Processes Instead of Promises: What We Can Learn from the Industrial Revolution for AI

"The AI revolution will boost productivity by 50%!" "ChatGPT makes knowledge workers twice as efficient!" Such euphoric predictions currently dominate the headlines. But these promises won't be fulfilled—at least not automatically. As the history of the Industrial Revolution shows, the key to true productivity leaps lies not in the technology itself but in the systematic industrialization of work processes.

These predictions have a crucial problem: They won't come true.

The promising figures from analysts only tell half the story. As the Industrial Revolution teaches us, new technology alone doesn't automatically lead to productivity gains. The real breakthrough only came with the systematic redesign of work processes - a lesson that is crucial for the AI era.

The productivity gains will only materialize if we understand what causes them.

Management Summary

The current euphoria around AI-driven productivity gains in companies overlooks important lessons from the Industrial Revolution: Not the technology itself, but the systematic industrialization of work processes is the key to success.

Key Insights:

Productivity Paradox: As with previous technologies (computers, internet), the mere use of AI tools does not automatically lead to measurable productivity gains.
Process Before Technology: The decisive lever for productivity gains lies in the systematic industrialization of knowledge work, not in AI technology itself.
Resource Reality: AI implementations require similar resources and structures as other major transformation projects - they cannot happen "on the side."
Hub-and-Spoke Model: Successful AI implementation requires a balance between central control (hub) and decentralized implementation in specialized departments (spokes).
Cost-Based Prioritization: The identification of AI use cases should be based on contribution margin accounting to address the largest cost drivers.
Two-Pillar Strategy: AI tools (central) and AI process applications (decentralized) must be considered and coordinated separately.
KPI Control: The combination of early and late indicators is essential for the successful management of AI implementation.
Avoiding AI Theater: Without clear processes and metrics, AI tools will at best increase work comfort but achieve no real productivity gains.
Hands-on Process Optimization: Direct understanding of work processes is essential - theoretical analyses are not sufficient.
Realistic Timeline Planning: The full impact of AI implementations, as with previous technological revolutions, only unfolds over longer periods.

The Illusion of AI Productivity: Why the Numbers are Deceptive

Let's consider an illustrative example: A company with 1000 knowledge workers implements AI tools. Software developers receive GitHub Copilot, others Microsoft Copilot. The costs average €30 per user per month.

This means the company must spend €360,000 annually, which appears immediately on the balance sheet. The vendors promise that the investment will pay off quickly: software developers should program 50% faster, other knowledge workers should be 50% more productive.

If this were true, the investment would be a no-brainer. But does this reflect reality?

Let's imagine knowledge work was like a digital assembly line. There would be standardized processes, clear division of labor, and efficient conversion of inputs to outputs in defined steps, supported or automated by AI tools.

Yes, in this ideal scenario, the promised productivity gains would materialize.

This does not reflect the reality of today's knowledge work in companies!

In typical companies, knowledge work resembles a manufactory with master craftsmen/artisans who complete their tasks according to their own judgment and expertise. Processes often exist for quality management reasons, but the lack of living these processes leads to excessive variance in working methods and productivity levels.

What happens when you give these "knowledge work artisans" AI tools? Tasks get completed faster. But this time savings doesn't necessarily lead to higher productivity. The reason: clear process definitions and metrics for productivity in knowledge work are missing. As a result, it remains unclear how the time saved through AI is being used.

Without productive alternatives, the freed-up time is often used for unproductive activities. These unproductive activities are excessive meetings, unnecessary emails, or browsing LinkedIn.

In practice, this becomes evident when asking knowledge workers about their AI usage. Two characteristic reactions often emerge:

AI aversion: "I work slower with AI than without!"
Limitation to tedious tasks: Only use cases for unpleasant activities that people want to get rid of are mentioned.

The reasons are multifaceted:

Desire for freedom: Knowledge workers don't want to be confined to a tighter process and tool framework.
Resistance to change: People avoid uncontrolled changes and protect their comfort zone.

