Open social innovation thrives on collective intelligence. But collective intelligence does not emerge by itself: it needs structure, continuity, and the ability to accompany many actors at once, before, between, and after the moments when people actually come together.
This is where Deep Inno comes in, as an additional partner in the process. The agentic AI system helps a team sharpen its problem statement before a co-creation workshop, so that time together is spent on substance rather than groundwork. It keeps reflection going between events, when momentum typically fades. And because it accompanies many teams in parallel, it can surface patterns and connections that no single facilitator could see.
For us at Dezentrum and the Future Urban Society team, the guiding question was never whether AI can replace innovation. It was: can AI help teams and facilitators ask better questions in the innovation process?
Good Questions, instead of shallow answers
Deep Inno supports innovation teams with targeted questions, drawing on methods from systemic problem exploration and ideation. Early in the process, the system uncovers gaps in an idea: Which barriers have not been considered yet? Which stakeholders are missing from the picture? Later on, it supports teams in developing their project and creates clarity, for instance when refining hypotheses or preparing proposals.
The content strategy sets Deep Inno apart from generic chatbots.
Language models alone are not predictable; they tend to move on before a problem is truly understood. Deep Inno makes that judgment explicit in code, keeping the process traceable and auditable.
Architecture: The main components
Deterministic components, agents with specific jobs
Technically, Deep Inno is a multi-agent workflow built in n8n. It combines deterministic components for routing and evaluation with specialized agents, each with a clearly defined job along the innovation process, from problem exploration through ideation to proposal development.
The agents make the system adaptable. Because each has a specific, well-bounded task, the setup can be reconfigured quickly to fit new contexts: different funding logics, languages, methodologies, or thematic focus areas. Wherever ideas and projects need sharpening, Deep Inno can be tailored with little effort. Essentially anywhere.
The deterministic components make it trustworthy. Because routing and evaluation run as code, every decision in the process is auditable: partners can see why a team was moved to the next phase, and the same input leads to the same assessment. Criteria can be inspected, discussed, and adapted to an organization's own standards instead of being buried in a model's behavior. For funding organizations that need to justify their processes, this reproducibility is the difference between an experiment and an instrument.
A further characteristic of the setup: all system components were chosen so that the whole stack can be switched to a self-hosted solution at any time. The intelligence engine itself is easily replaceable. A sovereign setup with open models in Swiss data centers has already been designed, with no prompt logging and no lock-ins. Costs only grow with actual needs: from standard operation to periodic fine-tuning to a dedicated model.
Testing and user feedback
At InnoDay 2026, we tested Deep Inno for the first time with a broad audience: experts from public and private innovation agencies. Participants worked live with the system on two case studies we had prepared in advance, so that every team had a concrete starting point and could test the system without first developing an idea of their own. Case A described an early-stage idea in local energy, with many open questions and unclear barriers; Case B a later-stage venture with a working platform whose scaling had stalled. In both cases, the task was the same: find a blind spot.
The demo supported our thesis: Deep Inno helps teams ask better questions earlier. By that we mean the right questions at the right moment, questions that point a team toward improving its innovation idea, whatever stage it is at: clarifying an unclear problem early on, or surfacing an overlooked system barrier later in the process.
The feedback helped to further develop the system: from bug fixes to the insight that a question-asking system needs a proper introduction, to the principle of generating more depth instead of speed.
Since then, we have been in talks with several innovation promotion organizations about individualized prototypes in which phases, agents, and evaluations are tuned to their specific language, funding logic, and priorities.
Outlook: From experiment to development infrastructure
Deep Inno is continuously being improved. Insights from every deployment flow directly into the next version, and we are step by step exploring how the agentic AI system and physical formats can best complement each other across the innovation cycle.
Behind this lies a larger conviction: social innovation needs not only project funding, but development infrastructure and ecosystems that truly support collective intelligence. Deep Inno is a small experiment within this larger question: Are we funding solutions today, or also the infrastructures to keep developing better ones tomorrow?
We personalize the system to your funding logic and priorities. Get in touch.
