I have used AlphaSense at work for nearly two years. At first, I thought its best feature was search: finding the sentence I needed in a mountain of documents. As the product moved toward generated answers, deeper research, and workflow execution, a different advantage became clearer. The better the AI became, the more I cared about what it was allowed to read.
That is a narrow observation from my own use, not a claim about every customer’s experience or contract. Still, the distinction is hard to miss. Public filings are accessible in many places. Expert interviews, broker research, licensed documents, and a firm’s internal work product have different access rights and provenance.
Then I looked closely at Rogo and had the opposite reaction: could it take over the whole financial intelligence stack? A product that turns research into a deck, model, memo, and email can own much more of a professional’s day than a search box can. Answering that question, however, requires separating value created for the industry from revenue available to software vendors.
Rogo wants the work surface
Rogo announced a $160 million Series D in April 2026. In September, it said it had been deployed to more than 50,000 professionals at 350-plus firms. Those are company-reported deployments, not independently audited paid active seats. Series D, September update
Its Felix agent is designed to turn a prompt into sourced research, PowerPoint presentations, Excel models, and Word documents. Rogo also places it in email and an Excel add-in. The important shift is that the product can start the task and finish the deliverable. If it becomes the default work surface, users may spend less time opening each data terminal or research portal directly. But controlling the surface is different from owning every input that flows through it. Rogo product update
How big is financial AI, really?
Four numbers often get presented as if they were successive estimates of the same TAM. They measure different things.
Financial AI: four numbers, four definitions
The underlying economic pool
McKinsey’s 2026 banking review mentions about $7.3 trillion in 2025 global banking-industry revenue and $1.3 trillion in bank net income. The two figures should not be combined into a margin without checking their different boundaries. Nor is the revenue figure a clean sum of every insurer, capital-markets business, and asset manager. A separate 2024 review broke the 2023 global financial-intermediation revenue pool of roughly $6.8 trillion into retail banking (33%), corporate and commercial banking (28%), payments (16%), wealth and asset management (14%), investment banking (5%), and other activities. That puts the 2023 economic revenue base for wealth and asset management near $0.95 trillion and investment banking near $0.34 trillion. Neither is AI-vendor revenue. I do not apply those older percentages mechanically to the 2025 number. 2026 review, 2024 review
This is an economic base exposed to AI, not software TAM. Much of any gain will stay with customers, employees, shareholders, and internal technology teams.
Annual GenAI value potential
McKinsey Global Institute estimated that fully implemented generative-AI use cases could add $200 billion to $340 billion of annual value in global banking. It also described that as 2.8% to 4.7% of industry revenue and 9% to 15% of operating profit. This is a potential based on a 2023 study, not realized 2026 revenue. Productivity is central to the estimate; the broader discussion also covers better decisions and lower risk. MGI research, profit comparison
Corporate and investment banking has credible use cases in pitch books, research, and client work. That opportunity sits inside the wider banking estimate; adding it again would double count. McKinsey on CIB
Subsector estimates need their own labels. McKinsey has cited about $60 billion of combined traditional-AI and generative-AI value potential in global asset management. Its 2026 European private-banking study models €4–5 billion of profit uplift from a redesign that includes AI. The latter concerns a regional scenario; the former is not GenAI alone. Neither is a clean decomposition of the banking-wide $200–340 billion, so I do not add them to it. Asset-management analysis, private-banking analysis
A $200–340 billion AI value pool is not the revenue pool available to financial-AI software companies. Even if AI creates $300 billion of economic value, the central investment question is what fraction outside vendors can price and retain.
