Full thesis · NYSE: FDS

FactSet Research Systems (NYSE: FDS) – Long

Adam Patrick Weir · 30 September 2026 · Original note (PDF) · Scored in the Q4 update

The Record on This Idea

• 05 Oct 2026 Q4 print: the pre-registered scorecard, graded Three positive, five neutral, none negative. Second half added per plan. Update
• 30 Sep 2026 Thesis published. Long at $266.18, blended target $350. This page. Kill-switch status: 0 of 6 triggered. Thesis

Every scorecard, kill-switch check and exit on FDS lands in this list, newest first, from publication to resolution.

1. Investment Thesis

FactSet trades at roughly half of its own five-year average multiple because the market believes AI-driven cuts to junior analyst headcount will decay its subscription base. At $266, a reverse DCF implies the market is pricing perpetual free-cash-flow growth of under 2%, which prices the business for permanent stagnation. That price is dependent on seat cuts actually transmitting into lower FactSet revenue – I believe they largely cannot. If the market's belief fails, the position re-rates: while it holds, per-share value still compounds at around 11% a year without the multiple moving.

The bear case argues that the reduction in junior seats across finance (particularly banking and research) will lead to a revenue reduction – but this relies on revenue being priced per seat to hold true. For most of FactSet's book, it is not. Market Infrastructure (17% of ASV, including CUSIP) is licensed per firm and system; Wealth (16%) is platform-priced on advisor headcount that is growing rapidly; the institutional buy-side (48%) is predominantly priced at the enterprise or AUM level (although to which degree is debated); only Dealmakers (19%) bills per analyst, and it is the channel my base case haircuts the hardest as a result (Exhibit 2). Management puts seat-linked revenue below 20% (although this claim is unaudited and contested – Redburn estimates approximately 75% is terminal-derived); ultimately, the market is pricing a far stronger transmission mechanism than the evidence supports. The 10 November Investor Day is the first opportunity for management to quantify the current disputed portion. Monetisation, meanwhile, is shifting from seat to the workflow itself: an AI agent does not require a terminal licence, but it does need entitled data – more than 20% of FactSet's top 100 clients now pay for its agent-facing data product. Organic ASV growth has accelerated for five straight quarters (4.1% to 7.1%) and held at 7.0% in Q4, supported in cash by deferred revenue running ahead of revenue.

The street models the FY26 margin trough of 34.5% as permanent through FY28, yet the cost programme behind it is documented: 27% of committed code is now authored by AI, the technology workforce has been reduced 10%, more than half of the targeted 100bps of annual productivity was captured by Q2, and a Google Cloud agreement caps the AI input cost through preferential token pricing. These numbers come straight from management and are corroborated independently: compulsory layoff filings disclose nothing other than the 10% technology staffing cut, and employee reviews complain most in the units being cut (Appendix A, tests 11 and 15). Compensation is roughly 60% of the cost base, so the market is pricing AI solely as a threat to FactSet without acknowledging that AI is reducing the largest cost. My margin path reaches 37.5% by FY31, which remains below the FY24 peak of 37.8%. The initial FY27 guide (34.75-35.25%) has since confirmed the trough, with management calling 34.5% a floor.

I also disagree with the street's buyback estimate of $333M. Their estimate is simply the unspent residual of the current authorisation ($356.0M remaining on 31 August). For the street to be right, the board would have to stop repurchasing entirely once the current authorisation runs out – nothing in FactSet's track record supports this. The firm has repurchased shares in every one of the last ten years (pausing only to repay the 2022 CGS deal debt), and the current programme – raised 2.5× to $1B in December – repurchased $644M in fiscal 2026, more than double the prior year. At that pace, roughly 5% of shares are retired each year at around 14× earnings.

If the market refuses to concede the seat question, margin recovery and repurchases still compound EPS at approximately 12% per year. What the thesis cannot survive is retention breaking, which the exit conditions laid out in Section 6 watch. Revenue throughout is modelled at consensus – the thesis is not dependent on beating revenue forecasts.

