No data, no AI, no value: Turning Artificial Intelligence into EBITDA

Data as a competitive moat

The quality of an AI solution depends heavily on the information on which it operates. 

Several panelists reflected on earlier machine-learning initiatives that ultimately became data-engineering programs. Companies first needed to consolidate fragmented information, expose quality issues and establish reliable performance metrics.

That investment is now becoming increasingly valuable.

Data can strengthen performance and exit readiness

Mikael explained how a common data platform can make information from different source systems visible across portfolio companies. This often exposes quality problems, but visibility gives management teams an opportunity to correct them.

A well-designed platform supports faster operational decisions and enables AI applications to work with company information. It can also provide greater transparency during an exit process.

If management can answer a potential buyer’s questions with reliable and traceable evidence, data infrastructure becomes part of the investment case.

Mikael also shared examples of proprietary data creating operational advantage. One company uses historical data to improve the pricing and recovery value of recycled mobile phones. Another analyzes movement data to optimize warehouse routing. 

General-purpose AI models are widely available. Competitive differentiation comes from combining them with proprietary data, embedded expertise, and workflows that competitors cannot easily replicate. 

Companies should not wait for perfect data

Data quality is essential, but companies should not postpone AI until every system has been modernized and every dataset cleaned. 

Marco explained that data transformation and AI implementation can progress together. A company can begin with a defined process, a focused dataset, and a measurable use case. Implementation then reveals which data gaps genuinely prevent the solution from working. 

This creates a practical cycle: 

• A prioritized use case exposes relevant data problems. 
• Targeted remediation makes the information usable. 
• The first deployment generates operational learning. 
• Better data supports broader scaling. 

Ivo illustrated the importance of operational context with an example from manufacturing. A model trained on sensor data appeared to perform well technically, but its recommendations were incorrect. 

A shop-floor review revealed that employees had changed the production process while the measurement setup still reflected the previous workflow. The data existed, but it no longer represented reality. 

Reliable AI therefore requires accurate measurement, process knowledge and continuous validation.

Critical information requires stronger protection

As AI becomes embedded in core processes, data sovereignty and provider dependency become strategic concerns. 

Nivesh emphasized that companies need to understand which data they possess, where it is stored, how it moves and which providers can access it. 

Customer information, proprietary manufacturing recipes and other critical intellectual property require stronger controls than anonymized or non-sensitive workloads. 

At SCHOCK, sensitive production know-how and customer data remain within locally controlled environments, while less sensitive workloads can use leading external models. 

Companies should also avoid making critical processes unnecessarily dependent on one model or provider. Regulation, pricing, performance and geographic availability are changing rapidly. Where practical, the architecture should allow companies to switch models or vendors without rebuilding the entire process.

Preparing portfolio companies for AI adoption

Technology is only one part of AI transformation. Workflow redesign, governance, and organizational change are often more difficult than selecting a model or platform.

Governance turns experimentation into execution

Successful AI programs need senior sponsorship and a governance process that forces prioritization. 

A practical stage-gate approach includes: 

1. Define the use case, business owner, baseline and financial ambition 
2. Confirm that the required data is sufficiently reliable 
3. Develop the solution with the operating team that will own it 
4. Validate the operational and financial results 
5. Scale successful initiatives and stop those whose costs outweigh their benefits 

Ending an initiative when the economics no longer work is evidence of sound governance, not failure. 

AI also belongs on the CEO and board agenda because it can influence the operating model, competitive position and equity story. Execution, however, must remain anchored in the functions that own the relevant workflows.

Adoption requires involvement across the organization

Mikael emphasized the importance of CEO sponsorship. Without visible support from the top, redesigning processes and overcoming resistance becomes difficult. 

Nivesh described how the focus is already shifting from tool adoption toward business impact. Usage can be a useful early indicator, but it should ultimately give way to measures such as EBITDA contribution and operational performance. 

Marco highlighted the difference between personal-productivity tools and the automation of business-critical processes. Employees often welcome tools that help with research or presentations. End-to-end automation can feel more threatening because it changes roles and established ways of working. 

Functional leaders and frontline employees should therefore be involved from the design phase. Ivo added that the discussion should begin with the customer benefit. When employees understand the purpose and help shape the solution, they are more likely to support it.

The objective is not to insert AI into every task. It is to redesign selected workflows so that employees can focus on judgment, customer relationships, and higher-value decisions.