Sofistic AI Lead Fayrouz Ben Soui on what her team builds for advisors, why AI models sound sure even when they're wrong, and where wealth management AI is heading
Wealth advisors have more client data than ever, but it's scattered across documents, accounts and systems. Sofistic brings it together into a single view of each client's full financial picture, giving advisors and their clients real clarity, and the insight to make more strategic decisions that grow client wealth and advisor AUM.
We've designed Sofistic to be as simple to use as possible. But simplicity takes a lot of work. Behind every feature is a team of engineers making sure each AI tool is accurate, secure and calibrated to what today's wealth advisors actually need.
Fayrouz Ben Soui is Sofistic's AI Lead. Just back from ALL IN in Montréal, where Sofistic was named to the event's Top 100, she told us how she got into AI, how her team builds AI for wealth advisors, and where AI in wealth management is heading.
How did you get into AI?
I studied physics engineering, so my path started with the math behind AI rather than computer science. My first AI project predicted energy use in buildings. Then I worked with medical data, and later at a consulting firm doing AI projects in legal, supply chain and education.
The industries were different, but many were highly regulated, and that shapes how you design AI tools. The first steps were always the same, too: gather the data, structure it, make it usable.
How would you explain what your team builds to someone new to AI?
Think about how a pilot flies. They don't check gauges scattered around the plane. Everything they need is on one dashboard, and the plane alerts them when something needs attention. That frees them to focus on flying.
We build that for wealth advisors. Everything about a client's financial life, including held-away and off-book assets, comes together in one view. Instead of piecing the picture together, the advisor can focus on the best path forward, and clients can see the same picture alongside them in meetings.
AI powers the alerts. Our Insights feature, part of SofiAI, looks across a client's full situation and flags what isn't obvious, like a life event that could affect their existing wealth plan. It doesn't replace the advisor's judgment. It surfaces what they might not think to look for, and they decide what to do with it.
What has working directly with wealth advisors taught you?
We work directly with wealth advisors as we develop the platform, including advisor teams at National Bank Financial, and that back-and-forth is my favourite part of the job. It shows us what really matters to them.
It's also full of surprises. In workshops, advisors tell us that features we found relatively easy to build have a huge impact on their day. Meanwhile, features they assume are easy to make are often the most complex for us.
What looks simple in AI for finance but is actually complex?
Financial document extraction. To an advisor, it's simple: upload a statement and the fields fill in. The hard part is being certain every number is right, because one zero more or less makes a very big difference.
Behind that one click are validation checks and weekly monitoring. Even the "simple" features need a rigorous process.
What do people get wrong about AI in finance?
A lot of people assume AI companies just build features that call ChatGPT or Claude. That's not how it works. Our use of these models is a tiny part of the work.
The real work is building the systems around them. Every feature is designed to our own criteria, then validated and tested so every answer is accurate and useful to advisors. That matters because AI models are confident even when they're wrong. When data is missing, a model fills the gap and sounds just as sure. In a regulated industry, you can't trust any output blindly, especially a number.
How does Sofistic keep wealth management AI safe and secure?
Safety is built into every feature from the start, not added at the end. Everything the platform surfaces passes through multiple validation checks, and we double-check that outputs are correct and within our industry's rules and regulations.
We also monitor our features every week to catch any drift in accuracy. When we test new AI techniques, we do it in a secure sandbox, never on live client data. And we keep working on the questions the whole industry is asking, like governance and data residency. You can review our security practices in our Trust Center.
What makes Sofistic's approach different?
We don't just plug in a powerful model and hope for the best. We choose specific models for specific tasks, study the latest research on the problems we're solving, and combine different scientific approaches to build the solution that best fits advisors.
And we build with advisors, not just for them. Their feedback decides what we prioritize. The result is a platform that's simple to use, with a lot of rigour underneath.
You recently represented Sofistic at ALL IN. What stood out for you this time around?
How much has changed. I attended the very first ALL IN, so I've watched the event grow alongside the technology, and every year the jump in what AI can do is striking.
Having Sofistic named to the Top 100 was exciting. But what stood out most was the people. I wasn't there to size up competitors. I wanted to see what everyone else is building. Some of the best conversations were with people from completely different industries who use techniques similar to ours. They understood our challenges right away, including the work most people assume is trivial.
How did people respond to Sofistic at ALL IN?
The interest came from everywhere. People from outside finance came up and asked whether they could use Sofistic for their own money. That says a lot about how much people want clarity about their finances.
We also met financial organizations and potential partners who were very interested in what we're building. It was encouraging to see that the problem we're solving resonates with so many people.
Where is AI in wealth management going?
The possibilities are wide open, and models are improving fast. The big question is how to use them safely, with the right governance and guardrails. We build that into every feature, and it will only matter more as the industry adopts AI.