How this dataset was built

This dataset was produced by an autonomous research pipeline operating under a written instrument. The instrument was fixed before any data was collected, which is what makes the resulting fields comparable across a hundred companies rather than a hundred separate judgment calls.

The sampling frame

The unit space is the 100 companies on the FinTech Global WealthTech100 list for 2026.

The list page carries no company data — no descriptions, no categories, no links in the markup. The roster is only in a gated PDF and in the filenames of the logo images:

wp-content/uploads/2026/04/1-WT10026-2ndEngine.png
...
wp-content/uploads/2026/04/100-WT10026-Zoe-Financial.png

All 100 positions were recovered from those filenames, with no missing index. Filename slugs that mangle real names were hand-corrected before becoming unit IDs — SSC → SS&C, ipipeline → iPipeline, Nice-Actimize → NICE Actimize, and roughly twenty others.

The leading number is alphabetical, not a ranking. The list runs 2ndEngine → Zoe Financial. It is recorded as list_position and must not be read as a competitive rank. Nothing on this site treats position 1 as better than position 100.

Because the list page itself is only a sampling frame, every substantive field comes from the company's own site or third-party sources — not from FinTech Global.

Every field carries its evidence

Each coded field is stored as a structure, not a bare value:

{
  "value": "aum_bps",
  "confidence": "high",
  "evidence": [
    {
      "quote": "verbatim text from the cited page",
      "url": "https://example.com/pricing",
      "accessed_at": "2026-07-24T00:00:00Z",
      "source_role": "company_official"
    }
  ],
  "gap": null
}

The quote must appear verbatim at the cited URL. This is the contract that makes the dataset checkable: any value on this site can be traced to a specific sentence on a specific page. The Evidence tab in the explorer exposes all of them.

Source roles, strongest to weakest

official_gov › international_org › company_official › professional_firm › specialized_press › general_press › wiki_or_blog

Different fields demand different roles. Products, value proposition, pricing, and IP claims prefer company_official — the company's own site or docs. Funding, headcount, and ownership accept specialized_press such as trade coverage.

Marketing copy is evidence of a claim, not evidence of a fact. It is coded as what the company asserts about itself, which is a different thing from something verified.

Gaps are a deliverable

When a value cannot be established, the dataset records a gap with a reason rather than filling the field with a plausible guess.

Reason	Meaning
`not_found_after_search`	Searched the company site plus at least two other sources. Nothing.
`not_applicable`	The field is meaningless for this unit — total funding raised for LSEG.
`unclear_definition`	The company's own description does not resolve to a codebook value.
`conflict_unresolved`	Two sources of equal standing disagree. Both are recorded in notes.
`out_of_scope`	The unit turns out not to meet the inclusion criteria.

A high gap count on a field is a finding about the sector's transparency, not a defect in the research.

Two known limits, declared up front

1 · Pricing is expected to be sparse. Most B2B wealthtech does not publish prices. A large number of undisclosed values and not_found_after_search gaps on pricing_detail is the correct outcome. undisclosed is a legitimate coded value — used with evidence that no public pricing exists, such as a "contact sales" page — and is distinct from a gap, which means the search itself came up empty.

2 · The roster mixes scales. LSEG, MSCI, FactSet, Morningstar, SimCorp, Temenos, Comarch, and Tietoevry are large incumbents sitting alongside seed-stage startups. Total-funding and last-round are not meaningful for a public company. A large_incumbent flag exists so those fields degrade honestly into not_applicable gaps instead of being forced into a number.

Verification

Three fields auto-enqueue a second verification pass whenever the first pass returns low confidence:

category — companies straddle boundaries, and the single-best-fit rule is inference-prone.
pricing_model — the highest-risk field for fabrication.
unique_ip — the field most likely to be filled with marketing paraphrase.

A patent claim should cite a patent office or the company's own patent listing. The phrase "proprietary AI" in a hero banner does not, by itself, establish IP.

Reproducing this

The study lives at RESEARCH/WEALTHTECH100_2026/ with an append-only ledger of every task the pipeline ran. The instrument is instrument/instrument.md; the coded output is data/units.json. This site is a rendering of that file — the site build reads the units directly and regenerates every figure, so no number here is typed by hand.