CRM

As of 2026-10-26 · partly unblinded

The answer. Salesforce is worth about $291 a share if AI agents shrink its customer-service seats slowly, prices per seat keep rising as customers move to premium editions, and its data and integration products keep growing about 8% a year. Frontier AI labs are a real threat, but to the interface and to seat-priced work, not yet to the customer record: in the world where the threat lands in full, which is our bear case, the stock is worth about $107, and most of that loss comes not from fewer seats but from agents reaching the data without paying Salesforce for the platform, and from the margin that goes with it.

Terms used throughout. The central case is our most likely path; the bear is the coherent world in which the AI threat lands; the bull is the world in which agents run through Salesforce. Seats are paid user licences. AI agent revenue is what Salesforce reports as Agentforce ARR (its agents, Slackbot, Headless 360), sold per user, per action or per resolution; we add Fin, the customer-service agent it bought in September. Headless means software reached by an agent through programming interfaces (APIs and MCP, an open standard for connecting AI agents to software) rather than by a person at a screen. Salesforce's fiscal year ends January 31: FY32 ends 2032-01-31.

1. Is Anthropic, and frontier AI generally, a threat to Salesforce?

Yes, and the forecast carries it as six drivers with ranges rather than a discount on the answer. What the evidence says, question by question:

Seats against agents doing the work. Salesforce is paid mostly per person: Sales Cloud ($9.0B in FY26) and Service Cloud ($9.8B) are seat products. An agent that resolves a support case removes the reason for a service seat. This is already visible inside Salesforce: its CEO said its support headcount fell from about 9,000 to about 5,000 as agents took about half of the cases (podcast, September 2025, as relayed by the press), and its own help agent resolves 64% of conversations without a person (Q2 FY27 call). Gartner forecasts that agents will resolve 80% of common service issues by 2029. Against that, Salesforce says Sales, Service and Slack seats all grew year on year in Q2 FY27 and attrition is near its lowest (CEO and CFO, 2026-08-26), and 87% of customers say a company must offer a human (Gartner survey, 2026-08-04). Our central has Service seats rising 1% in FY28, flat in FY29, then falling 2-3% a year (7% fewer by FY32); the bear has them falling to 64% of today. Price per seat offsets most of the central loss: list prices rose 9% (2023) and 6% (2025), and the September 2026 Core, Advanced and Max editions list 11-13% above the editions they replace, with Slack, agents and consumption credits bundled in.

Does the lab, or its enterprise partner, displace the system of record, or only the interface on top? So far only the interface, and Salesforce chose that itself. In Claudeforce (company release, 2026-08-26) Salesforce put its CRM inside Claude: a seller works in Claude, which reads Salesforce, Slack and other connectors; actions are routed back through Salesforce "to help ensure business rules are always enforced". The CEO's line was "the UI is the AI." Anthropic's S-1 (filed 2026-10-26) lists Salesforce among the data platforms through which customers adopt Claude, says Salesforce has built products on Claude that reach millions of users, and says Anthropic "may from time to time compete with enterprise software and application companies in select categories" (p. 237) and competes, and may increasingly compete, with its own software-as-a-service customers (p. 56). Its products for knowledge work (Claude Cowork, Claude for Financial Services) do work on top of records; the filing describes no plan to hold a customer database. OpenAI's Frontier platform (February 2026, press) competes for the same orchestration layer. The record itself is hard to move: it carries twenty years of a customer's data model, permissions, audit and integrations. What can move is who the customer pays for the work done on it. That is the bear.

Agentforce's real monetisation. No reported revenue line isolates it. Agentforce ARR, a management metric, passed $0.5B in Q3 FY26, $1.2B in Q1 FY27 and $1.5B in Q2 FY27, after Salesforce widened it to include Slackbot and Headless 360, so its growth rate is not like for like. We put FY27 revenue in that scope at $1.3B. Central: $6.0B by FY32. Bear: $2.5B, because the agent spend goes to the labs and their partners. Bull: $9.0B. We give it a 25% operating margin in the central (15% bear, 35% bull) because agent revenue carries model costs that seat revenue does not. The CFO said on 2026-08-26 that headless monetisation is still being worked through; MCP and command-line calls into Salesforce rose sixfold in that quarter. Calls are not revenue yet.

