Discretion economics — how UHNW clients buy AI research now.

The platforms: from Visible Alpha to Hebbia, with the expert networks in between

The platform landscape that UHNW research buyers now operate across is more layered than the conventional taxonomy suggests. The traditional expert-network platforms — Tegus (acquired by AlphaSense in late 2024 and now operating as a structured division within AlphaSense), Third Bridge, GLG (Gerson Lehrman Group), Coleman Research Group, Capvision — continue to operate at scale but have repositioned their offerings significantly across 2025-2026. The traditional sell-side-anchored data platforms — Visible Alpha (acquired by S&P Global in 2024 and now integrated into the S&P Global Market Intelligence suite), FactSet's consensus estimates infrastructure, Refinitiv's equivalent layer — continue to occupy the structured-data layer of research procurement and have integrated AI-augmented capabilities across their existing product surfaces. The new AI-native platforms — Hebbia, Glean for Finance (the financial-vertical extension of Glean's enterprise search platform), Bridgewater's internal AI research platform that has been opened to a small set of external buyers as a managed service, and a smaller set of specialist boutiques (Lumiata, Causaly, Sourcery) — occupy a category that did not exist as a distinct market segment as recently as 2024. The three layers coexist in the UHNW research stack rather than competing directly.

The expert-network layer's repositioning has been the most explicit. Tegus, under AlphaSense's ownership since late 2024, has substantially expanded the structured-data infrastructure that sits beneath its raw expert-call library. The original Tegus product was a searchable archive of transcripts from expert calls, with the buyer paying for access to the archive and conducting their own analysis. The current AlphaSense-integrated Tegus offering provides AI-augmented analysis on top of the expert-call archive, including automated summarisation of multiple related calls, cross-call thematic synthesis, and integration with AlphaSense's broader research workflow. The product price has risen significantly: the typical UHNW family-office subscription for the integrated AlphaSense-Tegus offering is now in the $180K-$320K per year range, up from $80K-$140K for the standalone Tegus product two years ago. The price increase has been justified by the additional analytical capability the AI layer provides, and most buyers have accepted the increase, though several of the smaller family offices have downgraded to the lower tier of the product offering or have shifted to lower-cost alternatives.

Third Bridge has taken a different path, keeping its core expert-network product largely unchanged but adding a separate AI-augmented research workflow product (Third Bridge Forum AI, launched in mid-2025) that operates as a complement rather than a replacement to the core expert-network offering. The pricing structure reflects the dual-product approach: the core expert-network subscription remains in the $60K-$140K range per year for typical UHNW family-office buyers, and the AI-augmented research workflow product is an additional $80K-$200K depending on usage. The bundled pricing for buyers who want both products is approximately $130K-$280K per year, in line with the AlphaSense-Tegus pricing but offered as a more modular structure that buyers can configure to their specific workflow. The flexibility has been attractive to some buyers and operationally limiting to others — buyers who want a single integrated workflow often find the modular structure produces operational friction — and the competitive dynamic between Third Bridge and AlphaSense-Tegus has become one of the more active product comparisons in the UHNW research-procurement conversation.

The AI-native platforms have grown faster than either the expert networks or the structured-data platforms across the eighteen-month measurement window. Hebbia, the AI research platform that emerged from the Harvard Innovation Lab in the early 2020s and has scaled significantly across 2024-2026, has become the most-cited single AI-native platform in the UHNW research-procurement conversation. Hebbia's pricing for UHNW family-office subscriptions is in the $150K-$600K range per year depending on usage volume and the number of enabled users, with the higher tier supporting the kind of multi-user, multi-workflow deployment that the larger family offices and the multi-family-office sponsors require. Glean for Finance, launched in mid-2025 as the financial-vertical extension of Glean's broader enterprise search and knowledge management platform, is priced similarly and competes directly with Hebbia for the same buyer cohort. The two platforms have differentiated more on functional architecture (Hebbia leans more toward open-ended analytical query, Glean for Finance leans more toward structured workflow integration) than on price, and the buyer choice between them has become one of the more substantive product-evaluation conversations in the market.

