From Scattered Data to AI-driven Action

I’ve been saying the same thing to anyone who’ll listen: the real bottleneck in wealth management isn’t the AI — it’s the infrastructure underneath it. The biggest gap sits at the data and architecture layer, and no model, however good, can deliver reliable insight when it’s sitting on top of fragmented systems and inconsistent data.

AI is now a strategic priority for the large majority of wealth and asset management firms — by most surveys, well over 70% of them. And most are still stuck in experimentation. The reason isn’t that AI is easy and they’re being lazy. AI is hard. Building agents that reason reliably over a client’s full financial life is genuinely difficult work. But it’s gated by something even more fundamental, and that’s where almost the entire industry is stuck: you cannot put intelligence on top of data you can’t even assemble.

What “fragmented data” looks like on the ground

Let me make the abstract concrete, because “fragmented systems” undersells how broken this actually is.

Picture a typical advisor. Positions live in Addepar. Held-away assets and performance sit in a custodian feed or Black Diamond. Private holdings are in Carta. Planning is in eMoney. Market data and research come out of Bloomberg. Banking flows through another system entirely. And the context that matters most — what the client actually said, wants, and fears — is scattered across email, calls, Slack, and a CRM that’s only as current as the last person who updated it.

Here’s the part that rarely gets named: the advisor doesn’t own any of this data. They can log into each system and look at it. But to answer one real question — “what is this client’s true exposure, across everything, right now?” — a human has to manually run a workflow: pull from each source, reconcile it, line it up, and assemble the answer by hand. It’s slow, it’s stale by the time it’s finished, and it’s inconsistent, because the quality of the answer depends entirely on who assembled it.

When aggregation is manual, expertise becomes a workaround for broken plumbing. The “great advisor” is too often just the one best at stitching fragments together in their head. The industry has quietly turned a data problem into a talent problem and called it a moat. It isn’t a moat — it’s a liability that walks out the door at 5pm. And it’s exactly why any intelligence you generate becomes unreliable the moment it touches this environment.

There’s a deeper reason this can’t simply be automated away: the systems themselves were never built for machines. Every one of these platforms was designed for a highly trained human to log in, navigate a screen, and interpret what they see — they assume expert judgment on the other side of the glass. What they don’t offer is the thing an AI agent actually needs: clean programmatic access, structured outputs, and the agentic protocols and interfaces — the APIs and tool layers an agent can call — to read data and take action without a person in the loop. So even where the data exists, it’s locked behind interfaces meant for people, not software. You can’t just point an agent at today’s wealth stack; there’s nothing for it to grab onto.

Share

The bar: how modern companies treat their data

The contrast with a competent technology company is stark, and it’s the standard wealth management should be measured against.

A modern company doesn’t wait for someone to need an answer and then go hunting. It runs pipelines continuously — 24/7 — pulling from every source: databases, payment systems like Stripe, attribution and conversion tools, revenue systems, support logs. Raw, unstructured data lands in a data lake; clean, structured data lands in a warehouse; increasingly it’s one converged lakehouse. Along the way the data is cleaned, labeled, given a consistent shape, and tracked with lineage so any number can be traced to its origin.

The difference isn’t the storage technology — it’s the posture. The advisor model is pull-based: a human triggers a scramble when a question arrives. The modern model is push-based: the data is already aggregated, current, and query-ready before anyone asks. By the time the question shows up, the answer is waiting.

That is what “fixing the foundations” has to mean in practice. Not a vague promise of “modernization,” and not a five-year rip-and-replace. It means one governed place where your data continuously lands, stays current, and carries enough lineage that you can explain any figure to a client or a regulator. A single source of truth that is fed, not fetched.

