Short answer
Real company culture is shaped by rewards, authority and accountability. Harry Lui explains how Hong Kong companies can work back from customer outcomes to capabilities, behaviours, reward allocation and AI governance.
Many companies display “innovation, collaboration, customer first”, but what staff really learn is usually not the words on the wall. They watch what the company actually rewards, what it tolerates, and who pays the price when something goes wrong.
A company may verbally encourage staff to raise risks while in practice rewarding only fast deal signing; it may ask the delivery team to own results while not allowing them to refuse unreasonable commitments. Over time, the team naturally chooses to hide bad news and push problems to the end, because that is the behaviour the system really encourages.
Once AI enters the company, this contradiction becomes even more visible. AI can speed up analysis and execution, but it cannot resolve management’s conflicting goals.
Organisation design should work back from customer outcomes
Before building culture, a company should answer a more concrete question: for which customers does it promise to deliver what verifiable result, and within what time?
From that result you can work back to the products, workflows, professional capabilities, collaboration and decision rights required. If the product promise itself is unclear, however hard the team works it will pull in different directions; if resources are insufficient, blaming failure on “not a good enough attitude” is unfair and will not improve results.
A complete governance chain can be understood like this:
Customer outcome → Product promise → Team capability → Daily behaviour → Reward allocation → Policies and procedures → Monitoring and correction
Culture sits in the middle of this chain. It explains which behaviours should take priority when trade-offs arise; rewards and systems decide whether those behaviours keep appearing.
Rewrite values as observable behaviours
For example, if a company says it values “growth”, it should not judge this by whether staff are willing to admit fault or accept criticism. More reliable behaviours are: acknowledging what is unknown, understanding evidence, revising work, testing new methods and reducing repeated mistakes.
If a company says it values “responsibility”, that does not mean asking staff to carry unlimited burdens. More concrete behaviours are: raising risks early, proposing alternatives, moving forward within one’s authority, and escalating when something goes beyond it.
Only when principles can be observed can management fairly give feedback, authority, training, bonuses or promotion. Otherwise “culture fit” easily becomes a subjective judgement of personal style and relationships.
A company’s real culture is hidden in its allocation system
Allocation is not only salary and bonuses; it also includes customer opportunities, important projects, learning resources, decision space and promotion.
If a company rewards only final revenue without accounting for refunds, delivery costs, customer complaints and cross-department support, the system is in effect encouraging short-term deals. If it rewards only “firefighting heroes” and not those who spot risks early and build processes, there may be more and more fires.
Good allocation design needs to answer: which contributions are recognised, what evidence supports them, who approves, when rewards are paid, how exceptions are handled, and how staff can request a review.
Most importantly, the same principles must apply to founders, trusted insiders, senior staff and top performers. Otherwise the team will quickly learn that the system is used to manage only some people.
AI suits monitoring, not acting as judge
AI can help compare work records, find missed commitments, flag whether similar incidents were handled differently, and organise staff explanations for managers to review.
But AI should not secretly assess loyalty, infer personality from tone, or automatically decide pay cuts, promotions or dismissals. Important decisions about people must keep clear criteria, relevant evidence, a chance for the person to explain, and substantive human judgement.
If a company uses AI to process staff or customer data, it also needs to limit purpose, who has access and retention periods. The aim of governance is not to watch more, but to make important decisions more consistent and traceable.
Fix one real break in the system first
A Hong Kong SME does not need to build a large HR system all at once. Management can pick one recent real incident: for example, sales promising something the delivery team could not do, a colleague raising a risk early but not being supported, or the same mistake causing repeated rework.
First reconstruct the facts, then check the customer outcome, responsibilities, authority, resources, rewards and review procedures. Find the single most critical break, trial one fix, and after thirty days check the effect using delivery quality, rework, decision time or customer outcomes.
Building culture is not writing one more slogan; it is making sure the behaviours the company wants are supported by its real systems.
AI can make this system easier to record and check, but management still has to decide first what result the company really wants, and which principles it is willing to pay a price for.
Practical supplement
System and behaviour alignment check
Check each item against one real event rather than reducing system gaps to personal attitude; for important personnel decisions, preserve an opportunity to explain and substantive human review.
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
What system support currently exists for one behaviour you want your team to demonstrate more often?
Smark Global perspective
Putting AI to work in a company is not only a technology project; it also involves workflows, authority, people and management systems.
Learn more: Smark Global enterprise AI solutions →Source: Original commentary by Harry Lui
Last updated: 2026-10-02(first published: 2026-10-02)
