A year and some on from the official launch of SOFI, the AI tool developed by mallowstreet for the financial services industry, CEO Stuart Breyer engaged in a media briefing for an update on its usage and an overview of the state of AI implementation generally.
As previously noted on International Adviser, the tool is different to others like ChatGPT in that it developed out of needs around meeting analysis and has taken a very different approach to how it is trained - it does not rely on client data for this. Such distinctions are likely to rise up the regulatory agenda going forward, noted Breyer, alongside other 'guardrails' against issues such as hallucinations - when AI makes up information because it does not know the answer.
Also shifting is the direction of AI technology, towards so-called Model Context Protocol and Agentic Model Design, which are expected to improve the accuracy and speed of AI in use, as well as less use of energy.
For any business thinking of implementing more AI, Breyer stressed that it is better to integrate into existing workflows, rather than force users to adopt new processes.
Leveraging off the discussion, the SOFI tool provided the following summary of questions and answers:
How has SOFI evolved since last year?
"SB explained that SOFI has grown significantly, now analysing over 220 meetings per day compared to 20 meetings a year ago. Last month alone, SOFI analysed 3,180 client meetings, saving clients approximately 397 working days. SB also discussed improvements in data security, the development of templates for systematic data extraction, and the implementation of safety checklists for compliance purposes."
How does SOFI handle client data security and privacy?
"SB explained that SOFI does not use client data for training the AI model. Instead, it learns from Mallowstreet's core business and events portfolio. SB emphasized that this is an important distinction from other AI tools that might use client data for training purposes. SOFI is designed to analyse meetings and provide outputs without retaining knowledge from those specific client interactions."
How does SOFI work across different regulatory jurisdictions?
"SB explained that SOFI is designed to be multi-jurisdictional, multi-model specific, and multi-hosting specific. They configure SOFI on a client-by-client basis to meet specific regulatory requirements. For example, they have German clients who require their data to stay in Germany, UK clients who want data to remain in the UK but allow processing in the EU, and US clients who want regional data separation. SB emphasized that clients own their data and can delete, extract, export, or integrate it into their systems as needed."
What is the average meeting length for wealth advisory clients?
"SB shared that across their client base, some clients average 36 minutes per meeting, while others average around 42 minutes. The aggregate wealth meeting is about 41 minutes, though there is significant variation ranging from brief 4-minute conversations to comprehensive estate planning meetings lasting 2.5 hours."
What will happen to advice firms that don't adopt AI?
"SB predicted that the competitive gap would become immense between firms that adopt AI and those that don't. This gap will manifest in their ability to effectively service and win business compared to the economic cost of running their businesses. SB also mentioned a potential regulatory tailwind, suggesting that regulators might eventually use AI for compliance monitoring rather than sampling meetings."
Is SOFI only for the wealth industry or can it be used for other financial topics?
"SB clarified that SOFI is built on financial services data broadly, not just wealth management. It can handle any conversations within financial services effectively because it's been trained specifically on financial services data. This specialization makes SOFI more accurate for financial services applications compared to general-purpose AI tools like Zoom and Copilot that need to cover many different industries."
How effective is AI at identifying vulnerable clients?
"SB explained that SOFI currently analyses vulnerability based on transcript text rather than tone or body language. They've chosen to focus on words rather than sentiment analysis because it provides more concrete evidence. They're building a specific agent to identify vulnerability based on regulatory definitions, firm policies, and advisor experience. SB emphasized the importance of supervised learning in this process, where they test the model against known cases of vulnerability and refine it based on results."
When will financial advisers see the true benefits of AI?
"SB admitted they don't have a definitive timeline but noted differences in adoption patterns. Large institutional clients are slower moving initially but have sophisticated long-term visions and resources to implement proper change programs. The advice industry is more fragmented, with some forward-thinking adopters, some who are uncertain how to proceed, and others resistant to change. SB suggested that defining clear use cases and measuring tangible savings are key factors in driving adoption."
What is Mallowstreet and how was SOFI developed?
"SB explained that Mallowstreet has been operating for 16 years in the institutional retirement savings market. Their core business includes malistreet.com, which serves about 4,500 institutional investors managing approximately £2trn in assets. They work with about 100 asset managers through research offerings and events. SOFI emerged from their events portfolio and is now a separate business with different objectives. The company has 25 employees and is based in the UK, not the US."
Does SOFI work in languages other than English?
"SB confirmed that SOFI currently works only in English. While simultaneous translation technology is improving, they are concerned about accuracy when translating back into native languages. They haven't focused on developing multi-language capabilities yet."
How well does SOFI handle different accents?
"SB stated that SOFI handles accents very well because it was trained on a wide variety of client data from across the UK and internationally. They mentioned that in 2018-2019, they had a China business where managers with heavy accents successfully used SOFI for creating marketing materials."
Should advice firms hire AI professionals to help advisors use these tools?
"SB suggested that most wealth advisory firms can't afford to hire AI consultants and questioned the value of such expertise. Instead, they emphasized that tools like SOFI are designed to provide immediate time savings (4-6 hours per week for advisors and power planners) without requiring specialized knowledge. SB advocated for finding efficiency first, proving the value, and then redeploying the saved time elsewhere as the path to organic adoption at scale."
Has SOFI expanded into the American market?
"SB explained that while clients with US colleagues and teams are using SOFI, they haven't actively approached US clients for two reasons: 1) Many well-funded tech firms have failed trying to crack the US market, and 2) They see sufficient market opportunity in the UK and want to focus on their existing network before expanding further."