Why this is harder in wealth — and why it matters more

I won’t pretend this is as simple as copying a SaaS playbook. Legacy infrastructure is a real structural barrier — firms can spend up to 70% of their technology budgets just keeping the old stack alive. But the deeper challenge is ownership: a startup owns its own Stripe data; a wealth manager mostly doesn’t own theirs. Custody, regulation, and contracts mean firms get permissioned, brittle, sometimes screen-scraped feeds, with real reconciliation problems — corporate actions, cost-basis, as-of dating. The upstream data isn’t just fragmented; its rights and its quality are worse to begin with.

That makes the foundational work harder, not optional — and it raises the stakes on the thing that matters most: data consistency is fundamental for both AI effectiveness and client trust. When an intelligent system makes a recommendation, the client and the regulator both need to understand why. That’s impossible on inconsistent, untraceable data. Governance and lineage aren’t a compliance checkbox bolted on at the end; they’re the precondition for being allowed to use AI on client money at all.

And the clock is running. With an estimated $70 trillion set to pass to a younger, digitally native generation over the next decade, the demand for real-time, personalized advice will not be met by hiring more people to manually reconcile spreadsheets. It’s a data-architecture problem before it’s anything else.

Share

This is exactly what we built Intellebox for

This is the gap I started Intellebox to close — and it’s why I think most of the industry is still describing the problem while we’re shipping the solution.

Think of it as an organizational brain. The blocker to AI automation is no longer the models; they got remarkably good, remarkably fast. The blocker is domain knowledge. Every wealth firm runs on critical know-how that’s scattered everywhere — some of it in advisors’ heads, the rest buried in old email accounts, Slack threads, client notes, support tickets, and databases. The firm works because humans vaguely remember where that knowledge lives and how to apply it. AI agents can’t operate like that. They need a new primitive: a system that pulls knowledge out of all those fragmented sources, structures it, keeps it current, and turns it into something an agent can actually execute — safely and consistently. Not a company-wide search box or a chatbot over documents, but a living map of how the firm really works: how a portfolio gets rebalanced, how a pricing or fee exception gets approved, how a client review gets prepared. That’s the missing layer between raw company data and reliable AI automation, and I think every firm is going to need one.

Intellebox built the infrastructure and the capability to support the scale of transformation this industry keeps talking about. We connect to the systems advisors already use — Addepar, custodians, Carta, eMoney, Bloomberg, banking, CRM, and the client communications that usually fall through the cracks — and stream them continuously into one governed source of truth: a lakehouse and secure vault, cleaned and structured, with a semantic layer and full lineage and audit. Push-based, not pull-based. Current by default. Explainable by design. And because this is a highly regulated environment, audit and compliance can’t be a quarterly scramble — every record carries its lineage and access trail, so the data stays continuously audit-ready and compliant as it changes, not just at the moment someone runs a report.

And because the foundation is solid, we can do the part everyone actually wants. On top of that unified data sits an agentic layer that retrieves a client’s full context, reasons over it, decides, and acts — surfacing recommendations, flagging risks, drafting the outreach, executing workflows — and writes the results back into the systems where the work lives. The data feeds the intelligence; the intelligence acts and feeds the foundation back. That loop is the product.

None of this works if you skip the foundation. An agent is only as good as the freshness and consistency of what it stands on; put a brilliant model on fragmented, stale data and you get confident nonsense delivered faster. The industry’s instinct has been to buy AI and bolt it onto a broken stack. We did it in the other order — foundation first, then intelligence — because that’s the only order that actually works.

The firms that win the next decade won’t be the ones with the cleverest model. They’ll be the ones whose data was already in one trustworthy place when the intelligence arrived. That’s the bet behind Intellebox: turn scattered data into a foundation solid enough that AI-driven action stops being a demo and starts being the default.

Meidad Glory is Intellebox’s Head of Engineering and a technology leader holding multiple patents in cloud computing and SaaS. Specializing in applied AI, cloud-native infrastructure, and enterprise SaaS, he is a former CTO and co-founder with prior leadership roles at VMware, Dell/EMC, and Skillz.