FDS earns ~29% ROIC on a subscription terminal/feed business with 95%+ client retention, yet trades at a 16x P/E versus a 33x ten-year average and 44% of base intrinsic value. The discount reflects a real reinvestment slowdown (5y ROIIC ~8%) and AI-disruption fear, not a broken compounder.

Plain English

FactSet sells a subscription to a research workstation that bankers, fund managers, and wealth advisors use every workday. They pay about thirty thousand dollars per seat per year, and once an analyst has built models, screens, and Excel formulas in FactSet, switching to a competitor like Bloomberg is a multi-month nightmare, so almost nobody leaves. That stickiness lets FactSet earn high returns on the money it invests. The risk is that AI tools may replace some of what FactSet does, slowly. The stock is unusually cheap today versus its history, which is why it is interesting.

Thesis

FactSet sells a workstation, data feeds, and analytics that buy-side and sell-side professionals use every workday. Customers are investment bankers, portfolio managers, wealth advisors, and corporate development teams. Once a research workflow, model, or content set is wired into FactSet, switching to Bloomberg, Refinitiv (LSEG Workspace), or S&P Capital IQ is painful: re-training, re-coding, re-permissioning, and migration of saved screens and Excel add-ins. The result is a 95%+ client retention rate and a 10-year average ROIC of 29% [TICKER FACTS]. Owner earnings of roughly $580M on TTM ending Feb-2026 are growing, the share count has shrunk modestly over a decade (-0.76%), and net debt sits at a comfortable 1.35x EBITDA with 11.7x interest coverage.

The market is pricing FDS at $227.58, a 16.15x trailing P/E versus a 10-year average of 33.07x, with a reverse-DCF implying owner earnings shrink at -1.38% forever. Against an IV-base of $522.52 (low $350.75 / high $662.49), the price-to-IV ratio is 0.44 — a standard Buffett-Munger margin of safety only if the moat actually holds. The legitimate worry is the 5-year ROIIC of 8.32%: management is reinvesting at returns far below historical, hinting that incremental dollars (acquisitions, AI build-out, sales growth) are not earning the franchise rate.

If FDS continues to grind out 5-7% revenue growth, sustains gross retention, and incremental capital eventually reverts toward 15-20% ROIIC as AI investment matures, the base case is a double from here over five years on multiple expansion alone. The price/IV math: paying $227 for $522 of conservatively appraised intrinsic value is the kind of setup Buffett describes when 'we'd rather own a wonderful business at a fair price.' Here we get a wonderful business at a meaningfully cheap price, conditioned on the moat narrative being intact.

Moat

FactSet operates in a four-firm financial-data oligopoly (Bloomberg, LSEG/Refinitiv, S&P Capital IQ, FactSet) with selective overlap from Morningstar, MSCI and Bloomberg-only datasets. Across the five moat types, FDS has unusually concentrated strength in switching costs and intangibles, with secondary strength in cost advantages.

Switching costs (DOMINANT). Damodaran's framing on Microsoft's success — that the most significant barrier to entry was 'the cost to the end-user of switching from one product to a competitor' [2] — applies almost perfectly to FactSet. A typical buy-side analyst has years of saved models, custom screens, Excel formulas, portfolio analytics templates, and PA (Portfolio Analysis) workflows hooked into FactSet codes. A bank's research department has compliance-approved workflows for distributing reports through FactSet. Switching to Bloomberg requires re-coding every BQL formula, re-training analysts (Bloomberg's command-line UX is notoriously idiosyncratic), and re-validating data lineage. FactSet's symbol-mapping dataset alone (mapping company A's tickers across exchanges, share classes, vendors) is a switching-cost moat in microcosm: once you've reconciled your portfolio system to FactSet's identifiers, ripping it out is a multi-month project. Reported ASV retention runs 95%+, and client retention runs 90%+ by count. Stress test: a $10B funded competitor over five years would still struggle, because the moat is workflow-embedded, not product-superior. Bloomberg has spent 40 years and untold billions and still only displaced FactSet on the trader side, never the analyst side.