Customer-service seat compression, in dollars. One driver at a time, it is smaller than the debate suggests: three more points a year of Service seat decline costs $9 a share; two points less price growth $12; AI agent revenue of $2.5B instead of $6B $10. Where the threat lands in full, the walk from the central $291 to the bear $107 (threat_bridge.txt) is:

Seats and price together are $47 of the $184. The larger losses are the data platform ($39), where MuleSoft, Tableau, Informatica and Data 360 are exactly what an agent with direct data access does not need, and the margin ($42) of a seat business that shrinks while its cost base does not. So the items to watch are data-group growth and the margin, not the headline seat count.

2. The business: who pays for what

Customers rent Salesforce by the seat for one to three years, billed in advance, so growth releases cash (unearned revenue less receivables and contract costs was 11.8% of FY26 revenue; 10-K). FY26 revenue $41.5B. From FY27 Salesforce reports two groups: Agentforce Apps (Sales, Service, Marketing, Commerce, Slack; $14.1B in H1 FY27, +8% in constant currency in Q2) and Data 360, Headless Platform and Other (Data 360, Informatica, MuleSoft, Tableau and the platform; $7.3B, +20% in Q2 with Informatica, about +6% without it in Q2 and +10% in Q1). Marketing and Commerce were flat to down in FY26 (CFO: "early signs of recovery"). Slack had its fastest new bookings growth since the acquisition in Q2 FY27.

Following the money. A $25B accelerated share buyback in March 2026 was paid for with $25B of new notes; debt is $39.5B against $11.4B of cash and securities at 2026-07-31 (10-Q). Since then Salesforce has paid about $5.1B for Contentful and Fin and agreed to pay $5.66B in cash for Braze (8-K 2026-10-20; close in FY28). Cash acquisitions over FY17-FY26 were $37B, 15% of the decade's revenue, and the after-tax return on the capital it added in each five-year window, goodwill included, was 0.6% to 12.1% (median 5.8%; output.txt): Salesforce has bought growth at or below its cost of capital. Strategic investments carry $11.3B, $5.1B of it Anthropic, marked only on observable transactions; a listing above that mark would add to value (each $1B after tax is about $1.25 a share). Stock pay is $3.5B a year (9.0% of revenue guided for FY27) and is charged as a cost.

3. Results against the bar

Salesforce guides close to its results: revenue beat the guide midpoint in six of the last eight quarters, by a median 0.51% and never more than 1.17% (output.txt). Q2 FY27 revenue was $11,345M against an $11.27-11.35B guide; cRPO (contracted revenue for the next twelve months) $33.5B, +14% in constant currency, a point above the guide, about 10% without Informatica. The FY27 guide is $46.1-46.4B. Its long-range targets have run high: "$50 billion in revenue in FY26" (2022) became $41.5B. The FY30 target is $63B+ including Informatica and deals announced by May 2026 (Investor Day, 2026-09-16), before Braze.

4. Our forecast and value

Model: model.py, output in output.txt; the exit multiple in exit_choices.py.

FY28 organic growth is 7.7%, close to today's 7-8%; Braze and further acquisitions (3.5% of revenue a year) add the rest. Sales plus Service grow 19% by FY32, nearly all of it price. Our FY30 $59.1B is 6% below management's $63B+ even with Braze. The GAAP margin rises because the core non-GAAP margin goes from 34.3% to 37.5% while amortization of acquired intangibles falls; stock pay stays at 9.0% and restructuring at 1.2%. Braze carries its own margin, not Salesforce's: 8.3% non-GAAP in FY28 (its FY27 guide, release 2026-09-08) rising to 20% by FY32 (our assumption), with stock pay at its Q2 FY27 15.7% of revenue; that holds the total non-GAAP margin to 36.0% in FY32.

Value. The five-year exit is the mean of two views. The long fade carries FY32 profit fifteen more years, fading growth to 3% and the return on new capital from the forecast's own 65% down to a mature 9.28%, the cost of capital. The method's rule holds each case at the return on its own FY32 capital (16.3% in the central, with goodwill counted because the growth is bought), never above the company's median five-year return and never below the cost of capital. Salesforce's median, 5.8%, is below the cost of capital, so the floor sets it. Growth after FY32 still adds value while the return is above the cost of capital and adds none once it reaches it (FY33 growth of 0% instead of 7% would cost $21 a share). After debt, the fade gives $201 a share. The outside view prices FY32 operating profit at 30.9x, which gives $561. The two differ by 179%, and the input that drives it is the mature return: the fade says Salesforce's acquired growth will in the end earn no more than its cost, the outside view prices the franchise as peers were priced. We hold both, and the mean is $381. With five years of owner cash, after interest, paid out and discounted at 10.14%, it is worth $291 today.

