90 % / 67%
Use and piloting keep rising
90% of companies “use” AI, 67% have been stuck in the pilot phase for some time.
insights · preparing for ai
How do I best prepare my company for the use of AI?
Many people still associate artificial intelligence only with ChatGPT. Yet it is much more — the era of AI agents begins with the autonomous takeover of activities.
Legally compliant use in the company requires the right mindset, AI user competence in the workforce, AI-consumable data, clear processes and a suitable IT infrastructure. Mindset. Processes. Data. IT infrastructure. Competence. Everything goes hand in hand — we show you how to get there.
the status quo · mckinsey ai report 2025
AI adoption is mainstream — real transformation has not yet begun
We are still far from truly using AI. It is also capital-intensive — companies should invest strategically.
62 % / 23%
AI agents are gaining acceptance
62% of organisations experiment with AI agents, 23% are already scaling them.
64 % / 39%
Big gap between impact and ROI
64% find that AI leverages operational innovation, but only 39% see a measurable effect on EBIT. AI requires not only investment in technology but a change in mindset.
Top 6%
High performers think bigger
Real process implementation across several business areas, large budgets, defined growth figures — not just POCs.
3×
Leadership as part of the solution
Leaders who use AI themselves deploy it three times as often in the company. The technology needs impetus from leadership.
51 %
Risk management is catching up
51% of users have experienced inadequate results, mostly due to inaccuracies. Lack of trust slows AI down.
What the successful do differently
They use AI to transform how work gets done — not just to speed it up. Better processes bring benefits from staff morale to cost savings. And they measure the speed at which information turns into assessable insights — that is the new competitive advantage. As for the workforce, 32% expect job cuts, 13% growth: activities with a clear process and predictable data flow disappear, new jobs from engineering through sales to risk and leadership emerge.
reasons for failure
Why it often fails
According to a study by the renowned MIT, 95 percent of all generative AI pilot projects in companies aimed at boosting revenue growth come to nothing. The five most common reasons:
Poor data quality
Insufficient data quality, data silos and no suitable IT infrastructure.
Weak understanding of context
Unsuitable tools do not learn from feedback and therefore do not improve continuously.
Organisational and procedural hurdles
Lack of alignment with day-to-day business, unclear processes, no integration into existing workflows.
Little focus on scalable solutions
Most projects get stuck in the pilot phase and never reach production maturity. Successful companies rely on focused individual projects and partnerships with specialised providers.
Lack of strategic direction
A systematic implementation strategy is missing. Companies that integrate AI purposefully into back-office processes belong to the successful 5%.
success factors
Eleven factors for successful AI use
According to the MIT study, cooperation with experienced partners raises the success rate to up to 67%. These eleven factors make the difference:
Measurable goals
Define KPIs, prioritise projects with quick wins, uncover shadow AI.
Build the data base
High quality is decisive: cleansing, standardisation, creating interfaces.
Build the technical base
IT infrastructure that can integrate AI tools — cloud and API are king.
Process-specific solutions
Generic tools such as ChatGPT or Copilot are often not enough — successful companies rely on specialised solutions.
External partnerships
Cooperation with AI providers or consultants with implementation experience significantly raises the success rate.
Seamless integration
Integration into existing workflows — no isolated solutions.
Build in feedback
AI systems should be able to evolve with user feedback.
Involve employees
Training and clear communication are essential.
Cultural adaptation
An open culture of learning from mistakes and a willingness to question processes.
Measure success
Review of goals: ROI, process improvement.
Scale successful projects
Once measurable results are delivered, transfer to other departments and areas.
in depth
The most important levers in detail
Establish measurability
Success can only be demonstrated through measurable key performance indicators (KPIs) — e.g. time saved, cost reduction, quality improvement or parts produced per unit of time. Dissolve shadow AI: employees should disclose hidden use — three quarters of STEM professionals use AI unofficially. Only then does the impact on the company become visible.
Initiate pilot projects
Define achievable goals and start quickly: identify and prioritise use cases for quick wins to make the workforce aware of the benefit. A quick win is the trade-off between technical feasibility, productivity gain and risk impact. Observe data protection requirements, provide budget for scaling.
Describe and optimise processes
Clear process description requires decomposition skills. Reframing helps to see more clearly how AI can support — processes should generally be evaluated. Knowledge of organisational design helps reshape structures so that AI can be integrated (e.g. supply chain planning at IKEA).
Make data AI-consumable
Break up data silos — knowledge is spread across employees, departments and tools. Convert analogue data into digital formats and choose formats with structure and token consumption in mind (Markdown, JSON, TOON etc.). Interfaces: provide API documentation, consider current protocols such as MCP and A2A.
Use partnerships
Conclude data processing agreements with data centres in the EU, find external providers for AI model hosting. 360° AI agencies support consulting, integration and training as a core competence — their experience is usually greater, and enablement by experienced outsiders accelerates.
Involve and train employees
Combine top-down and bottom-up approaches — every level of the company makes an essential contribution. Understanding AI automation and agentic behaviour has been mandatory under the EU AI Act since February 2025. Live agility (fail fast, learn fast), be able to evaluate results (e.g. with the NIST AI Risk Framework), define ethical and strategic boundaries — and learn to see AI as a sparring partner.
enablement
The four-stage model
Top-down and bottom-up: all levels work together — it takes the mandate and budget of leadership as well as the will and drive of specialists.
Step 1: Convince leadership
It takes the mandate to implement AI and budget from leaders — and the will and drive of specialists.
Step 2: Lay the foundations
Build IT infrastructure, identify, provide and prepare data. Build interfaces between AI and company-specific tools, document (and question) processes, train employees — because technical, legal and ethical aspects all play a role.
Step 3: Start pilot projects
Identify use cases, prioritise and implement quick wins. Measure ROI, check KPIs — and thereby create acceptance in the workforce.
Step 4: Scale, but properly
Integrate AI centrally into product development, business processes and decision-making and rethink value creation — data-driven automation runs through every part of the company.
learnings
What have we learned?
Data
… is the central component of every AI application. Without appropriate interfaces to data sources it stays unused.
Processes
… must be clearly described and should also be questioned as part of an AI strategy. The workforce has to form the necessary mindset.
Employees
… are the most important element of a transformation. All levels of the workforce are part of the transformation process — they need the necessary mindset.
Don't wait — shape
… serves as the motto for the AI momentum of 2025. Use the opportunity to secure competitive advantages.
This is how we implement it: The Sellium Method · AI Consulting · funded AI training · Funding options
contact
Let's talk
Happy to help with anything AI. We will get back to you promptly.
Get in touch
Thank you for your message.
We will get back to you shortly. If it is urgent, you can reach us at +49 371 524 99 140 or contact@sellium.ai.