At a closed-door Hubbis forum in Dubai, Murali Nadarajah examined how modern family offices are translating operational complexity into usable intelligence. The Chief Information Officer of Eton Solutions presented a practical walkthrough of how integrated platforms and Agentic artificial intelligence (AI) can process large volumes of fragmented information across funds, trusts and multi- jurisdiction structures, enabling firms to move from maintaining records to actively managing them.
His presentation, Platform Deep Dive & Agentic AI in Action, focused on the daily reality inside wealth organisations: documents arriving in different formats, repeated reconciliations and delayed reporting cycles. The challenge, he argued, is not a lack of data but the inability to use it quickly. The opportunity lies in converting raw information into structured, auditable intelligence that supports decisions rather than simply recording them after the fact.
Key Takeaways
- Most operational data is unstructured: around eighty percent of family office information sits in documents, emails and statements rather than databases.
- Agentic AI converts documents into usable intelligence: systems can classify, interpret and post data automatically while maintaining audit trails.
- Integrated workflows reduce operational drag: accounting, payments, reconciliations and reporting can be automated across entities and jurisdictions.
- Oversight improves when data becomes queryable: decision-makers move from waiting for reports to interacting directly with real-time information.
- The goal is decision support, not replacement of judgement: humans remain accountable, but manual repetition declines sharply.
The quiet operational crisis
Nadarajah began by describing what he called a “quiet crisis” inside wealth organisations. Unlike market volatility, operational inefficiency rarely appears dramatic. Instead, it manifests as everyday friction: delayed feeds, manual reconciliations, document chasing and ad hoc adjustments.
Over time, these routine tasks become accepted as business-as-usual. Teams spend significant effort maintaining continuity rather than improving oversight. Nadarajah argued that most organisations underestimate how much of their capacity is consumed by maintaining processes that were never designed for the complexity they now handle.
He identified four recurring patterns.
The first is operational drag. Staff manually monitor incoming statements, extract information from PDFs, compare capital calls to agreements and reconcile ledger mismatches. Each step is manageable individually but collectively absorbs disproportionate time.
The second is siloed systems. Accounting, investment reporting, tax and document management operate separately, forcing employees to act as connectors between systems.
The third is human patchwork. When systems disagree, staff intervene with manual corrections to keep workflows moving. This maintains continuity but reduces reliability.
The fourth is limited transparency. Only a small portion of data exists in structured form, meaning systems can process only a fraction of available information.
Taken together, Nadarajah said, these patterns explain why operational teams spend most of their time maintaining records instead of analysing them.
Why unstructured data matters
A central argument of the presentation was that family offices do not lack information. They lack structured information.
Nadarajah estimated that roughly twenty percent of operational data sits in structured systems such as accounting records or custodial feeds. The remaining eighty percent exists in documents: capital call notices, subscription agreements, valuation reports, invoices and statements. Traditional software cannot easily interpret these documents, so they are handled manually.
Agentic AI addresses this imbalance by reading documents, identifying their type and extracting relevant information in context. Rather than simply capturing text, the system interprets meaning: whether a file represents a capital call, a valuation change or a payment instruction.
The extracted data is then stored alongside structured records, creating a unified dataset that can be queried and analysed.
The distinction is important. Conventional automation digitises workflows that are already structured. Agentic AI expands automation into areas that previously required human judgement, such as recognising financial instructions embedded in narrative documents.
Automating workflows, not just tasks
Nadarajah emphasised that automation becomes transformative only when applied to complete workflows rather than isolated steps.
He illustrated this through payment processing. Traditionally, staff review invoices, verify authorisation, enter details into a system and initiate transfers. Each step may be supported by software but still depends on manual interpretation.
With integrated AI-enabled workflows, instructions can be extracted directly from documents, validated against permissions and executed automatically. The system records each step and produces an audit trail showing how the decision was reached.
A similar approach applies to reconciliations. Bank statements can be compared automatically to ledger entries, discrepancies flagged and adjustments posted. Manager reports can update valuations and generate entries without manual rekeying.
The effect is cumulative. Removing small manual steps across multiple processes reduces both effort and error risk while increasing traceability.
From periodic reporting to real-time interaction
Another focus of the session was reporting. Nadarajah argued that many family offices still operate in a periodic reporting model: data is collected, reconciled and presented after the fact. Decision-makers then react to what has already happened.
When structured and unstructured data are unified, reporting shifts from a scheduled exercise to an interactive process. Users can query holdings, exposures or liquidity positions directly, rather than waiting for prepared reports.
This shift changes the role of reporting itself. Instead of summarising history, reporting becomes decision support. For example, organisations can assess liquidity impact before meeting capital commitments or compare asset allocation dynamically against benchmarks.
Nadarajah noted that this capability does not replace human judgement. Instead, it enables questions to be answered quickly enough to influence action rather than merely record outcomes.
Supporting multi-entity and cross-border structures
Modern family offices rarely operate within a single entity or jurisdiction. They manage trusts, partnerships, special purpose vehicles and operating companies across multiple regions and currencies.
Nadarajah explained that integrated platforms support this complexity by maintaining relational records linking assets to entities and ownership stakes. This allows different stakeholders to see only their relevant exposures while preserving consolidated oversight.
The practical impact is improved governance. Reporting can be generated simultaneously for internal management, beneficiaries and co-investors without rebuilding data manually. Permissions and access controls are embedded rather than improvised.
He argued that this becomes particularly important in the Middle East, where families often maintain international asset bases while operating from regional hubs. A unified platform allows geographically dispersed holdings to be monitored consistently.
Oversight through transparency and auditability
Throughout the presentation, Nadarajah returned to the idea that automation improves oversight primarily through traceability.
Every automated action generates an audit record showing the data source, interpretation and resulting transaction. Humans still approve decisions where required, but they review structured reasoning rather than raw documents.
This reduces reliance on individual memory and increases organisational resilience. If staff change or workloads spike, processes remain consistent because they are embedded in workflows rather than personal habits.
The emphasis on auditability also addresses a common concern about AI: reliability. By making reasoning transparent, the system enables verification rather than blind trust.
Human judgement remains central
Despite demonstrating extensive automation, Nadarajah repeatedly emphasised that the goal is not replacing human decision-making. AI handles repetitive interpretation and processing, while accountability remains with people.
In practice, this means approvals, investment decisions and governance actions remain deliberate choices. The difference is that staff review structured information rather than assemble it.
This distinction reframes automation from a staffing discussion to a capability discussion. The value lies in freeing time for higher-level analysis, risk assessment and planning.
From reactive operations to proactive management
Nadarajah concluded by describing the broader shift enabled by integrated AI systems. When operational effort declines and data becomes immediately accessible, organisations move from reactive reporting to proactive management.
Instead of discovering liquidity shortfalls after commitments, they anticipate them. Instead of reconciling discrepancies at month-end, they resolve them continuously. Instead of compiling periodic summaries, they interact with live information.
The transformation is incremental rather than dramatic. Each automated workflow reduces friction slightly. But across hundreds of workflows, the cumulative effect reshapes how decisions are made.
Execution as the missing link
The presentation complemented the earlier strategic discussion by demonstrating how institutional- grade infrastructure operates at ground level. Where the operating model defines the destination, execution determines whether it works in daily practice.
For Nadarajah, the objective is straightforward: convert fragmented data into usable intelligence and embed it into workflows that support oversight across entities and jurisdictions. The result is not simply efficiency but confidence, the ability to act based on timely, reliable information.