Short answer
Why do companies keep repeating mistakes? The problem may not be a lack of data but forgetting why decisions were made. Harry Lui explains how to build enterprise memory, decision cards and strategy-drift checks.
Many companies keep large volumes of documents yet still repeat the same discussions.
Management may remember what the company did but have forgotten why; staff see the final decision without knowing the constraints, risks and options that were set aside. When an owner leaves, a team is reorganised or market pressure shifts, old problems return under new names.
This is not simply a document-management problem. The company has lost its own memory of judgement.
Many documents do not mean the company has a memory
Meeting minutes usually record who said what, reports record results, and policies record current rules. What really shapes future judgement is the relationship between them: what problem was faced, what information was available, what constraints applied, which options were considered, and why a particular cost was finally accepted.
If this is not kept, the company can later see only outcomes and cannot rebuild the context of decisions. A new manager may wrongly assume an old decision made no sense; the original decision-maker, having forgotten the conditions, may walk back down a path already shown not to work.
One high-value use of AI is helping a company turn scattered records into a decision context that can be verified and continually updated.
Enterprise memory should keep four things
Original evidence
This includes meeting records, customer feedback, operating data, proposal versions and formal decisions that are authorised to be kept. Summaries cannot replace originals, because later you need to know where a conclusion came from.
Decision cards
Every major decision should have a concise decision card recording context, goal, constraints, options, final choice, owner, expected result, review date and stop conditions. Unknown information should be marked as unknown, not filled in by AI to make a complete story.
Stage snapshots
Each month or quarter, a company should keep a record of the business situation, priorities, product versions, organisation, unverified assumptions and next-stage tasks. This avoids rewriting today’s understanding as something that was already known at the time.
Outcome feedback
A proposal being raised, approved, formally executed, winning orders, delivered successfully and being repeatable are entirely different stages. Enterprise memory must keep actual outcomes; otherwise a well-written proposal can easily be mistaken later for a proven capability.
AI can act as a strategy-drift monitor
Only with a continuous record of decisions can AI help management ask an important question: are the company’s recent actions still executing the original strategy, or have they drifted without explanation?
Drift is not necessarily wrong. Market change, customer evidence or cash-flow pressure may all require a change of direction. The real risk is that the company has already changed direction without reconfirming the reasons, resources and costs.
A practical strategy check can compare, in turn: what was originally decided, what it was based on, what has happened recently, which conditions have changed, whether new evidence has emerged, and which existing commitments the new actions affect.
AI organises the differences and points out gaps; management confirms the trade-offs. In this way, company history does not become baggage that blocks change but a basis for better-quality change.
Hong Kong SMEs need enterprise memory even more
Large organisations usually have fuller policies, committees and document management; in Hong Kong SMEs, major judgements are often concentrated in the founder and a few core colleagues. This works very well early on because decisions are fast and communication short; but as products grow, people move or handovers begin, three problems easily appear: everything goes back to the boss, the same mistake recurs, and different departments remember different versions.
Building enterprise memory does not mean buying an expensive system first. The minimum viable approach is to start with the three most recent major decisions: keep the original evidence, create decision cards, and check outcomes on a set date.
As this habit continues, the company gradually forms a basis for judgement that does not depend entirely on individuals.
The real value of a knowledge base is knowing why the company is where it is
The market often understands an AI knowledge base as “staff can ask questions about company information”. That is only the first layer.
A higher level of enterprise knowledge management lets the company answer: why did we choose these customers? Why did we stop a product? Why did we adopt this partnership model? Which constraints from that time still apply? What new evidence would justify a change?
When it can answer these questions, a company does not merely own data; it owns continuous management wisdom.
Practical supplement
Six handover questions
Record each question using one real piece of work and link back to the original evidence; mark anything unconfirmed as to be confirmed. This is a practical extension of the enterprise memory principle.
Your entries remain only on this page and will not be saved after you refresh or leave; use Copy worksheet to keep the text.
Discussion question
During handovers, which part of the decision context does your team most often lack?
Smark Global perspective
Before adopting AI, a company should organise its important knowledge, decisions and workflows into data structures that can be authorised, verified and improved.
Learn more: Smark Global AI Academy →Source: Original commentary by Harry Lui
Last updated: 2026-10-02（first published: 2026-10-02）