Intangibles (STRONG). FactSet's curated content — fundamental data going back decades, analyst estimates, ownership data, M&A databases, private company data via Cobalt and IRN — is built up over years of manual standardization. Damodaran notes that brand and accumulated knowledge become moats when 'the most significant barrier to entry is the cost to the end-user' [2] and when firms 'use accumulated know-how... as a competitive advantage' [3]. FactSet's content team is its competitive R&D. Damodaran's caution that 'the companies that will see the greatest increases in value are not necessarily the companies that spend the most on R&D, but those who have the most productive R&D' [5] is the right frame: FactSet's content investment translates dollar-for-dollar into stickier client workflows.

Cost advantages (MODERATE). FactSet operates at scale (~$2B+ revenue) and has slowly migrated content production to lower-cost geographies (India, Philippines). The fixed-cost-leverage moat is real but weaker than Bloomberg's, because Bloomberg's terminal+chat+messaging bundle creates network effects FactSet cannot match.

Network effects (LIMITED). FactSet is mostly a one-to-many data distribution business. There is a soft network effect through PortfolioAnalytics where investment consultants and their managers share workflows, and via the Open:FactSet marketplace where third-party data vendors plug in. But these are pale shadows of Bloomberg's IB chat network — the single largest moat in financial data and the reason Bloomberg keeps trader seats at $30k+/year.

Pricing power (MODERATE). FactSet's 30-year history of low-single-digit annual price increases, taken without losing clients, is the empirical signature of pricing power. But it is constrained by Bloomberg's ceiling and by buy-side cost discipline since 2018 (MiFID II unbundling, fee compression). Damodaran's note that legal/regulated monopolies often face price control [2] does not apply directly, but the analog is buy-side cost-cutting committees who can refuse increases.

Erosion risk. The moat's vulnerability is generational, not quarterly. AI-native research tools (BamSEC, AlphaSense, Hebbia, Perplexity Finance) reduce the cost of doing what FactSet does for $30k/seat, and ChatGPT-style copilots inside Excel reduce the value of FactSet's Excel add-ins. Generative AI is the first technology in 30 years that can plausibly reduce the value of saved templates — because users can re-create them in seconds. The 5-year ROIIC of 8.32% versus a 10-year ROIC of 28.99% is exactly what you would see if management were reinvesting heavily into AI/LLM offerings (FactSet Mercury, Pitch Creator) at returns below the franchise rate to defend the moat.

Moat verdict: WIDE.

Management

FactSet's capital allocation under CEO Phil Snow (since 2015) and his successor as president has been textbook for a mature compounder: reinvest in content and product first, bolt-on M&A second, predictable buybacks third, growing dividend fourth, with debt used judiciously. Across the five capital-allocation choices:

Reinvest. FactSet's R&D and content spend has stepped up notably in 2022-2025 to fund the AI/LLM stack (FactSet Mercury, GenAI workflows), modernize the cloud architecture, and integrate the Cobalt private-markets and IRN deal-management acquisitions. The 5-year ROIIC of 8.32% is the most important single data point in this analysis: it tells us incremental dollars are not earning the 29% franchise rate. There are two readings. Charitable: AI build-out is a 3-5 year capex bulge that will normalize as products monetize; this is what management says on calls. Bearish: structural — markets are saturated, large clients negotiate harder, and the easy land-and-expand has been done. The truth is probably in between, but ROIIC of 8% is below the cost of equity, so on a marginal basis FDS is currently NOT compounding intrinsic value through reinvestment.

Acquire. FactSet has done many tuck-in deals (Cymba, BISAM, Truvalue Labs, CUSIP Global Services for $1.93B in 2022, IRN, Cobalt, LiquidityBook). The CUSIP deal is the largest and most controversial: at ~25x EBITDA and a target IRR around 10-12%, it stretched the multiple and added the only modestly defensible asset (CUSIP is essentially a regulated monopoly on US security identifiers, with revenue from issuance and licensing). The acquisition was financed with debt, taking gross leverage briefly to ~3.0x before deleveraging back to today's 1.35x net-debt-to-EBITDA. Acquisition discipline grade: B-. The CUSIP price was high but the asset is genuinely durable.

Debt. Net-debt-to-EBITDA of 1.35x and interest coverage of 11.72x is conservative for a recurring-revenue business. FactSet has historically targeted gross leverage of ~1.0-1.5x and has consistently deleveraged after deals. No debt-funded buybacks at peak prices — a common sin of mature franchises that FactSet has avoided.

