In this blog post, we shift perspective: How did we experience AI implementation and evolution? After all, we weren’t AI natives or an AI-first company from day zero. Rather, through our transformation process, we learned extensively about change management and how to effectively design collaboration with AI systems.
We’re convinced that many executives currently face similar challenges and can recognize their own AI transformation journey in this post.
Transparency Note: This article is based on our practical experiences with various AI tools. The mentioned tools do not constitute recommendations, nor are we implementation partners of any provider.
The transition to “multiplayer mode” simultaneously became a team transformation for us, opening up many degrees of freedom and making us more efficient. Shared workflows and standardized prompts, along with cross-team AI strategies, have delivered real and measurable advantages.
Curiosity About AI and Our Experiences with AI Evolution
In the very beginning, there was curiosity. LLMs or Large Language Models suddenly became generally available. Curiosity matched our questions about what AI is capable of or will become capable of. The thought of using AI in a business context existed, but remained limited to translating or “improving” texts. Occasionally, an email was also formulated with ChatGPT’s help.
Unfortunately, many setbacks and deficits emerged. These early LLMs had no context memory. External interfaces were rare and often worked unreliably.
In this early phase, we developed long prompts instead of actually standardizing and documenting workflows as we do today. Prompt evolution helped, even if results sometimes sounded mechanical and typically stiff for AI.
The AI Transformation Journey
5 phases from chaos to strategic AI excellence
"Wild West"
"Copilot Purchased, Problem Unsolved"
"AI as Personal Sparring Partner"
"Multiplayer AI Teams"
"Human-led, Agent-operated"
Phase 0: Our Wild West
Finally came the phase we describe on our website about AI Evolution as the “Wild West.” Experiments with AI were largely unplanned.
The experience: One tool could solve this type of task better, while the next tool handled a different set of tasks better.
Frustration ran high because the AI suddenly started hallucinating and dragged previously achieved results into the digital abyss.
At that time, we had no strategy for dealing with AI. Only the idea that AI could help us create content more easily, based on existing drafts.
We didn’t consider more complex tasks such as designing new offers or using it as a strategic sparring partner in this phase. Also because the technology wasn’t yet that advanced and remained unreliable in many areas.
Our Change Management Learning: Phase 0 is normal and necessary. Every company must first experiment to understand possibilities. The mistake: Remaining in this phase instead of systematically moving forward.
Phase 1: Copilot Purchased, Problem Not Solved
One day comes the point where AI is introduced company-wide. We also reached this point. Google Gemini and ChatGPT were the two primary AI tools we used.
We supplemented these with NotebookLM, a potent solution from Google that makes it possible to evaluate large amounts of different sources: PDFs, audio files, websites, or YouTube videos.
Nevertheless, even in this phase, the question of necessary context remained unanswered. Context that could transform an AI from a prompt machine into an AI sparring partner.
We still managed long prompts. Additionally, we remained individual players. Of course, we exchanged ideas about which approaches worked reliably. The aspects of “memory” and “context” nevertheless remained limiting factors.
Our Change Management Learning: The transition from Phase 0 to Phase 1 often fails because tools are purchased centrally but not introduced strategically. Crucial: Define workflows first, then choose tools.
Phase 2: AI as a Personal Sparring Partner
We made a significant leap forward when we switched to Claude because we could implement a systematic approach.
Suddenly we could store personal preferences. Specifically: I could explain to the AI who I am, what my tasks are, and how it can support me.
Additionally, this was the moment when two of the biggest and simultaneously least visible deficits of AI systems could be overcome:
Uncertainty Matrix: In every chat and with every answer, the AI informs us how confident it is about its response. Instead of the usual tone of conviction that defined AI system answers, we now get insights into where the system is uncertain and how we can contribute to closing information gaps.
Opportunism: When working with an AI for extended periods, everyone eventually realizes that most AI systems are trained to support the user to the point where they’re clearly on their side. The AI then no longer says what is necessary, but what it believes the user wants to hear. We remedied this through a corresponding entry in the preferences. After all, we wanted a critical sparring partner who points out possible deficits to us.
Our Change Management Learning: In Phase 2, we achieved approximately 30% productivity increase in strategic tasks. The key: Define personal workflows and clear expectations for the AI.