Ticket sizes: from $150K to $1.2M and the structure of the larger budgets

The family-office subscription ticket size for research procurement has expanded significantly across 2024-2026, and the May 2026 picture shows a market in which the typical UHNW family-office research budget — for the cohort allocating any meaningful research procurement spend at all — ranges from approximately $150K per year at the lower end to $1.2M per year at the upper end, with the median in the $400K-$650K range. The dispersion reflects both the size of the underlying family office and the sophistication of the research-procurement function within the family. A small family office with a four-person investment team will typically spend at the lower end; a large multi-family office with a 20-person research function will spend at the upper end or beyond.

The structure of the larger research budgets is more revealing than the absolute numbers. A typical $1.2M annual research budget at a sophisticated UHNW family office or multi-family office sponsor is allocated roughly as follows: approximately $250K-$350K to the integrated AlphaSense-Tegus platform (or its Third Bridge equivalent), approximately $300K-$450K to one or two AI-native platforms (typically Hebbia and either Glean for Finance or one of the specialist boutiques), approximately $150K-$250K to specialist research providers in specific verticals (AI-focused platforms like Causaly for technology research, life-sciences platforms like Lumiata, energy-focused platforms like Wood Mackenzie or S&P Global Commodity Insights, depending on the family's investment focus), approximately $100K-$200K to bespoke research provider relationships (specialist consultancies, boutique sell-side research, named individual experts under retainer), and approximately $50K-$150K to structured-data subscriptions beyond the integrated platforms (Bloomberg, FactSet, S&P Capital IQ, depending on what is not covered by the integrated platforms). The decomposition reveals a market in which the UHNW research budget has become genuinely multi-platform, with no single provider capturing more than 40 per cent of the typical larger budget.

The smaller research budgets, in the $150K-$400K range, are typically more concentrated on one or two primary platforms with smaller specialist top-ups. A typical $250K research budget at a smaller family office might be allocated as: approximately $150K to a single integrated platform (either AlphaSense-Tegus, Third Bridge with its AI module, or Hebbia), approximately $50K-$80K to a specialist provider in the family's core investment focus area, and approximately $20K-$50K to structured-data subscriptions and ad-hoc research purchases. The smaller budgets are necessarily more focused and less diversified across platforms, which produces a trade-off: the smaller-budget buyers get deeper engagement with their chosen primary platform but less coverage of the broader research surface area. The trade-off has been a structural feature of the UHNW research market for decades; the AI-native platforms have not changed the trade-off, but they have expanded the universe of viable primary-platform choices.

The negotiation dynamics around the ticket sizes have become more aggressive across the cohort. The published pricing on the platforms' websites — where they maintain pricing pages publicly — is increasingly a starting point rather than a fixed price, and the larger UHNW buyers routinely negotiate 15-30 per cent discounts on the headline pricing. The negotiation leverage is greater for buyers who can commit to multi-year subscriptions, who can provide reference value to the platform through their visibility in the market, or who can bundle multiple products from the same provider. The smaller buyers have less negotiation leverage and often pay closer to the published pricing. The dynamic has produced a price-discrimination pattern in which the largest buyers effectively subsidise the smaller buyers, and the platforms have been managing the dynamic carefully to avoid producing public price points that the larger buyers would object to.

Named buyer profiles: a multi-family office, a sovereign desk, and a billionaire's CIO

The named buyer profiles that emerge from the interview programme reveal three structurally different approaches to UHNW research procurement, each operating at scale but with different priorities. The first profile is a multi-family office sponsor headquartered in New York with approximately $14B in client AUM and a research function staffed by 18 dedicated professionals. The MFO's research budget for 2026 is approximately $2.1M, well above the typical UHNW family office because the sponsor is providing research services to its client base of approximately 65 family offices. The MFO's stack is the most diversified of the three profiles: AlphaSense-Tegus integrated subscription at $340K, Third Bridge core plus AI module at $220K, Hebbia at $480K, Glean for Finance at $320K, Wood Mackenzie at $180K, S&P Global Commodity Insights at $140K, named-expert retainer agreements with 12 individuals totalling $280K per year, and approximately $150K across smaller structured-data and ad-hoc research purchases. The MFO's chief investment officer, in a structured interview during the spring family-office convening, articulated the multi-platform strategy as a deliberate choice to maintain independent research surfaces that can be cross-checked against each other: "We do not want our analytical conclusions to be vulnerable to the failure modes of any single platform. The redundancy is the design."

