At our annual Family Wealth Report Family Office Fintech Forum in New York, our US correspondent Charles Paikert explores the rapid evolution of artificial intelligence and what it means for family offices.

This month at the Family Wealth Report Fintech Forum in New York, Ken Gamskjaer, CEO of wealth management platform Aleta highlighted how artificial intelligence (AI) has gone from answering questions to executing workflows. “Instead of just asking AI what is my workflow? you can now tell it to rebalance, flag tax implications and draft a memo,” Gamskjaer said.

The emerging infrastructure for AI wasn’t even possible six months ago, according to Gamskjaer. Since then, OpenAI, Google, Microsoft, AWS and Anthropic have adopted MCP as the universal protocol for connecting AI to data. Once built, it can be connected to any data model, without vendor lock-in.

Consequently, a host of platforms have opened up in the last six months, as Morningstar, PitchBook, LSEG and Aleta have launched MCP integrations allowing private market intelligence, public markets data and portfolio analytics to be accessible to AI agents.

(For a report on Friday about the FWR investment forum, also held last week, see here.)

Build your own agents
As data becomes increasingly important and data ownership becomes non-negotiable, family offices will need to deal with their existing systems, Gamskjaer said, which are fragmented, closed and not built for machine-readable output. Then family offices will begin to build their own AI agents on top of their data to align with their specific structures and workflows.

While the family office profile of small teams, complexity, recurring workflows and fragmented source data is optimal for applying AI agents, the first winning use case is not autonomous decision-making, but supervised preparation work, according to Alex Lee, CEO of accounting software firm Truewind.

What agents can – and shouldn’t – do
Agents can collect relevant source files and data, draft schedules and summaries, map transactions and categories and flag missing support documents. But humans need to approve accounting treatment and thresholds; resolve edge cases and non-standard judgements; accept, reject or escalate open exceptions and own the final decision and accountability.

AI is general ledger agnostic and can help family offices with repetitive manual tasks, recurring workflows, categorization, reconciliation and consolidation. But successful agents “will not replace judgement,” Lee said. “They will reduce preparation.”

As Murali Nadarajah, global head of R&D and AI for wealth AI platform Eton Solutions put it, artificial intelligence has changed from “deterministic to probabilistic.”

Like a Golden Retriever
Family offices can now safely automate first mile activity, senior strategic marketing and business development executive Mark Wickersham said. For WE Family Offices, AI is a task accelerator, said Rocio Ortega, a firm partner and head of talent.

But conference speakers warned family offices not to use public AI models, and when they purchase proprietary models to ask what data those models were trained on.

Artificial intelligence is also biased to please the user, conference attendees were cautioned. “AI just wants to be like a Golden Retriever and bring you back what will make you happy,” said Greg Kammerer, principal and CFO of multifamily office 61 Holdings. “You have to tell prompts what [the application] can’t do, as well what you’re looking for.”

How family offices should deploy AI
The family office environment of small teams, high complexity, recurring workflows and fragmented source data make it an ideal incubator for artificial intelligence applications, according to software firm Truewind CEO Alex Lee.

How family offices should go about deploying AI, heed red flags and deal with vendors were major themes at Family Wealth Report’s annual Family Office Fintech Forum.

Structuring data is an ideal starting point and foundational, industry executives agreed. Then see where your needs are, which tasks are repeatable and can be automated. “What are the most manual procedures you’re doing?” said veteran family office executive Danielle Klimek.

Next, outline your strategy. “Why will you be using AI? What is the intent?” said Greg Kammerer, principal and CFO at multifamily office 61 Holdings. “Where your end needs?” said Ryan Kerry, CEO of Knowledger, a digital accounting firm for family offices. “Make your decision with the end in mind.”

After setting parameters for a strategy, family offices should set a policy for security and privacy, said Michael Maffattone, chief information officer and founder of fintech firm MultiMix IT. The policy should include not allow employees to use their own AI devices or public models that the company hasn’t licensed.

Vet every option
When it comes time to pull the trigger and decide what models to work with, don’t rush. “Vet every option, demand proof of concept,” said Janet Welch, managing director of operations for family office Trove. The final selection doesn’t have to be perfect, but make sure it’s at least an MVP (Minimum Viable Product).

Don’t be intimidated by vendors. “A common mistake is to not ask what the system can’t do that you might want it do in the future,” Welch said. “Don’t automate the past,” said Murali Nadarajah, global head of R&D and AI for family office software firm Eton Solutions. “Look to the future.”

How many of your clients look like me?
Decide if you want an all-in-one software solution or go with individual best-in-breed products. “Ask your software [vendors] if they will work with you as your firm evolves,” Kerry told conference attendees. “Ask what the product can’t do and what their AI roadmap looks like.” Also, inquire “How many of your clients look like me?” added Josh Kanter, CEO of family office fintech firm Leafplanner.