In an extreme case, a knowledge worker might simply call it a day after AI has helped them complete four hours' worth of tasks in one hour.

This assessment may sound harsh, but it reflects reality. Those who implement AI merely as an additional tool will not achieve measurable productivity gains. At best, workplaces become more comfortable; at worst, the amount of trivial activities increases.

The Productivity Paradox: When Technology Fails to Increase Productivity

This problem is not new and is not limited to AI. The productivity paradox is a well-known phenomenon from earlier technologies like the internet, computers, and smartphones. It describes how, despite substantial investments in new technologies, corresponding increases in productivity or economic output are often not observed.

This paradox was first observed in the 1980s with the introduction of computers in businesses. In 1987, Nobel laureate Robert Solow summed it up:

"You can see the computer age everywhere but in the productivity statistics."

In academic literature, various explanations for this phenomenon are discussed (Wikipedia):

Adaptation delays: New technologies require time for structures, processes, and capabilities to adjust.
Measurement problems: Conventional methods don't capture all benefits of new technologies, especially regarding quality and customer satisfaction.
Misdirected investments: Companies may choose unsuitable technologies or implement them ineffectively.
Cross-sector effects: Productivity gains in one area can be offset by losses in others.

Industrialization as the Key: Why Technology Alone Is Not Enough

I recently read "The Knowledge Factory," which addresses this question about digitalization. It's more relevant today than when it was published in 2018.

The Knowledge Work Factory: Turning the Productivity Parafox into Value for Your Business

The author comes to a surprising conclusion: The true driver of productivity gains is not technology, but industrialization.

The Industrial Revolution is often described in terms of three "great achievements": steam, electricity, and digital technology. But is this really the core of productivity growth? A closer look reveals that it wasn't the technology itself, but the industrialization of work that enabled the decisive productivity gains.

Steam and Electricity: Industrialization of Physical Labor

The first great achievement, "steam," freed humans and animals from physical labor. Steam engines, looms, and locomotives increased economic productivity. In Great Britain, the cradle of the Industrial Revolution, productivity quadrupled between 1760 and 1860.

The second achievement, electricity, enabled more efficient production processes and new tools for mass production. Electrification created new possibilities in factories and increased productivity in households. Between 1920 and 1970, the USA experienced a golden era of productivity growth.

Neither steam nor electricity were the actual drivers of productivity gains. The systematic industrialization of work processes made the difference. People created standardized procedures, broke down complex tasks into their components, and optimized workflows to maximize efficiency. These principles could be directly applied to the use of new technologies.

The Digital Age: Where Are the Productivity Gains?

Then came the third great achievement: digital technology. It promised a similar revolution to its predecessors. But contrary to expectations, productivity growth has stagnated since the 1970s. A brief phase of increased productivity from 1994 to 2004 remained the exception.

Why did the productivity gains fail to materialize? The answer lies not in the technology, but in the lack of willingness to industrialize knowledge work. While physical labor was systematically industrialized, intellectual work remained in a manufacturing state. Digitalization is visible everywhere – yet productivity statistics show no increase.

The real challenge is the industrialization of knowledge work.

The challenge lies in industrializing intellectual work like physical work. This means defining, standardizing, and optimizing work processes to achieve efficiency gains. Knowledge work often lacks clear processes and metrics, causing productivity gains to be lost.

History shows that every major technology introduction had a "productivity paradox" phase. Steam engines needed 60 years to displace old technologies, and electricity took decades to reach its full effect. Digitalization has taken the same path – without systematic industrialization of knowledge work, it will not bring notable productivity gains.

Management must implement technology and organize knowledge work in a way that allows these technologies to reach their full potential. The key lies not in the tools, but in their application and the organization of our work.

The insight from the Industrial Revolution is clear: The true force behind productivity gains is not technology, but the systematic industrialization of work.

