Every AML regime on earth rests on one deceptively simple demand: find the human.

Behind every entity, every trust, every nominee arrangement, there is a natural person who owns or controls the thing, and your job is to identify them, verify them, and decide whether you want their money. Everything else in the rulebook is supporting detail.

So it should worry the industry that this is precisely the task most programs do worst. When regulators publish enforcement actions, you tend to see things like ownership chains nobody followed to the end, a controller nobody named, or a trust nobody looked inside. The find-the-human part was missing.

And nowhere does it give up more often than in the investment world, where layered structures are most common. 

Why fund structures defeat standard UBO analysis

A retail bank's UBO problem is a local company with three shareholders. The investment world's UBO problem is a Delaware feeder into a Cayman master, owned by a Luxembourg holdco, held by a Jersey trust, administered by a Swiss trustee, with a protector in Singapore. Every layer is legitimate, common, and tax-driven. 

Three things make this hard in ways checklists don't capture.

1. Ownership must be computed. Effective ownership through a chain is multiplication: 60% of 50% of 80% of the top layer. Registry filings give you fragments in different formats and languages; someone has to assemble the arithmetic across the whole chain and notice when two paths converge on the same person, quietly pushing them over the threshold.

1. Control doesn't follow the equity. The 25% ownership test is the easy half. The person steering the structure may hold 5% and a shareholders' agreement, or 0% and a general partner role, or nothing but the power to replace the trustee. Trusts explode the category entirely - settlor, trustees, protector, beneficiaries all come into scope. A UBO analysis that only multiplies shareholdings finds the owners the structure wants you to find.

1. The answer decays. Structures are reorganised for tax, succession, or convenience - routinely, and without anyone notifying the fund's compliance team. A UBO determination is a photograph of a moving object. Most firms frame the photograph and file it for three years.

The "senior managing official" escape hatch

When a chain gets too hard, most regimes let you record a senior managing official instead of a beneficial owner. It exists for the genuinely ownerless edge cases.

In practice, this is where difficult analyses go to die. An analyst hits a nominee wall or a stubborn trustee, the deadline looms, and the file gets an SMO entry and a green tick. Everyone involved knows what happened: the structure won.

Here's the uncomfortable heuristic - the files most likely to end in an SMO shortcut are the complex, opaque, cross-border ones. Which is to say: your SMO-flagged files are a machine-generated list of exactly the places a bad actor would hide. Some firms have never read that list as what it is.

What rigorous UBO work actually requires

1. Draw the whole structure before judging any of it

The unit of analysis is the structure (go beyond the entity itself). Resolve every layer to a chart first - jurisdictions, percentages, control rights - and only then decide where verification effort goes.

2. Run ownership and control as separate analyses

Two questions, always: who owns more than the threshold, and who actually steers? The second requires reading constitutional documents, LPAs, shareholder agreements and trust deeds - the ninety-page PDFs that humans skim and shouldn't.

3. Verify against sources the customer doesn't control

A structure chart supplied by the investor's counsel is a claim, but without verification it lacks evidential substance. Every layer needs corroboration from registries, filings, or documents with independent provenance. Where registries are opaque, that's a risk factor to record.

4. Treat every UBO determination as perishable

Refresh on triggers - registry changes, filings, adverse media, transaction anomalies - go beyond the calendar. A three-year review cycle is a formal commitment to being three years out of date.

5. Audit your SMO population

Pull every file that ended in a senior-managing-official determination and ask honestly: was this analysis exhausted, or abandoned? The answer is a direct measurement of your program's real, as opposed to stated, risk appetite.

The economics that make this fixable now

Here's why this failure has persisted: rigorous UBO work is brutally expensive in analyst hours. Reading trust deeds, reconciling registries in five languages, recomputing chains after every change - no compliance budget on earth staffs that for every file. So programs rationed rigour and called the rationing "risk-based."

Where Steward fits

This is the core of what Steward does: our AI reads every document, builds the chart, runs the arithmetic, checks the registries, and keeps watching after onboarding - so "find the human" stops being an aspiration and becomes the default output of the system

That constraint just dissolved. Reading long documents, extracting ownership data, computing chains, monitoring registries for change - this is exactly the work modern AI does relentlessly and cheaply. The excavation scales; the human judgement stays where it belongs, at the bottom of the hole, deciding whether what was found is acceptable.

See it for yourself and book a demo here.