Operating model
Central intelligence: governed memory a company can reason with.
The question the company answered before
Stored knowledge is not yet usable judgement
A familiar question arrives in a meeting: why did we give this customer a different price last year? Someone searches the contract folder. Someone else looks through email. The person who remembers the discussion is away. The answer probably exists, but the reasons, authority and conditions around it are scattered across different records.
I use central intelligence to describe a governed layer that helps an organisation reason with those records. It retrieves relevant evidence, shows where it came from, connects it to decisions already made and identifies what still requires human judgement. It is not a single all-knowing brain. It is a way to make organisational memory usable without pretending that memory is automatically true.
The move beyond retrieval
Finding the document is only the first step
A chatbot, wiki and search system can all belong inside this layer. The distinction is not the interface. It is what happens after a source is found. Ordinary retrieval may return the contract, a policy page and three messages. Central intelligence should help compare them, distinguish the approved record from a draft, and explain which conclusion the evidence supports.
That extra step only earns trust when it preserves provenance, the trail back to the source. A useful answer says more than “the discount was approved.” It shows the approval, names the policy in force at the time and flags any conflicting record. The system may assist with synthesis, but the evidence remains inspectable and the accountable person remains visible.
Memory becomes useful through a governed reasoning loop
Retrieval is the beginning. A decision needs provenance, authority and a path back into operations.
- Retrieve relevant memory Select current policies, decisions and evidence.
- Show provenance Expose source, owner, date and review status.
- Reason in the current context Compare memory with the facts of this decision.
- Reserve authority A named human owns consequential judgement.
- Record outcome and trigger Store what changed, why and when it must be revisited.
A company can reason with memory only when the memory can also be challenged and corrected.
A worked organisational example
A decision needs evidence, an owner and a reason to reopen
Consider a hypothetical renewal. An account manager asks whether a twelve-percent discount can continue. The system retrieves the signed contract, the pricing policy that applied when it was approved and the decision record from finance. It explains that the exception was granted for a two-year commitment, cites each source and names the finance director as the owner of the decision.
The renewal is for one year, so the old decision does not automatically carry forward. The system flags the changed condition and routes the case for review. It can prepare the evidence and show the precedent. It should not quietly turn last year's exception into this year's policy.
This is harness engineering at organisational scale. The context spans systems and years. Permissions decide who may see commercial terms. A workflow decides when a human must approve. A durable record captures the new judgement and the conditions under which it should be reopened. The model helps assemble and compare; the organisation still owns the decision.
Where centralisation bends
One memory layer still needs many responsible owners
Central does not have to mean one database, one model or one team deciding what the company believes. That design becomes a bottleneck and can erase local knowledge. Legal, finance, sales and operations may need different source systems, permissions and review rhythms. What they share is a way to identify authoritative evidence, expose provenance and route uncertainty to the right owner.
The governance is the difficult part. Every important record needs an owner, an access rule and a reason to review it. Conflicting sources should be surfaced, not blended into a confident average. Sensitive evidence should remain invisible to people without permission. When the system cannot show why an answer is supported, it should say so and stop short of acting.
This is also where the idea can fail. Poor source material produces poor synthesis. A central layer can spread one mistaken policy faster than a disconnected wiki ever could. Teams may stop challenging inherited choices because the answer sounds official. Those are the governance costs behind the company brain, and they are reasons to design for challenge and revision rather than obedience.
The practical prize
Record the judgement once, reuse it until the facts change
The goal is not a company that decides each question once and never looks back. Good judgement changes when evidence, assumptions or responsibilities change. The goal is to stop rediscovering the same reasoning because nobody preserved it.
A useful decision record captures the conclusion, supporting evidence, accountable owner, date and review trigger. The next person can reuse that judgement while its conditions still hold, or reopen it with a clear view of what changed. This is the practical promise of the company brain: not an organisation that cannot forget, but one that can inspect what it remembers.
I think of that as a student as much as a systems builder. Learning is not keeping every answer unchanged. It is preserving why you believed something, noticing when the world no longer matches it and revising with evidence. A company deserves the same discipline. Central intelligence is useful when it helps people spend less attention reconstructing the past and more judgement deciding what should happen next.
A decision record is a living contract
Reuse is valuable only while the recorded choice remains current for the situation.
- Decision What was chosen and which alternatives were rejected.
- Evidence and rationale Why the choice made sense at the time.
- Owner and scope Who can change it and where it applies.
- Review trigger Date, threshold, incident or external change that reopens it.
- Supersession trail What replaced the decision without erasing its history.
Decide once means avoid needless repetition, not freeze judgement forever.