2. The Business, the Moat, and Management

FactSet is a financial data and analytics platform with $2.5B of annual subscription value across four main client channels (reported as a single segment). The moat, created by deep workflow integration, licensed data entitlements, and the retraining cost of moving users off the platform, is exceptionally strong – proven by the fact that ASV retention held above 95% through a −61% drawdown and an industry-wide AI panic.

Management's EPS and revenue guidance has been within a percent of actuals for three consecutive years, and capital allocation has been disciplined (with the one notable exception of CUSIP Global Services, bought for $1.9B at the 2022 peak and now the source of an antitrust class action from the customers who pay its licensing fees – Section 6). The stock is currently cheap following a collapse from $480 to $185 from December 2024 to February 2026, caused by a gradual de-rating through 2025 as AI fears built, then two sharp legs down – the September 2025 'invest' guidance reset and the February 2026 AI-agent panic. The register shows that growth funds sold during the decline and value institutions arrived in spring 2026, and short interest has doubled, too: the shareholder base is turning over at the bottom (Appendix D).

3. The Bear Case

The strongest version of the bear case has three parts. Firstly, the Dealmakers segment is genuine per-seat exposure at 19% of ASV, and its current 9% growth is partly a deal-cycle recovery rather than a durable growth rate. Secondly, ASV lags contracts as any seat cuts will only appear when multi-year enterprise deals are due for renewal (mostly between 2027 and 2029), so five quarters of accelerating growth cannot alone refute the bear case. Lastly, revenue per user is falling by roughly 4-5% (users +12%, ASV +7.1%), albeit this simply reflects the wealth-seat increase, and management is on record that, so far, extended contracts are carrying no price compression.

There are two ways in which this bear case can be challenged. The first is to stress test the damage the scenario implies and compare that against what the current price assumes already. If seat cuts remove a third of the dealmaker book by 2029 during the renewal cycles, total ASV growth drops 2% per year – this is approximately the rate currently priced in at $266, meaning even the bear case is roughly what I am paying for today. Furthermore, the dispute over the percentage of ASV that is seat-sensitive is limited: while management, which estimates it at sub-20%, naturally has an incentive to understate, Redburn's 75% estimate would require a third of ASV (Market Infrastructure and Wealth) to be priced per seat when it is not, and a bank reducing junior seats by a third to cancel a third of an enterprise contract. Exhibit 2b runs the range between these bounds and finds that the realistic middle produces the bear case of $240 rather than a broken thesis (I should note that if that scenario were developing, the kill-switches in Appendix C are designed to catch it early). This exact experiment also ran before during the bank failures and consolidations of fiscal 2009. While FactSet lost 2,800 users (7% of its seat base), ASV finished the year higher ($615M to $619M), retention held above 95%, and organic growth was back to 7% by fiscal 2010.

The second way to challenge the bear case is on its timeline. The bear's mechanism is most likely to rear its head during the 2027–29 renewals, meaning each period that passes with ASV holding above 6% narrows the window in which the thesis can come true – by late 2027, the argument that the damage has not yet arrived will no longer hold for either side. It is also worth noting that FactSet is not an outlier: the five subscription vendors serving the same client base (MSCI, LSEG, Moody's, Morningstar, S&P Global) are all accelerating, and none show any sign of the behaviour the bear fears.

4. Earning Power and Valuation

When I discount the street's own forecasts at the same 9% WACC and 3% terminal growth as my DCF, I arrive at $347 per share against a price of $266. This shows the valuation gap between me and the street does not come from estimates: my revenue matches consensus (FY27: $2,618M vs $2,615M), my GAAP EPS is within two cents, and the only real variance is margin and share count (Exhibit 3). Instead, for $266 to be right once the street's five-year path is credited, the market must be running terminal growth of roughly 0.8%, or a 10.6% discount rate, on a subscription book with 95% retention. This is the seat fear translated into discounting. Ultimately, margin and share count add roughly $33 of value on top of the $347; the remaining ~$81 of the gap comes from the discounting itself.

Blending four methods (Exhibit 4) gives a target of $350, 32% above the current price and equivalent to 17.5× FY27 earnings – still roughly half the stock's own five-year average multiple. The methods land close together ($315–372), and equal-weighting them gives $343, so the weights are not driving the answer. Against this, the bear case is worth $240 (−10%).