With the application in hand, train staff and encourage them to come up with their own  ideas for AI applications, Maffattone said. “You need the team to be fully on board from the beginning,” Welch emphasized. Once the new system is in place, “test for risk levels,” said Mark Aklian, chief information security officer for artificial intelligence platform Chiri.ai.

After careful preparation, WE Family Offices recently began deploying AI for its reconciliation team and advisors, who are using Microsoft Copilot to summarize reports from their in box, Rocio Ortega, a partner and head of talent told conference attendees. “Look at your tech stack, and determine what you need,” Ortega said. As Gamskjaer put it, “Be AI ready, not AI dependent.”

Data structure key to AI adoption
Family offices need to get their data in order before beginning to deploy artificial intelligence, according to industry executives at the forum.

Family offices are beginning to rapidly deploy AI, but AI is a tool that “is only as good as the data you have,” said veteran strategic marketing and business development executive Mark Wickersham. “Dirty data means dirty results.”

A family office’s data can be compared to books in a library which need to be organized to be useful and accessible, said Dario Sarmiento, chief information officer at Wilshire Law Firm. Preparing data is more important than adoption rate, according to Sarmiento. When deploying AI, structured data is king, he told conference attendees.

Fix your data before you buy AI
Solving the data problem is key to AI deployment, Gamskjaer said in his keynote presentation. Family offices increasingly want to own their data, instead of ceding control to vendors, as domain knowledge and the resulting data layer of artificial intelligence becomes more important, Gamskjaer continued. “Fix your data before you buy AI,” he told advisors.

Family offices also need to be really cognizant about data you’re allowing AI to see, Mark Aklian, chief information security officer for artificial intelligence platform Chiri.ai, said. “Determine what your crown jewels are and what you want to keep private,” agreed Paul McKibbin, executive director for consulting firm EY. “And make sure your data can’t be used to train other models.”

To avoid AI bias and mistakes, data must also be verified before it’s inputted, panelists said. The intersection of data and AI has also given birth to tokenized assets, resulting in a “wild new world of digital assets,” Alfonse Mandese, global head of revenue at digital asset trading platform Talos, said. This digital ecosystem will be maturing in the next five years, but legal clarity needs to be established around tokens as an infrastructure develops, according to Toby Thwaites, director of digital assets at BNY. “There’s still a lot of work to do,” Thwaites said.

The bottleneck is not AI
Data scattered across custodians, advisors and spreadsheets, along with different PE manager reports and data that is half manual and half automated, means that the bottleneck is not AI, according to Gamskjaer. “It is everything beneath it.”

Even if data is consolidated, it may be locked in a closed system controlled by a vendor, leaving family offices unable to connect its AI agents to its own data. As a result, you get the vendors’ features, not your strategy, Gamskjaer said.

Family offices should build one reconciled view across all asset classes and chose open architecture over closed platform, he advised. AI deployment, Gamskjaer said, should be a leadership decision, not an IT project.

To reach its potential, AI needs to expand geographically and embed in “all employee layers” of companies, according to BNY's Kevin Shea. AI investors should look for “where the bottlenecks are,” such as data centers, memory manufacturers, chip makers and networking optics. Investors should also make sure they have private exposure to AI-related companies, Shea said.

From fragmented data to an integrated platform: AI tech demos
In addition to panel discussions and fireside chats, Family Wealth Report’s Fintech Forum featured on-site tech demos focusing on how a range of AI applications and platforms for family offices can transform workflow and data management.

Centralizing family office management
Fragmented financial data, documents and stakeholder management for family office advisors can be centralized on an integrated AI platform.

Simran Kang, founder and CEO of MyFoTech demonstrated how capabilities such as real-time analytics and workflow management can be customized to reduce complexity and drive informed decision- making at scale.

Faster reporting workflows
Embedded AI can help family offices transform complexity into control.

Eric Switzer, director of sales for Masttro, demonstrated how the company’s platform can turn fragmented documents, alternative investment data and transaction activity into fast answers, cleaner workflows and reporting-ready intelligence.

Automating data entry
Manual data entry is nobody’s idea of fun. Krithika Rao, solution engineer for K1x, demonstrated how the firm’s AI aggregator maps data using a collaborative user interface and transfers the data into Excel-based reporting that can used for both tax compliance and client advisory opportunities.

Transforming records into action
How does AI navigate workflows such as reading documents, triggering processes and executing tasks? Stuart Guidry, vice president of platform consulting for Eton Solutions, demonstrated how a family office platform operates in real time. The EtonAI Web Agent showed how AI can turn a system of record into a system of action.

Scaling alternative portfolios
Operational bottlenecks become inevitable as alternative portfolios become more complex. Kelly Moore, executive director and head of strategic accounts for Arch, demonstrated how the firm’s AI application software automates reporting, reduces error and centralizes data across entities, enabling family offices to scale without adding headcount.

Simplifying multi-entity AP management
How can a family office’s approvals be simplified, controls strengthened and client relationships deepened? Phil Klopke, account executive for BILL, demonstrated how the firm’s AI platform can streamline the AP process, from bill creation to approvals and payments, and sync with accounting software.