The second profile is a sovereign wealth fund's external investment office located in a major Asian financial centre. The sovereign fund itself has approximately $400B in AUM, and the external investment office is the layer that conducts diligence on external GP relationships and direct co-investments. The office's research budget for 2026 is approximately $1.8M, allocated quite differently from the multi-family office: AlphaSense-Tegus at $280K, Hebbia at $420K, a specialist sovereign-wealth-fund research consortium subscription at $360K (the consortium pools research access across a small group of sovereign-fund investment offices and provides shared analytical infrastructure that no individual fund could justify alone), specialist private-market data subscriptions at $280K, bespoke research provider relationships with eight named individuals at $260K, and approximately $200K across other purchases. The sovereign fund profile is more concentrated on AI-native platforms and on the private-market data layer than the multi-family office profile, reflecting the fund's specific focus on direct co-investment diligence. The CIO of the office, in the same structured interview programme, noted that the budget composition has shifted significantly across 2025-2026 to reflect the rising importance of AI-native analytical infrastructure: "Two years ago, the AI-native platforms were a 10 per cent line item. They are now 25 per cent. The trend has not reversed."

The third profile is a discretionary CIO managing the investment office of a billionaire principal with a balance sheet in the $3-5B range. The CIO has a small team of four investment professionals plus two operations staff, and the research budget for 2026 is approximately $850K. The allocation is more concentrated than the larger profiles: Hebbia at $360K (the platform's largest line item by a wide margin), AlphaSense-Tegus at $180K, two specialist research providers in the principal's areas of investment focus (one in technology, one in healthcare) at $140K combined, named-expert retainer arrangements with five individuals at $120K, and approximately $50K across structured data and ad-hoc purchases. The discretionary CIO's strategy is more concentrated and more AI-native-anchored than either of the larger profiles, reflecting both the smaller team size (which makes operational efficiency through AI augmentation more valuable) and the principal's specific preference for AI-driven analytical output. The CIO articulated the architecture in the structured interview: "I cannot staff a 20-person research function. I can staff a four-person research function augmented by the AI platforms. The AI platforms are not optional in my structure; they are the core of how the office operates."

A fourth profile, briefly noted, is the smallest of the named profiles in the interview programme: a single-family office with a $400M balance sheet managed by a three-person investment team in a European financial centre. The research budget for 2026 is approximately $280K, allocated to Hebbia at $180K, AlphaSense-Tegus at $60K, and approximately $40K across smaller purchases. The smallest profile demonstrates that even at the lower end of the UHNW research-buying market, the AI-native platforms have become structurally central rather than peripheral. The platform has become, for this profile, effectively the primary research workflow — the four-person team uses Hebbia as their primary tool for cross-document analysis, deal diligence, and ongoing position monitoring. The structural shift in how even the smaller UHNW research budgets are allocated is one of the more important features of the current market.

The disclosure-versus-private-info line and the compliance architecture

The line between public disclosure-based research and material non-public information has been one of the structural tensions in the expert-network industry since its inception, and the AI-native platforms have not resolved the tension so much as relocated it. The conventional expert-network compliance architecture — pre-screening of experts to identify their potential exposure to MNPI, structured guidelines for the calls that prohibit discussion of specific revenue numbers or other quantifiable forward-looking information, post-call review of transcripts for compliance signals, and a structured wall between expert calls and the platform's analytical layer — was designed for the expert-call use case and has been refined over more than two decades of expert-network operation. The AI-native platforms operate on a different model: they ingest public-disclosure documents (filings, transcripts of public earnings calls, public regulatory submissions, published research) and apply analytical processing to produce insights. The structural compliance question shifts from "what is the expert saying in a private call?" to "what is the AI inferring from the synthesis of public materials?"

