GLG is the largest expert network by roster — >1M experts, >$400M of net revenue (CleverX) — and the most instructive company in the atlas for a reason that has nothing to do with its size.
It runs the oldest working version of the model this whole atlas is about: aggregate people, sell an outcome to institutional money, keep the difference. An expert who sets a rate of $200/hour produces a client bill of $800–$900 (Inex One forum). That 70–80% take has survived four decades, three compliance scandals and every attempt to route around it — the top of the What a rake can actually be league table, and the only entry on it observed from both sides of the trade.
The share collapse is the lesson
GLG's market share fell from 51% to 24% over a decade (Inex One), in a market that grew ~12%/yr over 2023–25 to roughly $3B in 2025 across ~11,200 client firms.
Read that carefully. The firm with the largest expert network on earth lost half its market to entrants who built rosters from scratch — AlphaSights (>$300M, 500k+ experts), Third Bridge (>$250M, 1.5M experts), Guidepoint, Dialectica, Atheneum. If the supply were the moat, that could not have happened. The million experts did not stop anybody.
What GLG kept is the take rate, not the share. That is the shape of the whole business: the rake is durable, the position is not, because the rake is paid for compliance and speed rather than for access to people. See Expert networks for the full argument and Which side you build first for why a competitor could bootstrap a roster but not a chaperone process.
The concentration contrast
No single GLG client exceeds 6% of revenue. That claim is in the research notes — one file carries it, together with top-ten clients at 19.0%, 90% of revenue recurring and 22 of the top 25 clients retained over a decade, all cited to the October 2021 S-1. Another file describes that same S-1 as the highest-value document nobody has opened. Both cannot be true, so the figure is carried here as directionally right and unconfirmed rather than as a filed number. See What we could not establish, item 24.
If it holds, it is the sharpest contrast available with the current cohort. Mercor took ~91% of H1 2026 revenue from AI foundation-model companies, dominated by OpenAI and Anthropic (The Information via BigGo). Appen was a $4.3B company with 80% of revenue in five clients; Alphabet terminated a ~$83M contract in January 2024 and the shares fell 40–41% in a day, on the way to a 97% drawdown (Appen FY2025 Annual Report). Eleven thousand client firms buying single calls is a completely different risk object from two labs buying nine-figure contracts. See One customer is a binary event.
The missing document
GLG filed an S-1 with the SEC in October 2021 and never listed (EDGAR). The document is public and sitting there.
The atlas is in two minds about it. Specific figures are quoted from it in the research — $589.1M of FY2020 revenue, $572.2M in FY2019, top-ten clients at 19.0%, no client above 6%, 90% of revenue recurring, 22 of the top 25 clients retained over a decade, segment contribution margin above 70%, $959.9M of debt, and the withdrawn IPO of March 2022 — and its non-exclusivity language is quoted on Getting cut out. A second research file simultaneously calls it the highest-value document nobody has opened. Both cannot be true.
Read in full it would give 2018–2021 revenue, gross margin, client concentration, expert payment costs and the then-ownership structure — audited, for the company that has run this model longest and at the highest take rate. Nothing else in the register would do as much work. Until someone opens it and reconciles it against the notes, every number on this page is provisional. See What we could not establish, item 24.
Do the labs buy from GLG?
The most valuable open question in the atlas, and the honest answer is that nobody knows.
No direct evidence surfaced either way. The research found no report of GLG selling expert time or transcripts to a frontier lab for training or evaluation — and no evidence that it declines to. The searches were run and returned nothing. [UNVERIFIED — searched, not found]
What is observable sits either side of the question. GLG and its peers have become buyers of AI rather than suppliers to it: AlphaSense bought Tegus for $930M in 2024 at a $4B mark (Reuters via Google News) and passed $400M+ ARR in March 2025 (AlphaSense via Google News) selling AI-moderated research. Meanwhile the labs' actual expert sourcing went to Mercor, Handshake AI and micro1 — none of which existed in this form three years ago, and all of which run at a 27–40% take instead of GLG's 70–80%.
So GLG has the supply, the compliance apparatus, the recruiting machine and the margin, and appears not to be selling into the fastest-growing buyer of exactly those things. Either it looked and declined — which would be worth knowing, because it would say something about what labs will and will not pay for — or it has not looked. See How much money is actually in the buyer pool and Expert data for frontier labs.
Where the rest of the record is thin
Founding date, headquarters, current ownership, current revenue, gross margin, headcount and capital structure are all unestablished here. The >$400M figure comes from a single industry blog and is net — an expert network invoices its fee and pays the expert out of it, so there is no gross billings line to confuse it with. The only filed revenue figure anywhere is the S-1's $589.1M for FY2020, five years stale and carried in the research rather than read from the document. The $200/$800–900 take comes from one forum answer plus one pricing blog. GLG has never printed a valuation. It is, for a company of its size and age, remarkably unmeasured — which is what happens when a business stays private, PE-owned and out of the trade press for forty years.