Buybacks. The 10-year share-count change is -0.76% — basically flat. Management buys back enough shares to offset SBC plus a small net reduction. They do NOT buy back aggressively at low prices, which is a venial sin for a Buffett-style allocator. With the stock trading at 0.44x base IV, this is exactly the moment a Henry Singleton or Mark Leonard would lever up and shrink the float; FactSet won't, and that's a meaningful demerit.

Dividends. The dividend has grown 25 consecutive years (Dividend Aristocrat status) at low-double-digit CAGRs. Yield is modest (~1%) but the growth trajectory and the signal of long-duration discipline are intact.

Communication. Phil Snow's earnings calls are unusually candid for a SaaS CEO — explicit about churn, ASV softness, and AI investment ROI. The IR pages disclose Annual Subscription Value (ASV) and client/user counts directly, which is more transparency than Bloomberg or LSEG offer.

The scorer notes flag 'Maintenance capex uncertain (>50% spread)'. This is real: FactSet's content-creation labor is partly maintenance, partly growth, and management does not split it. The IV-base of $522 implicitly assumes maintenance capex is roughly 50% of total reinvestment; a more bearish split takes IV-low to $351.

Capital allocator: B+.

Industry

Financial data and analytics is one of the great industries in capitalism: high gross margins, high recurring revenue, high switching costs, regulated counterparties, and a customer base that pays out of operating budgets rather than discretionary spend. Porter's Five Forces:

Threat of new entrants (LOW, with one asterisk). Building a competitor to FactSet from scratch requires (a) decades of normalized fundamental data, (b) thousands of content analysts, (c) integrations with hundreds of execution and OMS systems, (d) regulatory licenses, (e) sales relationships at every major asset manager. The asterisk is AI: AlphaSense, Hebbia, Perplexity Finance, and OpenAI's enterprise products are not trying to rebuild FactSet — they are trying to make 'FactSet-style retrieval' commoditized via LLM context windows. They are not full substitutes today but the cost of asking 'what was Microsoft's free cash flow margin in 2019' has gone from $30k/seat to $20/month for a marginal user.

Bargaining power of buyers (MEDIUM-HIGH and rising). Buy-side cost discipline since 2018 has compressed seat counts at long-only managers, hedge funds, and wealth platforms. Large clients (BlackRock, Vanguard, Fidelity) negotiate enterprise agreements that effectively cap price increases at 1-3%. The 95% retention is the right number to look at, but ASV growth has slowed from 8-10% pre-2022 to 4-6% in recent quarters.

Bargaining power of suppliers (LOW). Exchange data fees and content licensing are the main supplier costs and have actually become more expensive (NYSE/NASDAQ data fees are a structural drag). But labor costs in offshore content production are stable to declining.

Threat of substitutes (MEDIUM and rising). Three substitution vectors: (1) Bloomberg Terminal at the high end (FactSet has historically taken share from Bloomberg in research workflows but Bloomberg keeps trading), (2) Capital IQ at the mid-market (especially in private equity and corporate development), and (3) AI-native tools (AlphaSense, Hebbia) for the long-tail use case of 'find me the answer in this filing.' Bloomberg/LSEG/SPGI are stable competitors who behave rationally on price. AI-native tools are the dangerous ones because they can plausibly disrupt the 'I need to ask one question once a week' user, who is a real chunk of FactSet's seat count.

Intra-industry rivalry (LOW-MEDIUM). The four-firm oligopoly has competed on features and content, not price. Net new logos and ASV come from market growth, not from share-shifting. This is the classic 'rational oligopoly' configuration.

Value pool location and trajectory: the pool is large (~$35B globally for financial data and analytics) and growing 5-7% annually, with the fastest-growing segments being wealth management technology (FactSet has a strong position via the Wealth solution), private markets data (Cobalt, IRN), and ESG/regulatory reporting. The biggest risk to the pool is not its size — it's reallocation: AI may shrink the value of 'general purpose terminal seat' and grow the value of 'agentic workflows on top of structured data.' FactSet must be the layer underneath the agents, not be the agent itself.

Industry Verdict: Good.