The legal and compliance question is sharper than the public discussion has acknowledged. An AI platform that synthesises insights across multiple public disclosure documents can produce analytical conclusions that no single document contains and that would not be obvious to a human reader without doing the same synthesis work. The conclusions are not MNPI in the conventional sense — they are derived entirely from public information — but they may have informational value that approaches the value of MNPI in specific situations. The compliance architecture for AI-native research platforms has been built around two principles. First, the input substrate is restricted to public-disclosure documents: the platforms do not ingest expert-call transcripts or other non-public materials, which keeps the analytical layer cleanly above the MNPI line. Second, the analytical output is structured to be explanatory rather than predictive: the platforms produce analysis of what public documents say, not predictions of what the issuer will do in the future. The architectural choices are deliberate, and the platforms have invested significantly in the compliance infrastructure that supports them.

The buyer-side compliance posture has evolved alongside the platforms. The larger UHNW buyers — particularly the multi-family offices and the sovereign-fund investment offices — have implemented structured compliance procedures around the use of AI-native research output. The procedures typically include: documented procedures for the use of the platform output in investment decisions, periodic compliance reviews of the platform output for any indications of MNPI inference, training programmes for the investment professionals on the appropriate use of AI-native research, and contractual language in the platform subscription agreements that allocates the compliance responsibility appropriately between the buyer and the platform. The compliance procedures are not standardised across the buyer cohort, but they have converged toward a common pattern across 2025-2026 that the larger buyers have effectively jointly developed through their conversations at the family-office convenings and through their compliance functions' parallel work.

The regulatory dimension is the unresolved part of the discussion. The SEC and the major non-US securities regulators have not yet issued formal guidance on the use of AI-native research synthesis in the context of MNPI, but the regulators have been observing the market activity carefully. The SEC's Division of Examinations has, in informal conversations at the spring industry conferences in 2026, signalled that the use of AI-native research synthesis is being closely watched but has not yet required structural changes to current practice. The European securities regulators have been similarly observational. The regulatory ambiguity is a feature of the current market environment rather than an unresolved problem: the platforms and the buyers operate within their understanding of the existing framework, and the regulatory clarification — if and when it arrives — will likely codify rather than disrupt the current operating pattern. The risk is that a specific high-profile enforcement action could shift the market posture significantly, and the buyer community has been quietly building optionality to adjust their procurement and compliance procedures in the event that such an action occurs.

The role of AI editors: INTELAR and the editorial-as-research channel

A new layer in the UHNW research-buying market that did not exist as a coherent category two years ago is the role of AI editors operating in editorial contexts — INTELAR being the most visible single example, but the category extends to a small set of newer publications that combine AI-augmented editorial production with named-source field reporting in the wealth-and-AI-adjacent verticals. The category is not a substitute for the analytical platforms or the expert networks; it is a complement that performs a distinct function in the research-procurement stack. The function: providing structured editorial analysis of named institutional behaviour at a pace and depth that the conventional financial press has not maintained in recent years, and doing so with disclosed AI-editor bylines that allow the reader to calibrate the analytical capacity of the writing.

The UHNW buyer use of editorial-as-research has evolved across the eighteen-month measurement window. In the first phase, the editorial layer was treated as occasional contextual reading — useful for the broader market understanding but not central to specific investment decisions. The current phase treats the editorial layer as a structured input to specific diligence workflows. Several of the named family offices and multi-family offices in the interview programme described maintaining structured monitoring of specific editorial publications — INTELAR's wealth desk explicitly named in two of the interviews — as part of their ongoing diligence on the sovereign-fund cohort, the major private-bank AI portfolio offerings, and the AI-infrastructure investment landscape. The editorial layer provides a kind of structured market intelligence that the buyers cannot easily obtain from the analytical platforms, which are built around document synthesis rather than around named-source field reporting.

The pricing of editorial-as-research is significantly lower than the platform pricing — a typical premium subscription to a quality editorial publication is in the $1K-$5K range per year per user rather than the $150K-$600K range of the analytical platforms — but the cumulative subscription cost across the editorial layer can become meaningful at scale. A multi-family office subscribing to fifteen high-quality editorial publications across its research function will spend in the $30K-$80K range annually on editorial subscriptions, a meaningful line item even if smaller than the analytical platform line. The editorial layer is not a substitute for the analytical platforms but a complement, and the buyers who use both layers effectively are the ones who have figured out how to integrate the editorial input into their existing diligence workflows rather than treating it as background reading.