A side note: At the beginning of the 20th century, there was speculation that we would only need to work 4 hours per week in the 21st century. Had we used digitalization to industrialize knowledge work, this could have been the case. Since this didn't happen, we continue to toil away — a missed opportunity!

Guide: Successfully Integrating AI into Business Processes

If you've read this article up to this point, you're like me - dealing with the implementation of AI in your company.

You want to avoid making wrong decisions where effort and benefit don't align.
You want to avoid wasted investments and abandoned tools.
You only want to burden your company's workforce with another change initiative when it brings tangible benefits.

The risk isn't just about misguided investment. It's also about distracting your organization with an additional topic that keeps your employees from more important priorities.

AI must not become a smoke screen.

Based on these insights, we can derive recommendations and pitfalls.

Avoid distractions from AI use cases that only address "annoying tasks."
Start with the process and look for major cost drivers or overlooked process absurdities.
First search for unexplored non-technology improvements before considering AI use cases.
Avoid implementing Copilot or similar tools for all employees without clear usage guidelines. The introduction of M365 Copilot could be linked to having each employee follow only the AI summary for 2-3 meetings per month instead of attending the meeting. This would recover the license costs.
Don't get distracted by superficial AI use cases from content creators on LinkedIn, Instagram, etc. They only scratch the surface, think in terms of prompt engineering, and don't approach it from a process perspective.
Don't be driven by colleagues, shareholders, and the market. Even though ChatGPT will soon be two years old, technology adoption in companies is still in an early phase. The high frequency of weekly news shows this. Waiting can be a sensible strategy, and not everyone needs to be an early adopter.

Here's how your approach could look:

Step 1: Cost Analysis as Foundation

Analyze your company's previous year's balance sheet and identify significant costs from knowledge work in the contribution margin calculation.

The contribution margin calculation is a multi-level method for cost analysis. It examines costs and revenues layer by layer. Typical levels are:

Contribution Margin I is the difference between revenues and variable costs. This includes, for example, project resources (€/hour) for implementing a customer project.
Contribution Margin II: CM I minus product-fixed costs (e.g., IT platform investments).
Contribution Margin III: CM II minus division-fixed costs (e.g., Key Account or Service Management, Sales)
Contribution Margin IV: CM III minus company-fixed costs (staff functions, management)

This method helps you identify the largest cost drivers and potential inefficiencies in knowledge work within your company. These inefficiencies can be eliminated through AI.

A typical finding might be that AI use cases in marketing are frequently promoted on social media. However, the actual savings in marketing are often small compared to other areas. It makes more sense to focus on areas with greater savings potential. A 1% cost reduction in production could be more valuable than a 10% saving in marketing.

Don't forget that ROI in contribution margins is your north star. Other benefits through AI are also possible: e.g., increased quality (=fewer escalations and risks) or faster response times.

Step 2: Systematic Analysis of Potential

To organize AI use cases and internal contacts, you need an appropriate structure. The contribution margin calculation is too broad for this purpose.

Useful structures can include:

HR statistics (employee numbers by job families and levels)
Process management (list of all company processes with responsible parties)
Organizational chart
Portfolio structures

Note: These structures quickly become complex. Even medium-sized companies often have over 40 processes, 100 portfolio elements, or 300 departments.

Sensibly narrow down the search area by mapping cost sources from the contribution margin analysis to the more detailed structure. This helps you identify which areas need deeper examination and where to look for process improvements with AI.

Which teams and directorate areas are involved in which process or functional costs? Which processes are high cost drivers? Which sub-processes? Which similar or related job roles generate which costs? How many juniors benefit from AI assistance? Etc.

Step 3: Mobilize Teams and Resources

There's an interesting phenomenon in companies. When a new major client is acquired, a new factory is built, or a new market is entered, management immediately understands that massive investment in seizing opportunities is necessary. This includes coordinating the endeavor as a program or project. A factory isn't planned by a business developer alone, but by a team with various roles.

