Most wealth management firms have experimented with artificial intelligence (AI) in their advisory workflows, but the gap between using an AI and embedding it meaningfully is still substantial.
At The Wealth Mosaic’s AI in Wealth 2026: APAC Toolkit Roadshow event, held in Singapore on 24 September, Allen Chote, Founder & CEO of Advokit, addressed this challenge directly – exploring where he sees the gap between the two, and what closing it would require.
Chote opened with a working assumption about the division of labour between technology and adviser. “Although still somewhat debatable, most would agree AI is better now at analysing client attributes, reviewing portfolios, and making decisions on investment proposals and portfolio allocation,” he said. “It’s unlikely to ever be better at relationships, building trust, or going out for lunch.”
The implication for firms is that they need control over the client’s interface with AI. “Don’t allow the customer to be the one that engages with AI to try and make their own decisions, risking poor outcomes,” he said. “Control that. Become the trusted expert who is using AI effectively: put AI in your back pocket.”
He argued that AI is often discussed as a single thing, but that its adoption story should be seen in terms of three broad tiers of AI.
The first “frontier AI” tier is the most basic: using a frontier model directly, potentially by copying and pasting client information, notes, or screenshots into a chatbot, “hoping for consistent outputs”. Chote said this approach can produce “mixed outcomes” because it is not connected to the systems underpinning the adviser’s work.
The second “applied AI” tier involves AI being embedded into individual applications. In this model, AI can read data within a particular system and potentially orchestrate tasks, but remains confined to one or two applications.
“AI can read data in the system. Sometimes AI can even orchestrate the workload and take actions that a user could generally take,” he said. “But there’s no connection between those systems, so it’s not really orchestrating end-to-end and, in some cases, can add more time and effort to get things done.”
He added: “We’ve all been in the situation where we fought with AI across systems to automate things and ended up taking more time than if we’d done it manually.”
The third tier, “industry-specific applied AI”, is where AI becomes capable of orchestrating activity across systems and processes. For wealth management, that means understanding the industry-specific workflows involved in activities such as onboarding, KYC, investment proposals, and portfolio reviews, rather than simply providing a generic AI interface.
Despite the advances in AI technology and its widening range of applications, many firms haven’t got past the first tier – and that’s not the AI that can really transform wealth management or meet its specific demands.
“There’s a lot of different things that we have to do every day that are not supported by generalist multi-industry solutions,” Chote said. “It needs to speak to them openly, safely, and it needs to orchestrate each step along the chain. Agentic AI is only as good as the underlying applications it works with; AI-powered investment proposals are only as good as the AI-powered KYC and client attribute capture upstream.”
He also cautioned buyers about the increasingly common claims vendors make to be “AI-native”. A company that has built its technology around legacy architecture and subsequently incorporated AI may be AI-capable, but that is different from being built around AI from the outset.
He quoted Advokit’s Head of Product and Data Mike Hamilton: “Companies that existed before 2022 can’t have an AI-native platform in the same way that I can’t be a Chinese-native Mandarin speaker.”
From AI experimentation to embedded intelligence
The underlying technology is only part of the equation, Chote said. The challenge is creating an environment in which AI can access the relevant information and systems while operating within appropriate controls.
He described the four characteristics of embedded AI in the Advokit SAFE AI approach to agentic AI in wealth management: Security, Accountability, Focus, and Excellence. He argued that, in a regulated environment, the ability to explain how an AI-supported decision was reached is particularly important, but simply asking an underlying AI model to explain itself may not provide a sufficiently reliable record of what happened.
He pointed to the AI guidelines of the Monetary Authority of Singapore (MAS), which emphasise the explainability of decisions made with the help of AI. “That’s very difficult,” he said. “Even if you ask AI to tell you why it made a decision, it might not accurately tell you what happened, particularly if context is lost or a free model is used. So there needs to be a layer that can control that accountability.”
Discussing Advokit’s own approach, Chote described its agent as an abstraction layer sitting above underlying AI models. “The hard work is diligently building a control system on top of other models, rather than short-cutting and outsourcing to frontier LLMs entirely,” he said.
Chote argued that these frontier LLMs could increasingly recede into the background as commodities, while agents become embedded within the applications wealth professionals already use. Rather than an adviser repeatedly leaving their workflow to interact with a chatbot, individual systems could have their own agents which interact with underlying models and, ultimately, with one another.
“LLMs may fade to the back, and we'll never need to talk to Claude in a chatbot ever again,” he said. “Because every single solution you use will have an agent; that agent will connect to the underlying LLMs to harness the power of AI; then, hopefully, those things will become connected over time.”
This could also alter the role of standardisation in technology integration. Chote suggested that natural language gives AI agents an ability to work across systems that use different terminology. “Natural language becomes the integration,” he said.
The ambition, as Chote described it, is to move beyond standalone AI tools towards integrated workflow applications that agents can orchestrate within wealth advice businesses.
The importance of human data
A further theme of the discussion was the value of information that sits outside conventional structured financial data.
Portfolio holdings, account information, and other financial information are already captured by common wealth management applications. But the adviser-client relationship also generates a substantial body of contextual information: conversations, preferences, family circumstances, and other details that can contribute to a fuller understanding of the client.
Chote argued that AI creates an opportunity to capture and use more of this information, provided that it is handled appropriately.
“If we have said that AI will be better at analysing client data and portfolio data, then we also need to give it that data,” he said. “Human metadata can now be stored in ways that are totally unreadable by humans. This sounds scary, and perhaps it is, which is why the Advokit SAFE AI approach is so important.”
Chote described a deliberately controlled approach to data access. Rather than giving the agent broad access to a firm’s systems and then restricting it where necessary, Advokit designed its agent to start with no tools, adding them incrementally as the development roadmap progressed.
The principle, he said, is that the agent should only be given access to data or systems where there is a business need. This creates a more controlled framework for allowing AI to work across client information and workflows without treating unrestricted access as the default.
The proposition is that AI can help to organise a form of institutional memory, helping advisers retain relevant knowledge accumulated throughout a long client relationship. Even a good adviser with a significant number of clients won’t be able to remember everything a client has told them over many years.
That potential also creates a significant responsibility around data protection. Chote described an architecture in which personal client data can be extracted, tokenised, and anonymised before being made available to underlying models, with controls around data retention.
The principle, he argued, is that the use of richer client information should be accompanied by greater transparency and stronger safeguards rather than simply exposing sensitive information to a public AI service.
For wealth firms, the workshop’s central message was that getting there will require more than adding an AI capability to existing technology. It will require appropriate data, industry-specific AI-native partners, and a clear understanding of where AI can add value.
As Chote put it, “ultimately it's about partnerships”. The transition from experimentation to embedded AI may therefore depend as much on how wealth firms assemble and connect their technology environments as on the capabilities of the AI models themselves.
About The Wealth Mosaic
The Wealth Mosaic is a UK-headquartered online solution provider directory and knowledge resource, focused specifically on the wealth management industry.
For wealth managers, the buy side of our marketplace, The Wealth Mosaic is designed to enable discovery of key solutions, solution providers and knowledge resources by specific business needs.
For solution providers and vendors, the sell side of our marketplace, The Wealth Mosaic exists to support the positioning, exposure and business development needs of these firms in a more complex and demanding market.