Inversion

I am now a short-seller. My job is to write the most credible bear case for FDS, not to hedge it.

The single event that kills this. A major customer (think a top-10 buy-side firm: BlackRock, Vanguard, Fidelity, State Street, JPMAM, Morgan Stanley IM, T. Rowe, Capital Group, Wellington, or PIMCO) publicly announces it is replacing FactSet seats with an internal LLM-based research stack built on top of OpenAI/Anthropic enterprise APIs and a curated data lake from cheaper providers (S&P Market Intelligence APIs, FRED, Refinitiv Tick History, vendor-direct exchange feeds). The savings would be $30-80M annually for a top-10 client, large enough to attract CFO attention. Once one publicly leaves, the playbook is leaked, and Goldman, Morgan Stanley, and the rest follow within 18 months. ASV would not collapse overnight — but the growth narrative would, and the multiple would compress to 12x P/E even if earnings are flat. That alone is a 25% drawdown from here.

Why the moat is narrower than bulls think. The bull narrative (95% retention, sticky workflows) is a coincident indicator, not a leading one. Retention measures decisions made years ago; cancellation lead times are 12-24 months for enterprise contracts. The leading indicator is ASV growth, which has decelerated from 8-10% (pre-2022) to 4-6% (recent quarters), and net new client logos, which are flat. The moat is not eroding because of a single competitor — it is being slowly hollowed out by every analyst who learns to do 'fundamentals lookup' with an LLM and a CSV. By the time retention drops, the franchise has already lost its growth premium. Switching costs are also asymmetric: the cost of leaving FactSet is high, but the cost of NOT renewing a marginal seat (the junior analyst who left, the team that consolidated) is zero. That's where the silent attrition is happening today.

Why management is worse than it appears. The 5-year ROIIC of 8.32% is below the cost of equity. Management is shoveling cash into AI features (Mercury, Pitch Creator, GenAI workflows) at returns that destroy value on the margin, because they have no choice — to NOT invest is to concede the AI surface to AlphaSense and Hebbia. CUSIP at $1.93B was overpriced even by FactSet's own pre-deal IRR target. The buyback program is anemic given the price/IV ratio of 0.44; if management really believed the IV-base of $522 they would lever up and buy back 15% of the float. The fact that they don't tells you what they actually believe about the IV.

What bulls are extrapolating that won't hold. Bulls extrapolate 7-9% revenue growth, low-double-digit EPS growth, 25-30% ROIC, and reversion to a 25-30x P/E. Three of those are wrong. (1) Revenue growth: ASV growth is structurally below 7% in the buy-side compression era. 4-5% is the new normal. (2) EPS growth: requires either margin expansion (unlikely in an AI capex cycle) or buybacks at high rates (not happening). (3) Multiple reversion to 25-30x: the 33x ten-year average was set in a zero-rate environment that no longer exists; the reverse-DCF implying -1.4% growth is the market's verdict that the regime has changed.

Valuation trap. P/E of 16x is not cheap for a business with 4-5% organic growth and 8% ROIIC. At those metrics, a fair multiple is closer to 12-14x. The 'discount to IV-base of $522' is meaningless if the IV-base is wrong: rebuild IV using 4% perpetual growth (not 6-7% bull case) and 10% discount rate, and IV-base falls to ~$340 — almost exactly the IV-low. Margin of safety evaporates. The scorer's own note ('maintenance capex uncertain >50% spread') is the tell that the IV-base is overstated. If we mark the maintenance capex assumption to a more pessimistic 60% (versus current ~50%), owner earnings drop ~15% and IV-base goes to ~$440. Combine with a 14x exit multiple instead of 20x, and IV is $200 — below current price. This is a value trap dressed as a Buffett-Munger compounder.

If I am right, the stock could be worth $150-180 within 3 years.

Lollapalooza Bias Check

Biases active in me as the analyst right now:

Authority bias. FactSet is in every Berkshire-style 'compounder' watchlist and shows up favorably in Buffett-Munger heuristics (high ROIC, low debt, recurring revenue, dividend Aristocrat). I am inclined to give it the benefit of the doubt because it pattern-matches to 'Moody's-style toll booth' [Buffett 2013, [1] in the canon]. But Moody's has explicit regulatory protection (NRSRO designation); FactSet does not. The pattern match is weaker than it looks.

