The Token Trap: Why Most AI Investments Fail to Deliver (and What the Best Organizations Do Differently)

Many companies are currently trying to understand the automation potential of their workforce. They assess jobs, tasks and activities. They estimate which parts of work could be automated or augmented by AI. They identify where productivity could increase, where processes could change and where new skills may be required. This is important. But it is only the starting point.

But investment in automation in itself does not change a business. A dashboard showing that 30% of a role could theoretically be automated does not reduce cost. A report showing that a process has AI potential does not improve customer satisfaction. A heatmap showing automation exposure does not create innovation.

Automation potential is tangible and valuable insight. But realizing true, long-term value? Harder. Much harder.  

And how could it not be? Even now, this is an emergent technology; used by less than 1 in 5 of the world’s working age population. There is no playbook, no best practice, and no guide that will show you the way forward. However, this is what I have learned over the past year from organizations that I believe are at the cutting edge of AI-led workforce transformation.  

The Numbers Are Already Significant

McKinsey estimates that generative AI and other technologies could automate work activities that absorb 60–70% of employees' time today. Multiple studies make similar predictions. The potential is clearly there. But potential is not the same as impact.

BCG found that 74% of companies have yet to show tangible value from AI, despite years of investment, pilots and experimentation. Only a smaller group has managed to move beyond proofs of concept and generate real business outcomes.

That is the challenge many organizations are now facing. They have AI tools. They have automation assessments. They have pilots. They have dashboards. But they do not yet have a systematic way to turn automation potential into business value.

Automation Needs to Translate into Business Outcomes

If AI is treated only as a technology, it will often remain a collection of isolated tools, experiments and local productivity gains. But if AI is treated as a business transformation topic, automation data can become the starting point for very concrete outcomes.

• Greater innovation: AI can increase the speed at which ideas are developed, tested and turned into products or services. But only if organizations create the space, skills and operating model to use it.

• Cost leadership: If AI allows teams to do more with less effort, this should eventually show up in workforce plans, cost structures and operating models. Otherwise, the efficiency potential remains theoretical.

• Increased process efficiency: AI can reduce manual work, shorten cycle times and improve quality. This can lead to faster responses, better service levels and higher customer satisfaction. But again, only if processes are redesigned around the technology.

Where Many Organizations Fail

Many companies are introducing AI tools at speed. That is good. But implementation often stops too early.

The first mistake: no change in performance expectations

Companies introduce AI software without changing the performance expectations around work. A good example is Microsoft Copilot. Many organizations roll it out and celebrate adoption. But the more important question is: what changes because of it?

If Copilot helps people write faster, analyze faster or prepare faster, then this should lead to one of two outcomes: either the same amount of work requires fewer resources, or the same people can deliver more output. For a sales organization, that could mean more customer conversations or higher sales quotas. For a support organization, faster resolution times. For a finance team, shorter reporting cycles. But if nothing changes in the targets, processes or workforce plan, the productivity gain may never reach the business.

Microsoft and LinkedIn found that 75% of knowledge workers use AI at work. At the same time, 60% of leaders said their company lacks a vision and plan to implement it. That is exactly the problem. AI adoption is happening. But in many cases, it is not yet managed as a business transformation, with stakeholders from across the business invested in its success.

The second mistake: fragmented adoption

Different business units buy different AI tools. Teams experiment independently. Some functions move fast, others wait. The result is a fragmented AI landscape without a clear enterprise view of where automation should create value. This creates activity, but not necessarily impact.

The third mistake: adding AI to existing processes instead of redesigning them

The first cars looked like horse-drawn carriages with an engine attached. It took time before people realized that a car did not need to look like a carriage. The same is happening with AI. Many companies are still putting AI into old workflows, old approval chains and old job structures. They automate single tasks, but they do not rethink how the process should work if AI is at the core. That limits the impact dramatically.

The Right Question

A lot of organizations start with the wrong question. They ask: where can we use AI? That is understandable, but it often leads to tool-driven experimentation. The Token Trap.

A better question is: where should AI change the way our business works?

That question forces a different conversation. It connects AI to strategy, workforce planning, process design, cost targets, customer experience and future skills. And this is where automation data becomes powerful as a trigger for a business conversation.

The best practitioners that I have worked with, follow a simple model:

Step 1: Estimate Your Automation Potential

The first step is to create transparency. Which roles are most exposed to AI? Which tasks can be automated or augmented? Which processes have the highest efficiency potential? Which capabilities will become more or less important?

This does not need to be perfect. The goal is not to create a mathematically precise automation score. The goal is to create a structured view of where AI could realistically change work, to identify where the biggest opportunities and risks are, and where leadership attention is required.

But the output should not only be a list of exposed roles. The output should be a business hypothesis. For example:

If AI can automate 20% of activities in customer service, what does that mean for cost, speed, quality and customer satisfaction? If AI can augment 30% of work in sales, does that mean fewer resources, more pipeline, higher conversion or higher sales quotas? If AI changes the skill requirements in software engineering, does the company need fewer developers, different developers or faster product cycles?

That is the level where automation potential starts becoming useful.

Step 2: Challenge the Business Through Workforce Planning

The second step is where many organizations fail. Automation potential should not stay in HR, strategy or technology teams. It needs to become part of the business planning conversation. Business leaders should be challenged with external market data, internal workforce data and automation potential.  

For example: if the market is already moving toward AI-enabled service models, what does that mean for your team? If 20–30% of activities in a function can be automated or augmented, how much of that is realistic for your business? If competitors are using AI to reduce cost or increase speed, what investment would you need to keep up?

This is where the real value starts. Because the moment business leaders are asked to reflect on automation potential for their own area, they start thinking more deeply about what AI actually means for their processes, their people, and their goals.

This conversation should lead to concrete workforce assumptions: How many FTE will be needed in three years? Which roles will grow or decline? Which tasks will disappear? Which new skills need to be built? What productivity improvement is expected? Which process KPIs should improve?

The output should not be a vague ambition like "we want to become more AI-driven." The output should be a quantitative workforce and business target for the next three years.

Step 3: Monitor Whether the Value Is Actually Realized

The third step is ongoing monitoring. Once workforce targets are defined, companies need to track whether the expected change is actually happening. Are the planned efficiency gains materializing? Are processes becoming faster? Are business units actually changing the way they work, or are they only adding AI tools on top of existing processes?

This is where workforce planning becomes much more dynamic, becoming a mechanism to monitor whether the organization is actually capturing the value of AI.

This also helps avoid one of the biggest risks in AI transformation: mistaking adoption for impact. A company can have high AI tool usage and still limited business value. The only meaningful measure of success is whether AI changes the cost base, the speed of work, the quality of decisions, the customer experience and the skill profile of the organization.

AI automation data can be powerful. But only if it leads to decisions about what exactly work should look like in the future. The Token Trap: measuring activity and mistaking it for progress. The organizations that escape it are those that make better decisions, not better dashboards.

‍

Florian Fleischmann is SVP of AI and Business Transformation at TalentNeuron, where he works with global enterprises on translating AI strategy into workforce decisions. He focuses on the gap between automation potential and business impact.

‍