We constantly talk about the AI revolution, but in practice, most medium-sized companies use the technology to optimize outdated processes.
Henry Ford once said that if he had asked people how to improve transportation before the invention of the car, most would have asked for a faster horse.
He understood a fundamental human weakness: our brains naturally take the path of least resistance. When we sit down to innovate, we subconsciously base our ideas on the experiences and systems we already know. The result is that our âinnovationâ is quickly limited to simply electrifying existing habits.
In my daily work as a digital marketing specialist in the financial sector, I see this exact pattern unfolding right now. We talk endlessly about how AI will turn the world upside down. But the reality is that in the vast majority of companies, AI is “just” about simple process optimization. People build an extra CustomGPT, set up a new Copilot Agent, or launch an internal Gembot that lets employees solve the exact same tasks as yesterday â just a little bit faster.
In other words, we are building faster horses.
If we want to exploit the true potential and avoid being overtaken by those who dare to invent the car, we need to dig deeper.
Constraints are the fuel of creativity
But how do we break the deadlock of habit-thinking in a busy marketing or IT department?
Ironically, the answer is rarely total freedom. A blank piece of paper and an open prompt box can paralyze any organization, while sharp constraints force the brain to think in entirely new ways. Creativity must be provoked to wake up.
Next time you gather the team to develop new solutions, try setting up radical thought experiments with a simple âwhat ifâ:
- What if we weren’t allowed to use our existing martech systems at all from tomorrow? How would we build lead generation from scratch?
- What if we had to deliver the same result, but the marketing budget was cut by 90 percent?
- What if an AI model had to drive the entire customer journey autonomously? How would we redesign data collection and touchpoints to enable that?
The alternative to these exercises is to work consciously with innovation crossings. This is where we combine AI capabilities with other existing technologies to create an entirely new value proposition, rather than just optimizing a single step in a process.
The challenge with Citizen Development and poor data discipline
Let me give a concrete, technical example from my own daily life of what such an innovation crossing can look like.
Many larger companies have embraced citizen development â the idea that end-users can build their own Copilot Agents to solve their specific departmental needs. On paper, itâs a fantastic democratization of technology.
However, my personal realization is that it often fails in practice. Why? Because people’s data discipline and ability to prompt vary incredibly. When the input is diffuse, the output from the Agent follows suit, and then the adoption rate drops drastically. People lose trust in the tool.
The solution: A controlled innovation crossing
To solve that problem, I recently built an internal Martech Toolbox for our in-house marketing department, which now includes a home-built AI chat.
Here, we haven’t just given users an open chat interface. Instead, weâve built a system that enforces data discipline before the free conversation with the AI begins. The solution is based on an integration with Google Vertex â an AI API that can be used by developers (and vibe-coders like me) as an âAI componentâ to build AI into services, data processing, and much more.
Instead of letting the user start the conversation with an empty text field, I chose to have the conversation with the AI initiated via an âold-fashioned,â dynamic form.
This is how the flow works:
- Qualification Through the form, the user is forced to decide on the precise task type (e.g., SEO optimization, content drafting, or data analysis).
- Structured input The form adapts dynamically and requests all necessary variables in dedicated fields (target audience, tone-of-voice, existing data sources, etc.).
- AI activation Only once the structured data is collected is a well-defined payload sent to the Vertex API, and the actual conversation with the AI begins.
Because the AI is now fed with a razor-sharp, structured starting point, the quality of the output is sky-high from the first response. At the same time, we have built a backend structure that makes it quick and easy to set up new task types, while ensuring central learning and logging across all the department’s users.
Itâs a classic innovation crossing: We take a well-known, strict data collection method (the form) and combine it with advanced generative AI (Google Vertex).
This is where we force ourselves out of the horse-drawn carriage, away from messy citizen development, and into the future.
FAQ
It is the intersection where you take two or more existing technologies, processes, or concepts and combine them to create an entirely new value proposition â rather than just optimizing an existing silo. It is also called combinatorial innovation.
Historically, the biggest breakthroughs have occurred in exactly these intersections.
The Printing Press: Johannes Gutenberg didn’t invent printing. His stroke of genius was an innovation crossing where he took a traditional wine press from agriculture and combined it with movable lead type and oil-based ink. It created mass communication.
The Smartphone: When the first iPhone was launched, it wasn’t just an improved phone with better buttons. It was a deliberate cross between a mobile phone, a music player, and an internet browser.
Uber: They didn’t invent the taxi or carpooling. They made an innovation crossing between an established service industry and the (then) new GPS and smartphone technology, creating a completely new business model.
In a modern martech context, this means you shouldn’t just use an LLM to write emails a little faster. You should look for the crossing where advanced AI is combined with, for example, strict, dynamic forms and your ERP system to create a fully automated problem solver that wasn’t technically possible to build yesterday.
Open AI agents can be fine… but when we base our AI initiatives on open agents, we effectively make the quality of the work very dependent on the individual employee’s data discipline and technical capability. The ability to write the perfect initial prompt and feed the AI with the right context varies incredibly from person to person.
An open chat interface basically functions as a blank piece of paper. For the few power users in the company, itâs a fantastic, free playground, but for the vast majority of employees, the total lack of structure can be paralyzing. If the user doesn’t know in advance which specific datasets, formats, or variables the model needs, the output quickly becomes superficial, generic, or imprecise.
This often means that the adoption rate and excitement drop significantly as soon as the honeymoon period is over. When people experience that they can’t rely on getting a consistent and useful result every time, they fall back into old habits and systems.
In this way, the potential of agents is limited to merely functioning as glorified spell checkers or for writing a quick email.
An API-driven approach via Google Vertex (or similar enterprise models) gives you full control over the backend, security policies, logging, and the ability to build custom frontends (like forms) that tightly control the user journey.
