Institutional Knowledge in Banking: Why AI Makes It More Valuable
CCO Managing Partner
For most of my career in banking, knowledge was a real source of advantage.
Not insider information. Market research. Understanding client behaviour. Seeing structural shifts earlier than others. Banks that could turn observation into insight created value both for themselves and for their clients. Information asymmetry was a legitimate competitive edge.
I experienced this directly at Werthstein. One of our differentiators was giving clients structured access to the latest knowledge and market insights. On 22 March 2017, we published an investment theme we called “Zeitgeist Driverless Cars”. One of the key stocks in the basket was NVIDIA. The share price that day was roughly $2.70 on a split-adjusted basis. Today, it trades at around $229. The video is still available on YouTube if anyone wants to watch it.
That is what early, usable knowledge can do.
AI Has Made General Knowledge Abundant
General knowledge is no longer scarce. Anyone with a decent model can access macro analysis, sector overviews, historical data and coherent reasoning in seconds.
The old arbitrage on publicly available information is disappearing fast. What used to take a research team days can now be produced in minutes, and often with greater consistency.
This does not make knowledge irrelevant. It changes which knowledge still carries value.
What Knowledge Is Becoming Scarce in Banking
The scarce resource is no longer general insight. It is the specific knowledge that lives inside the institution.
The institutional memory. The definitions that were never fully written down. The judgment rules experienced people carry in their heads. The way a particular client segment actually behaves over time. The exceptions that never made it into the official process handbook. The proprietary data that reflects real decisions, real trade-offs and real outcomes across years.
This knowledge still exists in most banks. It is just fragmented, spread across systems, local files, email threads and people. When those people leave or move roles, parts of it disappear. When different teams use slightly different definitions of the same concept, the organisation quietly operates with multiple versions of the truth.
The problem is becoming more urgent. Large restructuring programmes release not only people but also the knowledge they carry. Even without layoffs, an aging society means experienced colleagues are leaving.
In both cases, the organisation loses specific knowledge that was never fully captured. Once it is gone, it is extremely hard to reconstruct.
Why Institutional Knowledge Matters More with AI
Humans could often paper over inconsistency with context and experience. Agents cannot.
They operate on whatever foundation they are given. If the underlying definitions, rules and institutional knowledge remain fragmented, agents simply scale the ambiguity, faster and with more confidence.
This is the quiet risk many institutions underestimate. The technology is ready. The models are capable. The bottleneck is the quality and accessibility of the specific knowledge the models are allowed to work with.
What a Knowledge Layer for AI Needs to Do
A usable knowledge layer is not a glorified document repository or another SharePoint site. It has to solve three practical problems at once.
First, it must make proprietary knowledge explicit and structured. This includes business definitions, decision rules, process logic, client segment characteristics and the informal judgment that experienced people apply every day. If it stays only in people’s heads, it cannot be reused by colleagues or by agents.
Second, it must create clear ownership. Someone has to be accountable for key definitions and for keeping them current. Without ownership, knowledge decays. Different teams quietly create their own versions, and the organisation loses a single source of meaning.
Third, it must be usable by both humans and machines. People need to find and trust the knowledge. Agents need consistent, versioned and permission-aware access so they can reason on top of it instead of guessing.
This is the direction we have taken with HICO VAULT Knowledge Management. The goal is not to create more documentation. The goal is to turn institutional knowledge into governed, reusable infrastructure that both employees and AI agents can work with, while keeping control, auditability and clear ownership in place.
How Banks Can Start Building a Knowledge Layer
Most successful approaches we see do not begin with a multi-year transformation programme. They start smaller and more concrete. They pick one high-value domain, for example advice preparation, credit processes or management reporting, and make the critical knowledge in that domain explicit.
They define the key concepts, document the decision rules that actually matter and assign ownership. Only then do they connect agents or advanced analytics on top of that foundation. They also treat knowledge as something that needs governance, not just collection.
Versioning, access rights and the ability to see where a definition came from become as important as the content itself. Without this, the knowledge layer slowly turns into another uncontrolled data swamp. Finally, they accept that the work is never fully finished.
Institutional knowledge evolves. Client behaviour changes. Regulation shifts. A static knowledge base becomes outdated almost as quickly as it is built. The real capability is the ability to keep the knowledge living, owned and usable over time.
AI Is Changing What Knowledge Is Valuable
The scarcity has moved. General knowledge is abundant. Specific institutional knowledge is not. Banks that recognise this early and invest in making their own knowledge usable will compound advantages. Those that continue to apply powerful models to incomplete or inconsistent context will keep producing impressive outputs that still feel unreliable when real accountability is required.
Knowledge was always an advantage in banking. The form of that advantage is simply changing.
If you are looking at how to make your bank’s institutional knowledge usable for both employees and AI agents, book a 30-minute exploration call with us. We can look at where critical knowledge sits today, where it is at risk of being lost and which domain would be the most practical place to start.