Where do you stand on AI? Discover our five phases.
AI is everywhere—as hype, as an opportunity, as a risk, and sometimes even as a silver bullet. At Cube, we prefer to look at what’s really happening. Because behind the scenes—and increasingly, in the spotlight as well—we see that many organizations are going through the same evolution. From cautious experimentation to AI becoming an integral part of software and processes. Where do you stand right now? And what step can you take today? In our blog, we’ll walk you through the phases of AI adoption and show you what’s already possible.
Phase 1 - AI on one's own initiative.
Colleagues are using AI tools, but no one knows exactly who is using what and for what purpose. AI comes up as a topic of conversation in meetings, but it isn’t listed as a project on the agenda anywhere. Three departments are experimenting independently of one another, with no one coordinating what’s happening. And when asked, “What are we actually solving with this?” there’s silence.
What step can you take right now?
Start with a baseline assessment: Which AI tools are your colleagues already using, for what purposes, and what information are they sharing? That insight forms the basis for clear agreements and policies. Next, choose one team where AI is already being actively used. That’s often where the most energy is focused, and you’ll quickly learn what works in practice. For one process, describe what’s currently going wrong, how AI can improve it, and how success can be measured. This is how you make AI concrete, safe, and applicable.
The value of AI at this stage.
At this stage, the value of AI lies primarily in direct support. AI takes over small, repetitive tasks, such as drafting an email, summarizing a document, or filling in code. This allows you to quickly identify where time is being wasted and where AI can immediately ease the burden of your daily work.
Phase 2 - Initial rules and guidelines.
There’s an AI policy on the intranet, but few people know what’s in it. Tools are either allowed or prohibited, but the nuances in between are missing. Colleagues ask themselves, “Can I use ChatGPT for this?” But those questions aren’t being addressed anywhere. Meanwhile, one colleague has a prompt that lets him finish his work two hours faster. The rest of us don’t know anything about it.
What step can you take right now?
Turn your AI guidelines into a version that colleagues will actually use. Avoid legal jargon and be clear: what’s allowed, what isn’t, and when should you be careful? Also, create a single place where colleagues can share prompts, questions, and experiences. This will form the basis for a shared AI library. Also, schedule a regular review of your policy—for example, every six months. AI is evolving rapidly, so your guidelines need to be flexible enough to adapt.
The value of AI at this stage.
At this stage, AI becomes accessible to more people than just early adopters. With clear guidelines, colleagues know what is and isn’t allowed, enabling them to use AI with greater confidence. This broadens its use and reduces the risk of errors.
Phase 3 - Shared skills, working methods, and pilot projects.
One team has a working AI pilot, but there is no clear path to a broader rollout. The prompt library and use cases are there, but no one owns them. As a result, they quickly become outdated. Knowledge is also fading away. Training sessions have been held, but two months later, much of it has already been forgotten. And sometimes the first pilot is immediately applied to a strategic process, even though the people who will have to work with it haven’t been properly involved yet. As a result, the initiative falls by the wayside before AI can truly deliver value.
What step can you take?
Appoint an AI ambassador for each team: someone who knows the work and understands how colleagues use AI. This way, knowledge doesn’t get stuck with just a few early adopters. Next, choose one working pilot and define the rollout in concrete terms: who will be working with it, what data is needed, and what problem is it supposed to solve? Collect use cases that deliver value and discuss monthly what didn’t work with AI. Choose the next pilot for a task that involves a lot of repetition or frustration. That’s where AI is most likely to be put to real use.
The value of AI at this stage.
At this stage, you can see the structural benefits AI delivers—not just what it promises in a demo. By running a pilot using your own data, you make that value measurable, repeatable, and relevant to real-world applications.
Not sure where you stand or what you're already capable of? Let's talk.
Phase 4 - AI in production, but disconnected.
We have an internal AI assistant that runs on our own documentation, but no one is sure if the answers are always correct. In addition to the official AI features, colleagues also use their own versions of Claude or ChatGPT. Meanwhile, sales, marketing, and support are each building their own AI solutions: three isolated silos that don’t communicate with one another. The features work, but there’s no overall view.
What step can you take right now?
Track what happens for each AI feature: how many questions are asked, how satisfied users are, and how often answers are still being verified. This way, you can see not only whether a feature is being used, but also whether it’s reliable enough. Also, keep track of what your colleagues are still doing in their own Claude or ChatGPT. That list shows where there still seems to be a need. For each feature, map out which data and links are being used and who is responsible for them. That overview forms the basis for the next step.
The value of AI at this stage.
In this phase, AI is integrated into the software people are already using. Not as an additional tool alongside it, but right where the work is done. This eliminates the need to switch between systems and makes AI an integral part of the process. This is also where the difference between SaaS and custom software becomes apparent. While off-the-shelf software often adds AI as a separate feature, custom software allows AI to work exactly where it’s needed in the process. That’s why Cube doesn’t build AI alongside your software—it builds it right into it.
Phase 5 - A strong foundation with our own AI architecture.
In this phase, AI is no longer operated as a standalone feature or pilot, but on a collaborative basis. Data is made centrally accessible via a Data Management Platform . On top of that, an MCP layer determines, on a per-role basis, what AI is allowed to see and do. This gives AI secure access to systems, knowledge, and actions, without requiring a separate integration for every new application. The question thus shifts. It is no longer just a matter of “Is this allowed?” or “Does this work?”, but rather “How do we maintain control as AI supports more and more processes?”
What step can you take right now?
First, map out the foundation. Which data sources are connected? Which roles are authorized to view which information? Which actions are logged? And where does a human need to monitor or approve the process? Next, check whether new AI features can build on the same foundation. Can you switch the underlying model without causing functions to fail? And is it clear who is responsible for access, data, and process quality?
What are the benefits of AI in this context?
At this stage, AI agents can independently carry out more complex processes based on an existing foundation. New applications build on that foundation, access is managed by role, and you can scale without being locked into a single vendor. This way, AI grows alongside your organization without you losing control.
Ready to implement AI on a structural basis? So are we.
Often, an organization is spread across two or three phases at the same time. Which step makes the most sense and will yield results the fastest depends on where the focus lies. An hour of brainstorming usually provides more direction, and we’d be happy to brainstorm with you.
Worth reading next...
Vibe coding: what it is, how it works, and how reliable it is.
Digitizing processes: 7 steps to start without chaos
MCP integration: connect AI securely to your business systems.
Questions? No problem.
Most organizations go through five phases: from standalone AI tools implemented on their own initiative, through policy and pilot projects, to AI integrated into existing software, and ultimately to their own AI architecture with a Data Management Platform and an MCP layer.
SaaS AI works alongside existing systems. Custom solutions integrate AI directly into the software people already use, resulting in fewer switches and more useful output.
What is an MCP layer, and why is it relevant to AI?
AI needs to work with your data, your processes, and your systems. Off-the-shelf tools are quick to set up, but they operate alongside your software rather than within it. Custom solutions integrate AI directly into the applications people already use, which eliminates the need to switch between systems and delivers better results.