Observable external vendor spend
Reported segment revenue gives us a firmer, though imperfect, view of what customers already buy. LSEG reported £3.978 billion of 2025 Data & Analytics revenue excluding recoveries: £1.925 billion Workflows, £1.822 billion Data & Feeds, and £0.231 billion Analytics. Those components are already inside the total. LSEG annual report
Observable financial data and workflow revenue
USD · 2025
GBP · 2025 · separate scale
| Public-company segment | 2025 revenue | Why include it | Boundary |
|---|---|---|---|
| LSEG Data & Analytics | £3.978B | Data, feeds, Workspace | Excludes recoveries; GBP |
| S&P Global Market Intelligence | $4.916B | Capital IQ, data and workflow | Also contains credit and enterprise solutions |
| FactSet, consolidated | $2.322B | Financial workstations and data | Includes other products and possible resale |
| Morningstar Direct Platform | $0.831B | Investment analytics surface | Asset-management focus |
| PitchBook | $0.672B | Private-market data and workflow | Separate Morningstar segment |
The four US-dollar entries total about $8.74 billion. I leave LSEG’s £3.978 billion in its original currency rather than introduce an arbitrary exchange date. Bloomberg has no directly comparable public segment disclosure. MSCI’s broader revenue includes asset-based index fees, so I leave it out of this narrower basket. AlphaSense’s reported ARR is not recognized revenue and is excluded as well. This basket illustrates an existing multi-billion-dollar external-spend order of magnitude, not a definitive market total or even a strict mathematical floor: resale and cross-licensing could duplicate some end-customer spend. S&P filing, FactSet filing, Morningstar filing
A narrower SAM for financial knowledge-work AI
For Rogo- or AlphaSense-like products, I count seats that repeatedly read documents, analyze companies or transactions, and produce research or client deliverables. The table is my scenario, not an official occupation census. I keep investment banking, PE, public-market investing, credit, wealth, and corporate strategy/IR in separate buckets and exclude most trading and operational roles. The public scale checks are FactSet’s 237,324 professionals using its service in FY2025 and Rogo’s 50,000-plus announced deployments in September 2026. Neither is the number of potential global buyers; deployment need not equal active paid use. FactSet filing, Rogo
| Vertical | Addressable seats, thousands (low/base/high) | Annual spend per seat, $thousands (low/base/high) | Base case |
|---|---|---|---|
| Investment banking | 80 / 120 / 180 | 4 / 8 / 15 | $0.96B |
| Private equity | 60 / 100 / 140 | 3 / 7 / 12 | $0.70B |
| Asset management, public equities, HF | 120 / 200 / 280 | 3 / 6 / 12 | $1.20B |
| Credit | 60 / 100 / 150 | 2 / 5 / 10 | $0.50B |
| Wealth | 150 / 275 / 400 | 1 / 2 / 5 | $0.55B |
| Corporate development, strategy, IR | 100 / 175 / 250 | 1 / 3 / 6 | $0.525B |
| Total | 570 / 970 / 1,400 | Varies by vertical | $1.23B / $4.44B / $12.74B |
I multiply seats by annual spend within each vertical, then sum. $1.23–12.74 billion is a scenario range for potentially addressable annual spend, not current revenue or a forecast. Bundled data subscriptions and pure incremental AI software spend may be mixed. Whether a new sale comes from an existing data budget or a new one is the largest unresolved question. The assumptions and calculation are recorded in the repository’s research dossier.
External counts offer scale checks, not a global seat census. The US Bureau of Labor Statistics reports 377,200 financial and investment analysts and 299,400 personal financial advisors in 2025; CFA Institute counts more than 200,000 charterholders worldwide. Dividing FactSet’s FY2025 $2.322 billion of company revenue by its 237,324 reported users at fiscal year-end gives about $9,800 per reported user. That is a revenue-to-point-in-time-user check, not a subscription price. The user count excludes users from FY2025 acquisitions, so numerator and denominator do not fully align. The US occupations are broader than the eligible seats here, charterholders overlap with those jobs, and the samples cover different geographies. I neither sum them nor extrapolate them into the SAM. In wealth, the scenario targets a subset of advisors with repeated, valuable research and document work. US analysts, US advisors, CFA Institute, FactSet filing
What does Rogo read?
Model quality and a smooth interface matter, but a financial answer also depends on source access, freshness, rights, and citations. Output ≈ Model × Context is a conceptual heuristic, not a statistical formula. As models improve, scarce context may become more visible as a source of product differentiation.
AlphaSense’s current company pages describe more than 500 million documents, 300,000-plus expert interviews, and 8,000-plus new interviews a month. Its broker-research pages use differing definitions: 1,500 partners on one and 1,700-plus sources on another. I use the conservative 1,500-plus here. The moat is more than quantity. Licensing and access rights, a historical corpus, permission-aware retrieval, and traceable citations determine how much of that content an AI can safely use. Content, Expert Insights, broker research
AlphaSense’s $930 million agreement to acquire Tegus in 2024 makes strategic sense in this light: it added expert interviews and private-company material to the context the platform can search and reason over. The price paid is not a standalone appraisal of the content. Company announcement
Rogo connects to LSEG
The 2025 LSEG–Rogo agreement is an instructive counterexample to simple disintermediation. Rogo users who hold an LSEG Workspace license can access LSEG fundamentals, estimates, and M&A data inside Rogo. LSEG describes AI applications as distribution channels for its data. Rogo can win the work surface while LSEG retains the data relationship and rights. Partnership, LSEG annual report
PitchBook similarly says its Premium Connector in Rogo requires both a PitchBook license and a paid Rogo account. Preqin lists Rogo as a partner, while Rogo has announced Third Bridge expert-transcript integration. Access terms differ by source; none of this implies a universal free pool of financial data. PitchBook, Preqin, Third Bridge integration
AlphaSense is moving up the stack
Calling AlphaSense a static search product misses its own expansion. It announced Generative Search / Grid in March 2025, Deep Research in June 2025, Custom AI Agents in March 2026, SuperAnalyst in June 2026, and Work Products in July 2026. These are announcement milestones, not a claim that every feature was generally available or part of my personal use. Generative Search, Deep Research, Custom AI Agents, SuperAnalyst
Verified AlphaSense product milestones
AlphaSense is moving from content and search into reasoning and workflows. Rogo starts with the agentic workflow and reaches down into data and integrations. They meet in a real competitive zone. It does not follow that either must own every layer.