These endeavors are tangible and understandable. Decision-makers know what needs to be done. It's different with process improvements, digitalization, and AI implementation. Although the potential EBIT benefit can be greater, decision-makers often believe this can be handled by one person or a small team on top of their existing duties.

A comparison makes this clear:

A new business model brings €10 million EBIT per year. It requires hiring up to 385 employees, assuming salaries of €80,000 (+30% additional costs) and an EBIT margin of 10%.
Implementing AI for process improvement brings €10 million EBIT per year. Then it's said: "Klaus from Business Development can handle that."

Why this discrepancy?

Process and digitalization topics seem abstract and intangible. Often there's a lack of imagination about what specifically needs to be done and what resources are necessary. This leads to collectively underestimating the project and inconsistent implementation.

As a driver of these topics, it's up to you to make clear to decision-makers what's required - both strategically and operationally. Use pilot projects to create concrete examples:

Strategic: Decide with management whether a "top-down" or "bottom-up" organization makes sense.
Operational: Define clear task lists. What needs to happen in the next ten weeks? Who takes on which tasks?

It's crucial to make clear that this isn't a one-person job. For successful process improvements and AI implementations, you need a team with diverse competencies, e.g.:

Technology experts for selecting and implementing the right tools.
Change managers to prepare the organization for change.
Department representatives who know the process and requirements.
Project managers to coordinate tasks**.**

Use initial experiences from pilot use cases to illustrate what needs to be done to ensure all success factors for a successful AI use case. Examples include:

Assigning roles,
Moderating workshops,
Bringing stakeholders together,
Developing smart make-or-buy strategies,
Designing solution concepts and mockups,
Integrating partners and service providers,
Conducting process analysis,
Developing product vision,
Making KPIs measurable,
Identifying specific process activities where AI should support,
Designing training,
Clarifying operations,
Orchestrating budget decisions,
Managing implementation team,
Conducting user tests,
etc.

Warning Signs: When Resources are Lacking

What should you do if management cannot provide resources but still wants AI successes? In this case, you should drop the topic.

Why?

Without adequate resources, you will either:

not be able to access the crucial processes with the greatest efficiency potential, or
become a bottleneck that holds back the organization.

In both cases, you risk causing more harm than good. You'll be condemned to putting on an AI show for employees, customers, and shareholders instead of generating real progress.

It's better to tackle the topic only when the necessary resources are available. This is the only way to ensure professional execution. Invest the necessary time to clarify these resources with decision-makers.

Digitalization and AI are not side projects. They require the same structured approach and resource commitment as building a new factory or introducing a new business model. Only with a clear plan, realistic resources, and a well-positioned team can you unlock the full potential and gain the trust of decision-makers.

Step 4: Two-Pillar Strategy: Separate Tools and Processes

Structure your AI implementation project into two areas. The first area concerns which AI tools and base technologies you want to introduce in your organization. The second area is how you use these tools in use cases to create value in processes.

The difference is: The AI tool question can be resolved centrally, e.g., by the CIO, while AI process applications are coordinated and driven decentrally with the respective process owners.

Coordinating both workstreams involves different questions.

AI Tools (e.g., implementing M365 Copilot):

Rollout concept: Who gets which license and for what?
Which tools and partners do we use?
When do we choose make or buy for solutions?
Ground rules: What can be decided decentrally, what falls under CIO authority?
How do we shape AI adoption? (Training, communities, ...)
How do we technically organize the ordering and change processes?
How do we ensure compliance with EU AI regulations?
Monitor costs and benefits of AI implementation holistically.

AI Process Applications (e.g., contract management AI):

Identify and prioritize use cases.
Develop AI application visions and align with stakeholders.
Define KPIs and objectives; calculate ROI.
Develop solution concepts and make make-or-buy decisions.
Allocate resources and organize budget decisions.
Manage implementation and operations.
Quality Assurance.
Monitor KPIs and ensure cost/benefit ratio.
Ensure knowledge exchange and reuse between use cases.
Identify personnel development needs.