5. Catalysts

The multiple will only move when the seat worries are resolved – I believe three things can do so: the 10 November Investor Day, where management can quantify seat-linked vs consumption revenue, and collapse the bear case by doing so; the 2027-2029 renewal clock, which forces the bear's mechanism to either appear on schedule or expire; and AI revenue becoming a disclosed line, converting the narrative from a key business threat to a revenue driver.

As I have mentioned, even before this there are three dated events that pay the position regardless. The 30 September print brought the initial FY27 margin guide in at a 35.0% midpoint, above the street's flat 34.5% (the print is scored against my pre-registered expectations in Appendix B). A new buyback authorisation – none yet announced, with $356M remaining – would mechanically retire the street's $333M assumption. Finally, the court's redacted class-certification opinion – the ruling on whether the CUSIP plaintiffs can sue as a group – publishes around 2 October.

The entry was structured around the 30 September print: options implied a ±10% move, so the plan was half a position before it, the second half conditional on an in-line-or-better FY27 margin guide – a bar the guide met. Finally, at 11× EBITDA with 95% retention, FactSet is a plausible target for strategic or private-equity buyers. The market prices this possibility at zero, so it is effectively a free option for the holder.

6. Risks

Each of the four main risks is dated or has a pre-set response. Seat-linkage disclosure may disappoint at the Investor Day, in which case resolution waits for the 2027–29 renewals. An adverse CUSIP ruling, or an injunctive change to the licensing model, ends the position. If rates stay high, multiples on long-duration assets may stay compressed regardless of fundamentals – the position still compounds at around 11% a year while I wait. And there has been no CEO open-market purchase, although officer buys in autumn 2025 at $253–296 are on the record. Exit discipline is set in advance: any two of the six kill-switch conditions in Appendix C (for example, ASV below 5% for two consecutive quarters, or retention below 95%) and the position is closed regardless of price. The switches fire with a lag, so a disciplined exit realistically costs 15–20% from entry; the half-position start is sized against exactly that loss. The credit market, meanwhile, sees none of this risk: ratings are stable at BBB+/Baa3, spreads have not moved, and the banks upsized the revolver in August – the bond market is not pricing the disruption story the equity market is (ownership and balance-sheet detail in Appendix D).

Appendices: A: Evidence ledger (15 pre-registered tests) · B: 30 Sep scorecard · C: Kill-switches and catalyst calendar · D: Positioning and credit · E: Supplementary exhibits.

Appendix A: Evidence Ledger – 15 Pre-Registered Tests

Every test pre-registered with bull/bear predictions before looking; nulls and warts reported. None fatal; one contradiction (no CEO open-market buy) disclosed.

Appendix B: The 30 September Scorecard – Pre-Registered 27 Sep

Options imply a 9.7% move (~1.8× normal). Entry plan: half position before the print, half after an in-line-or-better FY27 margin guide; no add on a scorecard-negative print.

Scored 30 Sep: revenue (neutral-high), ASV add ~$82M (positive), adj EPS $4.52 (positive), ASV guide (neutral), margin guide (bar met), EPS guide (neutral), authorisation silent (neutral), retention/AI (positive) – three positive, five neutral, none negative; second half added per plan.

Appendix C: Kill-Switches and Catalyst Calendar

Facts, not price, end this position: any two falsifiers → exit regardless of price. Price weakness with fundamentals intact = add.

Falsification Conditions (4–8 Quarters)

1. Organic ASV growth < 5% for two consecutive quarters
2. ASV retention drops below 95% - or the disclosure quietly disappears
3. Dealmakers ASV turns negative, or management cites bank seat reductions
4. FY27/28 margin guide < 34.5% without ASV acceleration (the 'permanent investment' trap)
5. CUSIP: adverse ruling, settlement > $300M, or injunctive change to licensing economics
6. Wealth growth < 7% (penetration stall kills the offset engine)

Appendix D: Who Is on Each Side, and Can the Balance Sheet Wait

Appendix E: Supplementary Exhibits