A business decision needs a memory

An investment proposal may arrive as a spreadsheet. A customer insight sits in an email. The objection that could change the decision appears halfway through a meeting. When the team returns to the question a month later, somebody has to reconstruct the reasoning.

That is the problem I want to work on with Decision Ground. I am developing a workspace for business owners and their teams that keeps the question, evidence, options and eventual decision together. An early prototype is taking shape. The pitch shows what we plan to test next.

What needs deciding?

Consider a hypothetical SME deciding whether to use AI to prepare customer quotations. “Which model should we buy?” is only part of the discussion. We first need to know which part of the work causes difficulty, which data may be used and who can judge an acceptable result.

A limited trial might answer more than another broad presentation. The team could compare preparation time, corrections and actual use before deciding whether to continue. The criteria need to be recorded before the result arrives.

Research should expose the gaps

The planned Deep Research function starts with a defined question and scope. It should gather relevant sources, retain counterarguments and show where evidence is thin. A useful dossier gives a person something to inspect. It should never turn a weak source into a confident recommendation through polished wording.

The same principle applies to the decision barometer. I want to explore Jev by TypeSafe as a way to make structured, gradual judgements: goal fit, evidence strength or reversibility. These are separate dimensions. They should not be collapsed into an impressive-looking percentage of business success. The model still needs evaluation for our tasks and language.

A second brain that learns and stays current

Knowledge becomes more useful when a team can connect it to what actually happened. It can also become misleading when an old assumption keeps returning as a fact.

The planned second brain should take in new evidence, decisions and later results, so the knowledge stays current. It should flag assumptions that no longer hold. Personal notes stay private. For the shared wiki, the system proposes changes with a source and date; the team reviews them before everyone relies on them. That is how it learns from experience without building a hidden personal profile.

Keep the route open

I want the same decision record to be accessible from compatible AI applications, including Mistral Vibe, ChatGPT, Claude, Copilot Studio and local open models through Ollama. MCP is the intended connection layer. Availability will depend on each application and its permissions.

German hosting and explicit control of data routes are requirements, not claims about an already deployed service. Local models are an interesting option to test, particularly for conversations with organisations that want to keep sensitive information on their own machines. A cloud decision model and a local model need separate evaluation.

The next test

The useful test is a complete decision cycle: an open question, a reviewed decision and a later look at what happened. Then a second decision, to see whether the recorded learning helps. We do not yet have evidence that the product improves outcomes.

The questions can be about much more than technology: an offer, a partner, a process or an investment.

Explore the Decision Ground prototype and pitch. What’s on your mind as a business owner? You can book a 30-minute conversation with me from that page.