Phase 3: Multiplayer AI Teams – “I” to “We”
The next major evolutionary step was collaboration on projects within an AI system. This made it possible to optimally use two of Claude’s capabilities. First, project-specific instructions can be defined for each project. Second, projects can be shared. This has brought us significant advantages:
Distributed Tasks: Instead of constant copy-paste, all relevant information can be stored in a project. For example, if we’re working on a blog post or white paper, one team member can do the research. Another team member continues working with the results and creates the texts. Finally, another team member takes over and creates the graphics together with Claude (the graphics on this page are also created with Claude and embedded as HTML code).
Artifacts: Copy-paste becomes superfluous because central information is stored in artifacts that all team members can access.
Memory: What to do when a chat reaches its capacity limit? Previously, we had to extract the chat and import it again as a PDF. Today, the AI has access to earlier chats and seamless continuation of work is possible.
The transition to “multiplayer mode” simultaneously became a team transformation for us, opening up many degrees of freedom and making us more efficient. Shared workflows and standardized prompts, along with cross-team AI strategies, have delivered real and measurable advantages. Specifically, this meant establishing an error culture (hallucinations are normal!), regularly questioning workflows and continuously re-documenting them, and sharing what works within the team.
Our Change Management Learning: The transition to Phase 3 doubled our productivity gain to 50-60%, especially in content creation and project coordination. Crucial: The entire team must be brought along.
Phase 4 (In Progress): Human-Led, Agent-Operated
As a team, we’re working on the next stage of AI evolution. We see this in the use of autonomous AI agents deployed specifically for operational tasks. Human team members then concentrate on strategy and leadership. This remains “work in progress,” but we’ll report on our progress.
Our Change Management Learning: Phase 4 requires fundamental rethinking: From “We use AI” to “AI works for us.” This is less a technical than a cultural challenge.
The transition to team-based AI use acts like an additional multiplier because not just individual team members but entire teams participate in the transformation process.
The Question of ROI
A noticeable ROI emerges at the latest when AI can be used as a personal sparring partner. For us, it showed:
- Phase 2: approximately 30% productivity increase in strategic tasks
- Phase 3: 50-60% efficiency gain through team collaboration
- Specifically measurable: From 10 days to 2 days for proposal creation
The transition to team-based AI use acts like an additional multiplier because not just individual team members but entire teams participate in the transformation process.
Furthermore…
There are plenty of other things we’ve learned to appreciate in working with AI systems.
For example, the research functions. Instead of starting directly with a prompt to solve a problem, the AI can search the internet for relevant information and prepare it as project knowledge. For us, this means we no longer need to pay attention to our own information bias. We have a large data pool within a short time that helps us illuminate questions from different perspectives. We’ve also developed workflows for this that we use daily.
Another example is branding. We’ve developed guidelines on how our wording and external appearance should be designed. We have the AI check every blog post and every other marketing product against these guidelines. This helps us recognize when we sound too alarmist or too vague.
And we’ve also found our own answer to the question of the “best AI tool”: A mix of tool agnosticism, because tool choice is often secondary, combined with a clear view of the specific challenges that need solving in each case. Notebook LLM is an example of how large amounts of data can be processed quickly. This is also possible with more generalist tools, just slightly better with this specific tool. As with almost all tools, it depends on the problem that needs solving as efficiently as possible. Which is why we decided on a tool mix.
What Comes Next?
Our journey is far from over. But we’ve already learned extensively along the way and we learn just as much daily in interactions with our customers. Constantly expanding this knowledge pool and remaining curious about new capabilities is what drives us.
Let's get into conversation
Are you facing the challenge of anchoring AI strategically in your company? Let’s talk about your individual transformation path.
Our promise: We know every step of this journey from our own experience – with all the ups and downs.
You can arrange a suitable appointment for a free initial consultation on our contact page.
Would you like to dive deeper into the strategic aspects of AI transformation? Read our article on Change Management as an underestimated success factor.
AI Evolution – Practical Experiences in AI Implementation
[…] This post is part of our “AI in Practice” series. We show how we use AI ourselves – not just in theory, but in everyday work. Because we believe: Those who offer AI consulting should also live it. Here you can find our blog about our own AI integration. […]