This summer we spent 4 weeks building Precuro Oracle, a living AI system that has become central to how Precuro operates. It is not a folder structure or a static knowledge base. It runs on tools we already had, such as Claude and Notion, with no coding or custom development involved. It automatically captures, connects, and makes practical use of the information we generate across research, customer conversations, product development, claims practice, regulation, and market work staying current as we work and feeding every AI task we run.
For a very small company building a new insurance-specific risk category, this matters. We are not applying an established model to insurance; we are building Cognitive Risk Intelligence in an industry where no ready-made category, buyer language, or implementation model exists.As we keep learning, we cannot afford to leave that in inboxes, meeting notes and our heads. Every customer conversation, research finding and regulatory constraint must inform the next decision we make.
From compounding insight to opportunity
The recent pain-triage breakthrough shows what Oracle makes possible.
In a few focused weeks, we moved from an initial question — whether Precuro’s cognitive model could identify clinically meaningful pain-rehabilitation patterns — to a focused, globally relevant new application of our technology.
Oracle supported the sprint at every stage, with our clinical and commercial judgement applied to any output it produced.
Structuring the initial brainstorming and testing the strategic hypothesis.
Comparing established psychological pain assessments with Precuro’s cognitive model.
Supporting research design, evidence review, and analysis.
Formulating the clinical and commercial case for early pain triage.
Identifying priority customer groups, including insurers, rehabilitation providers, specialist pain clinics, and private care organisations.
Mapping relevant organisations and the senior clinical, research, rehabilitation, and claims leaders to approach.
Developing accurate, audience-specific material for clinical, insurance, and investor conversations.
The result: a new proposition in which seven short, non-clinical questions can identify fear-avoidance and boom–bust patterns early in a pain pathway, then guide people toward scalable digital cognitive and behavioural support matched to their needs.
Built for scale
Oracle turns scattered information into usable intelligence. It helps us retain what we learn, separate evidence from assumption, and act while the opportunity is still alive. Its value is not the volume of stored content but the connection between research evidence, customer signals, market context, regulatory constraints and product logic, held as a single evolving view of what Precuro should build next.
Precuro Oracle does not replace human judgement, relationships, or rigorous research. It makes them cumulative.
As Precuro moves across insurance lines and geographies, what we have learned travels with us. For a small team in a complex, heavily regulated market, that is a real strategic advantage. Less time reconstructing what we already know, and more time turning validated insight into applications.
Why we built Oracle this way
The system was conceived and developed by Jelena Marjanovic, PhD, AI Adoption Consultant, in close collaboration with our team. Building Oracle at this speed, tailored to Precuro's highly specific needs and capable of producing consistently precise, reliable output, reflects both the depth of Jelena's expertise and her exceptional ability to turn complex knowledge work into a practical operating system.
We cannot think of another individual who could have built anything comparable with us. The sophistication is in the design. Oracle is built from tools we already had, with the connections between them drawn from Jelena's business analysis of how Precuro works, and it is deeply aligned with how Precuro researches, validates, prioritises and builds.
Our team adds to it continuously from day-to-day research, customer work and market development, with Jelena advising as we extend it.
We asked Jelena to explain the thinking behind Oracle, the mistakes she sees most often, and why this applies well beyond insurance.
In her words: The most common mistake I see is treating generative AI as a source of truth. It is actually a prediction engine. Ask it for a business proposal and it predicts what a generic proposal looks like, because it knows a great deal about the world and nothing about your business, besides the limited context you can realistically feed it in one session.
The usual response is better prompting, which helps to an extent only but doesn't solve the limited knowledge problem.
One field where the cost of this is visible is Venture Capital. Andre Retterath, a GP at Earlybird, wrote in August that "the gap isn't the model, it's everything the firm has built around it." Most of what a firm knows, he says, "sits in people's heads, scattered documents, and individual chat histories." Public databases show which rounds closed. Only the fund knows which rounds were attempted and failed, and why. That knowledge decides whether the next deal is a good one, and no AI can reach it.
A living knowledge base like Oracle closes that gap. It is where the firm's knowledge is written down in a form the AI reads before every task, so the output starts from what you know rather than from what the model guesses.
If you want to build your own version of Oracle, you need three things. This is the approach I take with every client:
The first is a business analysis of your workflows and your goals, so you know which processes to prioritise and which use cases return the most for the least effort.
The second is getting the knowledge out of people's heads. Most of it was never written down. You look at a draft and know within seconds something is off, but you struggle to say why. Extracting that judgement takes structured questioning, usually by someone outside the team, because the people in the workflows stopped noticing what they know.
The third is building capability in the team, so your people keep finding use cases and improving the output after the engagement ends.
Human judgement stays designed into the system throughout. That is why we built Oracle this way.
Jelena’s contact:
Jelena Marjanovic Consulting
AI at Work
https://www.linkedin.com/in/marjanovicjelena/