Step 5: The Hub & Spoke Model: Central Control, Decentralized Implementation

The introduction of AI tools and process applications requires a clear structure to balance the tension between top-down and bottom-up approaches.

The bottom-up approach enables the identification of practical use cases through proximity to processes and detailed knowledge from departments. However, it carries the risk of creating an "AI zoo": many Proof of Concepts (PoCs) without transition into productive applications.

The top-down approach can bundle resources and skills, but there is a risk of being detached from reality and producing impractical results.

A Hub & Spoke model is ideal for reconciling these opposites.

Hub = central, coordinated team // Spokes = AI use case owner

The central project team ("Hub") takes on a coordinating, supporting, and steering role. It ensures transparency, strategic priorities, and governance. This team acts as an enabler and catalyst for the departments.

Hub responsibilities:

Identification of high-impact use cases: Prioritize use cases with the highest potential for ROI and strategic value.
Central support: Provision of templates, tools, and best practices for departments.
Coordination and transparency: Consolidation of activities to avoid duplicate work and leverage synergies.
Governance: Ensuring IT security, AI compliance, and overarching AI governance.
Enablement: Training of AI owners and departments.

Recommended resources in the Hub:

Project coordinators and conceptual designers: Lean core team for steering and planning.
CIO/IT representatives: Responsible for governance, security, and compliance.
Implementation support: Assistance for prototype development and transition to productive operation.
HR: Support for personnel development and building AI-relevant competencies.

The "Spokes" are the departments and process owners who implement concrete AI projects and process applications. They bring detailed knowledge and proximity to business processes.

Spoke responsibilities:

AI Owner: Responsible for a specific use case, including implementation, rollout, and ensuring ROI.
Process Owner: Driver for process improvements (with AI).
Employees with AI ideas contribute pain points from their daily work.
Product Owner: Responsible for technical management of AI applications and integration into the IT landscape.

Collaboration between Hub and Spokes:

Regular synchronization meetings to align on progress and challenges.
Hub support in scaling successful use cases.
Continuous knowledge transfer between Hub and Spokes to spread AI competency throughout the organization.

The Hub & Spoke approach provides the structure for successful AI implementations. It combines central control with decentralized execution and prevents pitfalls like an uncoordinated AI zoo. Companies achieve quick results and ensure long-term value creation.

Benefits of the Hub & Spoke Model

Balance between strategy and practice: The Hub ensures strategic control while Spokes guarantee practical implementation.
Prevention of an "AI zoo": Central governance prevents uncoordinated PoCs and promotes productive applications.
Efficiency and effectiveness: The Hub bundles resources and creates synergies while Spokes build on their process expertise.
Scaling of best practices: Successful use cases can be quickly and efficiently transferred to other areas.
Clear responsibilities: The Hub provides impulses, the Spokes deliver results – everyone knows what to do.

Step 6: Think Big, Act Small

The implementation of AI in business processes cannot be planned like a classic waterfall project. The dynamics of the technology and business environment are too high.

A realistic assessment is important. If progress in Large Language Models stagnates, cost savings will not materialize in the short term. Currently, the implementation of AI use cases often fails in the crucial final 20 percent.

Enormous investments are flowing into AI basic research. Therefore, further exponential progress is expected in autonomously acting AI systems, larger processing capacities, or better model reliability.

Since development is difficult to predict, a step-by-step approach is recommended.

Plan a multi-year budget framework for the overall project that is oriented toward the expected ROI. This is the theoretical upper limit of investments but not yet a binding expenditure. The concrete investments are decided case by case for each use case separately.

The operational Hub & Spoke concept should then be implemented in concrete, tangible 10-week sprints. Control the activities through a control board with various action areas and two development paths:

Planning action areas:

Community Work
Use Case Identification
Use Case Piloting
Use Case Rollout
AI Governance & Compliance
and others

Development paths for reporting:

AI Tool Development (Idea → Proof-of-Value → Scaling)
AI Process Applications (Idea → Proof-of-Value → Rollout/Adoption)

The funnels are an important tool and visualization to achieve a realistic picture of progress.