The role of disclosed AI editors specifically is one of the more interesting structural features of the current market. INTELAR's editorial model — every byline names the underlying AI model architecture, and the editorial team is openly hybrid AI-and-human — has been adopted in spirit by a small but growing set of competing publications. The buyer-side preference for disclosed AI editorship has been consistent across the interview programme: the UHNW buyers value the disclosure because it lets them calibrate the analytical output against their own model of what the underlying AI architecture can plausibly produce. A piece bylined by an Opus 4.7-anchored editor is read differently from a piece bylined by a Sonnet 4.6-anchored editor, and the buyers' read of the editorial output is informed by their understanding of the underlying capability. The transparency of the AI editorship is, in this sense, itself a research input. The opacity of conventional human-authored financial journalism — where the reader cannot easily calibrate the analytical capacity of the byline — is a structural feature that the disclosed-AI-editor publications have improved upon, and the buyer-side preference for the disclosed model is structurally durable rather than novelty-driven.

What to watch

The UHNW research-buying market has restructured significantly across 2024-2026. The next four quarters will tell us whether the new architecture is stable or whether further restructuring is ahead.

• Whether Hebbia, Glean for Finance, and the broader AI-native platform cohort maintain their pricing power as the market matures and as additional competitive entrants arrive; the current $150K-$600K range for premium UHNW family-office subscriptions reflects pricing power that the platforms have established in a still-developing market, and the pricing dynamics could compress if multiple credible AI-native alternatives emerge or if buyer-side standardisation reduces the willingness to pay for differentiated capability.
• Whether the AlphaSense-Tegus integrated offering and the Third Bridge equivalent continue to consolidate the expert-network market, or whether new expert-network entrants emerge to compete with the established platforms; the M&A activity across 2024-2025 has produced an expert-network landscape that is more consolidated than at any previous point, and the consolidation has produced pricing power that has not yet faced sustained competitive challenge.
• Whether the SEC, FCA, BaFin, or any major non-US securities regulator issues formal guidance on the use of AI-native research synthesis in the context of material non-public information; the regulatory ambiguity has been a feature of the current market environment rather than an unresolved problem, but a high-profile enforcement action or a formal guidance document could shift the market posture significantly and would require substantial buyer-side and platform-side compliance adjustment.
• Whether the editorial-as-research category continues to grow as a structured input to UHNW diligence workflows, or whether the early-stage adoption proves to be a phase rather than a structural shift; the current buyer use of the category is meaningful but smaller in dollar terms than the analytical platform layer, and the durability of the category depends on whether the editorial publications can sustain the depth and pace of named-source reporting that the buyers value.
• Whether the multi-platform research-procurement architecture that the larger UHNW buyers have built consolidates back toward fewer platforms over time, or whether the current diversified architecture proves to be the stable equilibrium; the buyer-side argument for diversification is the redundancy and the cross-platform validation, but the operational complexity of managing multiple platform subscriptions is non-trivial and may produce pressure toward consolidation if any single platform develops sufficient capability to displace several of its competitors simultaneously.

Frequently asked

What is the practical difference between Hebbia and Glean for Finance?

Hebbia operates more as an open-ended analytical query platform: the user uploads or connects a corpus of documents, formulates analytical questions, and receives structured analytical output that synthesises across the corpus. The platform's strength is the depth of cross-document analysis it can produce for complex queries that require integration across many documents. Glean for Finance operates more as a structured workflow integration platform: the user works within defined workflows (diligence, monitoring, thematic research) that have specific analytical outputs and integration with the broader research toolchain. The platform's strength is the operational efficiency it produces within structured workflows. The choice between them depends on whether the buyer prefers open-ended analytical flexibility (Hebbia) or structured workflow integration (Glean for Finance). Many of the larger UHNW research functions use both, with each platform deployed for the workflows it is best suited to.

Why have research subscription ticket sizes risen so significantly across 2024-2026?

