Consider an analyst preparing an investment committee pack. AI summarises the public filings in minutes. The analyst then searches previous papers, retrieves the committee's earlier concerns and stitches that material together by hand. The firm's knowledge has never been organised and connected to the AI-assisted process.

That is useful progress. The analyst has found a way to save time. But the work of finding and connecting the firm's own context is largely unchanged.

The analyst still owns the investment argument. But they shouldn't have to reconstruct the firm's knowledge each time they prepare one. They may be using AI every day, yet still spend hours finding and connecting information manually.

When I assess where a firm is on its AI enablement journey, I want to understand how those individual gains can help the wider team. Can colleagues use the same method? Can they build on what someone else has learned without starting again?

Some teams need help getting started. Others have capable people whose useful experiments are ready to become part of everyday operations. For investment firms in Hong Kong, that means making expertise available across the team, under conditions that let people use it responsibly.

The Starting Point Matters

An AI setup used by a colleague in London or Singapore cannot simply be assumed to work on the same terms in Hong Kong.

As of October 2026, OpenAI lists the UK, Singapore and New Zealand among its supported locations for ChatGPT. Hong Kong is absent. This is a provider availability restriction, rather than a general prohibition on AI use in Hong Kong. It nevertheless changes the options available to a team. OpenAI supported countries

I would establish the permitted access route before building a business process around it. The particular service matters, including the models available through it and the terms under which it processes company information.

There is also a local regulatory context. The SFC supports responsible AI adoption by licensed corporations and expects controls proportionate to the use case. Its guidance covers testing and ongoing oversight, including responsibility for third-party providers. SFC guidance on generative AI

These considerations belong in the design from the beginning. Otherwise, a promising demonstration can leave the firm with something its staff cannot use for the work that justified it.

Make a Useful Method Available to Colleagues

Consider what would help a second analyst prepare a good IC pack using the first analyst's approach.

Sharing a prompt would be a start. They would also need access to approved research and a clear understanding of what the committee expects. An example of a well-reviewed pack could show how the firm distinguishes evidence from assumptions. Previous committee challenges could help them recognise where a proposal needs closer examination.

An AI-assisted process could bring that material together, with links back to the evidence and gaps flagged for investigation. The analyst would still own the argument. But they would have a stronger starting point, informed by the firm's experience.

The test is whether colleagues can use the method on different proposals and recognise when it needs adapting. A pack produced faster is worth less if reviewers spend the saving untangling its claims.

Allocator due diligence questionnaires, or DDQs, illustrate how a shared capability can keep improving. A colleague may already have an effective way to draft responses. Making that method reusable means connecting it to current, approved fund information and identifying who resolves unanswered questions.

If compliance corrects an answer, that correction should reach the approved source used for the next questionnaire. It should not remain buried in one person's email. Someone must also withdraw answers that are no longer valid.

Last month's answer can be beautifully written and still be last month's answer.

That is how one person's useful experiment can become a dependable starting point for others. Each review can improve the method, provided someone checks and approves what changes.

Give People a Route They Can Use

This is where architecture and governance become practical.

An analyst needs to know which internal material can be used for a particular task. An investor-relations colleague needs an approved way to work with meeting records. The firm needs to know who can change the process and who supports it when it stops working.

A route that requires permission for every routine action can leave capable people stuck. Unrestricted access creates a different problem. I would define the permitted work clearly enough that people can proceed within agreed boundaries.

Zero data retention can form part of that design. But a provider's promise not to retain content does not settle whether the firm may use that information for the proposed purpose. Where personal data is involved, Hong Kong's PDPO also requires attention to purpose and security, including safeguards over processors. PCPD explanation of the PDPO

The same care applies to what AI can do inside connected systems. The SFC's June 2026 guidance specifically addresses limiting connector permissions and applying maker-checker controls to high-impact actions. SFC cybersecurity guidance

For me, the aim is to give people useful freedom within boundaries the firm can explain and enforce. The technical choices should serve that aim.

Help the Team Own and Improve It

The person who developed the approach is a valuable starting point. I would involve them in explaining which sources matter and where they still apply judgement. That gives colleagues a way to understand the method, rather than simply repeat its instructions.

Let another team member try it on a real case within the approved scope. Where do they need help? What looks obvious to its creator but is missing from the guidance? Those questions reveal what needs to be made explicit before the approach can travel further.

A shared method should leave room for professional judgement. An unusual investment proposal may warrant a different line of enquiry. Colleagues need to know when to depart from the usual approach and how to bring a useful improvement back to the team.

Ownership also needs to be practical. Someone must maintain the sources and test changes to the workflow. Support cannot depend on the original enthusiast always being available.

The workflow shouldn't take annual leave with its creator.

Put the Capacity to Work

Our guide to what fund managers should implement first covers choosing a starting workflow. Here, I would judge progress by whether more people can use the resulting capability reliably, with less repeated assembly and avoidable rework.

Then decide what that recovered capacity is for. An investment team might investigate a question it previously had to leave unresolved. Investor relations might spend more time with investors, or tackle work that keeps being deferred. The choice should reflect what the firm needs, rather than assume every saved hour must become more of the same output.

By the next committee meeting, the aim should be a pack that is easier to prepare and gives the committee a better basis for discussion. Retain the lessons from preparing it and from the review, so the next analyst benefits too.

Today's AI can support this. How far a firm takes it depends on its ambitions and the state of its knowledge. At ELab, our AI advisory and implementation work helps organisations assess that starting point and embed changes into everyday operations.

The opportunity is to make the firm's experience useful to more people, then give them room to do something worthwhile with it.