augmentation. This is where the conversation typically becomes more ambitious. AI has the potential to influence more complex, judgment-based work (e.g. identifying patterns, generating recommendations, and reducing the need for manual analysis). It starts to touch roles that were previously considered relatively insulated, including analysts, specialists, and parts of middle management.

However, there is a consistent pattern here. Organisations introduce AI, productivity improves, and then…very little else changes.

Roles remain largely intact. Structures stay the same. Workflows are only partially redesigned. In effect, AI becomes an additional layer of capability rather than a catalyst for simplification. The organisation gets smarter, but not necessarily leaner. 

To translate augmentation into meaningful workforce reduction, something more uncomfortable needs to happen. Roles need to be redesigned, spans of control need to increase, and duplication needs to be actively removed.

Which brings us to the third, and most critical, layer - operating model simplification.

Even where technology has created efficiency, complex operating models have an extraordinary ability to absorb it. Multiple management layers, duplicated roles across functions, and fragmented accountability create a level of friction that quietly consumes any gains that have been made. Given the numbers of projects we have been involved in our experience is that many organisations are, in effect, highly efficient at being inefficient.

To capture the full benefit of automation and AI, the way work is organised has to change. This often means moving away from traditional functional silos towards more flow-based or value stream-oriented models, where accountability for outcomes is clearer and the number of handoffs is reduced. Fewer handoffs, fewer layers, fewer people involved in each piece of work. It sounds obvious when stated plainly, which is perhaps why it is so often avoided in practice.

The final layer, and the one that receives the least attention, is demand reduction.

Most organisations focus on how work is performed, but far fewer ask how much work should exist in the first place. Workload is driven by product complexity, service variability, and internal processes, much of which have grown over time with very little deliberate design. Reducing demand requires making choices that are often more strategic than operational. Simplifying product portfolios, standardising offerings, and eliminating low-value activities all reduce the volume of work flowing through the system.

This is also where things tend to get slightly more political. It is one thing to automate a process; it is another to remove the need for the process altogether.

The key insight is that these layers are interdependent. Automation removes basic tasks. AI enhances productivity in more complex areas. Operating model simplification captures and consolidates those gains. Demand reduction lowers the total volume of work. 

Focusing on only one or two of these layers, usually the more comfortable ones, is why many organisations invest heavily in technology and see only modest reductions in workforce.

Even when all four layers are addressed, there are practical constraints. Regulatory requirements, customer expectations, and the organisation’s capacity to absorb change all shape what is possible and how quickly it can be achieved. Significant reductions, where they do occur, tend to play out over several years and are rarely evenly distributed.

A more constructive framing, therefore, is not to ask how to reduce headcount by a given percentage (tempting as that question is for boards/ investors), but to ask what would have to be true for the organisation to require materially fewer people.

It is a subtle shift, but an important one.

Because it moves the conversation away from cost-cutting targets and towards system design. It forces a discussion about how work is structured, how demand is generated, and how technology is actually applied. And, perhaps most importantly, it exposes whether the organisation is willing to tackle the harder, less comfortable changes or whether it is hoping that technology alone will do the heavy lifting.

In most cases, it will not.