The financial intelligence value stack
↑ AlphaSense · content to workflow | ↓ Rogo · workflow to data integration
Two companies meet from different directions
Finance is not one workflow
An investment-banking pitch, a PE data-room review, a hedge-fund research process, a credit memo, and a wealth client’s proposal use different inputs, outputs, time horizons, and controls. The matrix is a conceptual comparison; latency ranges describe the character of the work, not industry service standards.
Finance is not one workflow
| Vertical | Task | Scarce data | Output | Latency | Control | AI opportunity |
|---|---|---|---|---|---|---|
| IB | Pitch/deal | Deal and market data | Deck/model | Days–weeks | Client approvals | Drafting/comps |
| PE | Sourcing/diligence | Private co./VDR | IC memo | Days–weeks | Confidentiality | Diligence synthesis |
| Public/HF | Ideas/monitoring | Research/experts | Investment memo | Minutes–days | Sources/MNPI | Signal detection |
| Credit | Underwriting | Contracts/borrower | Credit memo | Hours–weeks | Audit trail | Covenant extraction |
| Wealth | Advice/portfolio | Client/product data | Client proposal | Hours–days | Suitability/privacy | Tailored advice |
| Corp. strategy | Market/M&A | Competitors/internal | Strategy report | Days–months | Internal data | Market mapping |
A general interface can launch all of these tasks. Deep data, review steps, and accountability within each task are harder to standardize. The durable niche is not simply a vertical label. It is a specific data right, recurring team workflow, or regulatory context that a broad agent still needs to respect.
Three moats and two investment theses
Three moats in financial AI
Scarce content, rights, history, provenance
AlphaSense starting pointModels, deals, research, deliverables
Rogo starting pointSecurity, compliance, seats, procurement
Both are building itThe data moat is about rights, historical depth, provenance, and permission-aware access as much as corpus size. The workflow moat lies in finished models, decks, emails, and deal processes. The distribution and trust moat is built through security review, compliance, installed seats, and procurement relationships. AlphaSense has a distinctive content starting point; Rogo has a workflow starting point. Both are investing in trust. That is an interpretation of public evidence, not a winner call.
AlphaSense said in June 2026 that it had raised $350 million at a $7.5 billion valuation and had exceeded $600 million of ARR in Q1 2026. Dividing the headline numbers gives less than 12.5×. This is a private-round headline valuation / reported ARR ratio, not EV/revenue: ARR is not recognized revenue, and the enterprise-value bridge is not public. The company also reported 7,000-plus enterprise customers. A separate company marketing page claims 90% of the S&P 100 as customers; that is not an independently reconstructed client count. The premium, if sustained, would be a bet on licensed content, distribution, switching costs, retention, and more workflow spend per customer. The risk is that workflow-first products take the front door and leave AlphaSense closer to a premium content supplier. The counter-thesis is that AlphaSense owns both content and the work surface. Company announcement, S&P 100 claim
Rogo’s $2 billion valuation is reported by Bloomberg; Rogo’s own Series D release does not disclose it. I do not construct a revenue multiple without a reliable public revenue figure. The positive thesis requires repeat use across workflows, data orchestration, model flexibility, and enterprise deployment economics that improve with scale. Dependence on licensed inputs, incumbents building their own agents, costly implementation, and workflow fragmentation are the tests on the other side.
Winner-take-interface, many-take-the-layers
One interface can sit above many layers
Financial professionals could converge on one or a few AI work surfaces while many specialized providers remain underneath: market data, expert networks, broker research, private-market data, internal knowledge, and dedicated workflow tools. Interface consolidation does not imply data-source consolidation. A world where Rogo becomes very large and a world with many successful niche financial-AI and data vendors are compatible.
For an investor, the useful questions are narrower than “who wins financial AI?” Who controls context competitors cannot easily reproduce? Who owns a completed, recurring workflow? Who has the security and distribution to keep capturing value? AI may clarify which niches are genuinely scarce rather than erase them all.
Reading the numbers
Economic value ≠ software TAM. Deal volume ≠ vendor revenue. ARR ≠ recognized revenue. Private headline valuation / ARR ≠ EV/revenue. Public-company segment revenue is a proxy with different boundaries and possible overlap, not a complete TAM. The seat-based SAM is assumption-sensitive; read the low, base, and high cases together. Data cutoff: October 3, 2026.
Discussion
Discussion
Have a question, correction, counterexample, or another interpretation?