Step 7: KPIs as a Compass: Using Early and Late Indicators Correctly

When introducing AI, there is a risk of getting lost in everyday use cases that offer no strategic value, such as the often-requested automated travel expense reporting.

Therefore, it is important to plan and monitor the AI implementation program using KPIs.

In the Hub & Spoke model, we achieve this by having the central Hub resources focus on KPI-driven activities.

In discussions with the Spokes (decentralized AI leads), each use case directly clarifies where in the process AI should have an impact and which process KPIs should improve.

An example: A sales manager in the tender business typically monitors KPIs such as effort per tender or processing times, ideally through a CRM-based analytics dashboard.

Here we can set concrete goals: "Let's reduce the processing time for tenders from six to five weeks" or "We want to increase the success rate for tenders from 35% to 50%".

The Hub creates a global AI KPI dashboard. Here, the entire organization can track how KPIs develop through various AI applications - and take corrective action if needed. Everyone can see whether the AI mission is winning or losing and which KPI they can contribute to.

When building the KPI system, distinguishing between two KPI types is important:

Lagging KPIs show the desired improvements, e.g., cost savings or revenue increases. These metrics are measurable with a time delay.

Leading KPIs are early indicators that show us during implementation whether we are on the right track.

Example: In service management, cost reduction is a Lagging KPI. The proportion of AI-supported service tickets is a Leading KPI that signals early on whether we will achieve the desired cost savings.

Step 8: Hands-on Process Optimization: The Direct Path to the Goal

An important note for everyone in the central hub of the AI implementation program: The Hub & Spoke model only works if the Spokes fulfill their tasks effectively (meaning you are only an indirect enabler).

In practice, AI leads in the departments often lack sufficient resources or competencies, which leads to the program missing its objectives.

Therefore, it's important not to shy away from choosing the most direct path to process optimization. Look at concrete processes yourself. Understand why a service ticket that could theoretically be processed in one day actually takes six weeks in reality. Form your own opinion and don't just rely on others to assess and understand situations correctly. As a rule, they don't.

To use AI effectively, you - or another expert - must analyze the work processes. Observe daily operations and identify inefficient workflows, if necessary through direct observation over the shoulder. You can address these weak points specifically with AI solutions or through simple process improvements without technology.

Conclusion: The Path to Sustainable AI Productivity Gains

The discussion about the productivity effects of Generative AI shows that its true impact lies not only in the technology itself, but in the organization of knowledge work. As the Industrial Revolution proved, it is not the innovations themselves, but the industrialization of work processes that enable decisive productivity improvements.

The introduction of AI tools will not work miracles as long as companies do not define clear processes, metrics, and goals. They must create the foundation to use these technologies effectively:

Process-oriented approach: Before implementing AI use cases, analyze the biggest cost drivers and inefficiencies in the processes.
Hub & Spoke model: Combine strategic control through a central team with practical implementation in departments to leverage synergies and avoid an uncoordinated AI zoo.
Focus on KPIs: Plan and monitor the introduction of AI with clear leading and lagging KPIs to ensure measurable results.

Companies that want to successfully implement AI must abandon the notion that productivity gains automatically occur through new technologies. Even if that's what the salespeople of these technologies promise. They must actively work on the industrialization of knowledge work, optimization of work processes, and preparation of employees for the changes.

History shows that technological innovations need time to develop their impact. Companies that act now with a smart strategy and structured approach can harness AI's potential and secure a long-term competitive advantage. The real task lies not in the technology, but in its meaningful integration into the working world – for real, measurable improvements.

I hope these key points help with the introduction of an AI program and convey how to move from AI zoo and AI theater to real, measurable improvements through AI.

If you have questions or challenging issues that require deeper discussion, feel free to contact me at any time.

Best regards,
Martin Weitzel