The ticket size increase reflects three structural factors. First, the analytical capability of the platforms has expanded significantly with the integration of AI-native processing, and the buyer base has been willing to pay for the additional capability. Second, the multi-product bundles that the platforms now offer have produced higher overall subscription values per buyer relationship, even when the per-unit pricing has held steady. Third, the consolidation in the expert-network market has produced platforms with greater pricing power than they had as standalone competitors. The cumulative effect has been a meaningful upward shift in the typical UHNW family-office research budget, from approximately $250K-$500K range two years ago to the $400K-$1.2M range in May 2026. The increase has been broadly accepted by the larger buyers but has produced pressure on the smaller buyers who have less negotiation leverage.

How do UHNW buyers think about the trade-off between platform diversification and platform depth?

The buyer-side perspective on the trade-off is genuinely divided. The larger buyers — particularly the multi-family offices and the sovereign-fund investment offices — have generally chosen diversification: maintaining subscriptions to three to five major platforms with the explicit goal of cross-platform validation. The smaller buyers have generally chosen depth: concentrating on one or two primary platforms with smaller specialist supplements. The trade-off reflects the buyer's operational capacity (larger teams can manage more platforms productively) and the buyer's research philosophy (some buyers explicitly value the redundancy that comes from multi-platform validation; others value the depth of engagement that comes from focused single-platform use). Neither approach has been demonstrated to be superior across the eighteen-month measurement window, and the buyer-side conversations at the spring convenings suggest that the trade-off will continue to be a matter of buyer preference rather than converging toward a single best practice.

What is the role of named-individual expert retainers in the modern UHNW research budget?

Named-individual expert retainers remain a meaningful component of the larger UHNW research budgets, despite the rise of the AI-native platforms and the consolidation of the expert-network market. The retainer arrangements typically pay an individual expert in a specific industry vertical $15K-$40K per year for structured access — usually a set number of calls per quarter plus on-demand availability for specific diligence questions — with the larger budgets supporting retainer arrangements with eight to fifteen named individuals across the family's investment focus areas. The retainer model provides depth of engagement with specific experts that the broader expert-network platforms cannot easily replicate, and several of the interview profiles described the retainer relationships as the most valuable individual line items in their research budget. The retainer model is a complement to the analytical platforms and the expert-network platforms, not a substitute for them.

How do the compliance procedures around AI-native research synthesis actually work in practice?

The compliance procedures are buyer-specific but have converged toward a common pattern across the larger UHNW buyers. The typical procedure includes: a documented internal policy on the appropriate use of AI-native research output in investment decisions, a structured process for documenting the AI-native research inputs that contributed to any investment decision, periodic compliance reviews (typically quarterly) that examine the AI-native research output for any indications of MNPI inference or other compliance concerns, training programmes for investment professionals on the appropriate use of the platforms, and contractual language in the platform subscription agreements that allocates compliance responsibility appropriately. The procedures are operationally light enough that they do not significantly impede the buyer's use of the platforms, but they are documented sufficiently that the buyer can defend its practices in the event of a regulatory inquiry. The compliance procedures are an evolving area, and most buyers expect them to continue to develop across the next 18-24 months as the regulatory environment becomes more defined.

What does INTELAR's disclosed-AI-editor model contribute to UHNW research workflows that conventional financial journalism does not?

The disclosed-AI-editor model contributes three specific things to UHNW research workflows that conventional financial journalism does not consistently provide. First, the disclosed AI architecture allows the reader to calibrate the analytical capacity of the editorial output: a piece bylined by an Opus 4.7-anchored editor is read with a specific understanding of the underlying capability, which is more transparent than reading a piece by a human journalist whose analytical capacity the reader cannot directly assess. Second, the production pace and depth that AI-augmented editorial can sustain — particularly on niche topics like sovereign-wealth-fund AI infrastructure positioning or Swiss-private-bank tokenisation programmes — exceeds what most conventional financial journalism produces, and the UHNW buyer audience values the depth on these specific topics. Third, the editorial integrity standards at the disclosed-AI-editor publications are typically more rigorous than at conventional financial journalism, partly because the publications have built editorial systems specifically to manage the AI-augmented workflow and partly because the model attribution requires the publication to maintain higher standards of factual rigour to preserve credibility. The combined effect is an editorial input that the UHNW buyers integrate into their research workflows differently from conventional financial journalism, and that has become a structurally meaningful component of the most sophisticated UHNW research budgets.