Anchoring. The IV-base of $522 is anchored on a maintenance-capex split that the scorer itself flags as uncertain. The reverse-DCF implied growth of -1.38% is the OTHER anchor, and these two anchors are inconsistent. I should not split the difference; I should weight the reverse-DCF more heavily because it incorporates the market's collective view of the AI threat, which I cannot know better than the market.

Confirmation bias. I came into this analysis looking for a sticky, switching-cost compounder at a discount to IV. FactSet fits that template, so I am cherry-picking confirming evidence (95% retention, 29% ROIC, dividend track record) and underweighting disconfirming evidence (8% ROIIC, 4-5% ASV growth, $1.93B CUSIP deal at 25x EBITDA, anemic buybacks at 0.44x IV).

Recency. The 8.32% 5-year ROIIC is recent and may be a temporary AI-investment cycle, not a permanent regime. I should weigh both possibilities equally; I am inclined to charitably read it as cyclical because the bull case requires it.

Social proof. FactSet is widely owned in the 'high-quality compounder' value-investing community (multiple Buffett-Munger newsletters, value subreddits, and quality-focused funds hold it). That creates a reflex of 'smart people own this so it must be a good idea.' Most of those smart people bought at $300-450 and are now sitting on losses; their continued ownership is partly endowment effect, not a fresh thesis.

Deprival super-reaction. Looking at price-to-IV of 0.44 creates a mild fear-of-missing-out: 'this is the cheapest the franchise has been in a decade and you'll regret not buying.' That feeling is exactly what the market is using to clear inventory from holders who bought at $400. I should resist it.

Incentive bias does not apply directly (I am not paid to publish a buy). Commitment bias is mild — I have no prior public position on FDS in this report.

The lollapalooza vector here points UP toward optimism: authority + confirmation + social proof + deprival are all pushing me to call this a Buy. The corrective is the inversion section above and the explicit weighting of the reverse-DCF over the IV-base.

10-Year Outlook

Same fundamental business model in 2036? Probably yes, but with a different surface area. FactSet in 2036 will still be in the business of selling structured financial data and analytics workflows to buy-side and sell-side professionals. The terminal-as-pixels surface will shrink; the API/data-feed and agentic-workflow surface will grow. Customer base larger? Yes — wealth management and private markets are the growth tailwinds. Profit per customer higher? Uncertain. AI compresses the value of the 'long tail' user (the analyst who logs in twice a week) but may grow the value of the 'power' user (the PM running portfolio analytics at scale). Net is probably flat to modestly up.

Moat wider? Probably narrower. AI is the first technology in 30 years that can plausibly substitute for FactSet's core value proposition (structured data + saved workflows). The moat will not break, but it will compress: pricing power weakens, ASV growth stays in the 3-5% range structurally, and the 95% retention erodes to 90-92%. The franchise survives but does not expand.

Single biggest threat: a generational workflow shift away from terminal-style UIs toward agentic AI workflows that consume FactSet-equivalent data from cheaper or open sources. The historical analog is the slow death of Reuters' terminal business as Bloomberg's chat network became dominant — except now FactSet is in Reuters' position and the LLM is the new chat network.

Key assumption I would need to monitor: ASV growth, gross retention, and Mercury/GenAI product attach rates. If ASV growth stabilizes at 5%+ and retention holds at 94%+, the bull case is intact. If ASV drops below 4% sustainably or retention slips below 93%, the moat is compressing faster than expected and the analysis flips.

CONFIDENCE: medium

Position Guidance

• Recommendation: Buy
• Conviction: medium
• Target buy price: $220 (current price of $227.58 already meets the threshold; size into weakness)
• Target trim price: $480 (approaches IV-base of $522 with 8% buffer; trim aggressively above $550 which is 105% of IV-base)
• Position sizing: 2-4% starter position. Add to 4-6% if ASV growth re-accelerates above 6% or if price drops below $200. Do NOT make this a top-five position; the AI tail risk is real and the 8% ROIIC is a warning that may not be cyclical. Pair-trade hedging via SPGI (S&P Global) long or AlphaSense competitive intelligence is reasonable for a hedge-fund construction.