The exit multiple and the judgment range. The method's default, anchor (d), is the median of six peers' own ten-year medians, 34.4x (Adobe 41.2x, Oracle 20.9x, SAP 23.7x, Intuit 41.6x, Autodesk 49.3x, Microsoft 27.6x); Salesforce's own history has too few years with meaningful GAAP profit to count. Those peers earned their multiples at a median 13.9% revenue growth; our central grows about 7% at the exit. We do not take Salesforce's own multiples from its faster years, and no shared growth adjustment predicts better out of sample. What we use: software companies whose peers' multiple was earned at 10% growth or more and which grew 5-10% five years later fetched 0.891x (1996-2009) and 0.909x (2010-2021) of what all companies of their own era fetched. 34.4x × 0.900 = 30.9x. The support is weak, weaker than in our Veeva revision: the sample leaves out companies taken over or failed, it requires each company's own ten-year history, which Salesforce lacks, and its ratio is measured against anchors that include the company's own leg, while Salesforce's anchor is the peers' alone. The range shows what else is defensible:

The judgment range is $237-312. Below it sits only the fade alone. The lower rows are defensible: the anchor's six peers ended their latest fiscal years at a median 21.8x, and Salesforce itself at 23.7x (FY26). They are single year-ends, which the method's ten-year window was built to smooth; the 24.2x row rests on a cross-section of companies of all qualities, and our Veeva work found cross-sections predict realised multiples worse than the unmoved anchor. The call depends on this choice: every row at 24.2x or below is a Hold.

Cases.

Case-weighted (25/50/25, our weights) $316, above the central because the bull is further away than the bear.

Cost of equity 10.14% = 5.29% risk-free + 1.155 beta × 4.20% premium (the method's dated Damodaran pair, 2026-10-01). The beta is the Blume-adjusted median of Adobe, Oracle, SAP, Intuit, Autodesk and Microsoft; Salesforce's own 1.25 gives $283. Owner cash is paid out after after-tax interest at the debt's 4.62% weighted coupon ($1.82B a year before tax), and the $39.5B of debt and $0.6B of finance leases are deducted from the exit, held flat and refinanced; cash is counted at face today, less the acquisitions already paid and 3% of revenue for operations; strategic investments at carrying value less tax on their gain. The exit is then an equity value, discounted at the cost of equity like the payouts. Shares 797.8M: 823M on 2026-08-20 less our estimate of the buyback's final delivery (26.2M, from the "≥14%" reduction management expects) plus 1M dilutive awards.

Sensitivity of the central (output.txt, review_answers.txt): core margin about ±$12 a point; data-group net retention 99% / 111% -$31 / +$26; stock pay 8% / 10% +$13 / -$13; mature return at the central's own held 16.3% +$7, at 20% +$9; acquisitions at 6% of revenue -$11; growth after FY32 of 4% -$10, of 0% -$21; FY32 amortization of past acquisitions at $0 or the whole $1,354M the 10-Q puts after FY31 +$6 / -$15 (we use $400M, our assumption); a debt rate one point higher in the cost of capital -$1; each $1B of revenue above our path held as a level +$7.0.

Odds ranges, changed after the price was seen. The odds tool refused three driver ranges whose central sat too near their top (71-74% of the way from bottom to top; it allows 29-71%). We kept every central and widened the top only: Service seats to +3 points a year (from +2.5) and Sales seats to +2.5 (from +2), both within the seat growth FY23-FY25 revenue implies after list-price rises, and FY33 growth to 10% (from 9%), the bull's own value. The acquisitions solve range now starts at 1.5% (from 0%), because below about 1.2% of revenue the shared method cannot value the resulting harvest. No value moved; the odds are run on the widened ranges. The model auditor notes that the refused ranges cannot themselves be simulated, so there is no valid before; the advocate capped the published draws at the old tops and P moved from 54.0% to 53.7% (on the pre-review central). On the final model the chance of a value at or below the bear ($107) is 3.9%, below the 5-25% the odds tool expects of a one-in-five-to-ten case: our ranges may be too narrow on the downside, which the advocate's combined world below also suggests.

5. The market prices our forecast at a lower exit multiple

This section was written after the price, targets and estimates were read (unblinded). Sections 1-4 and the model changed after it only through the review answers in section 10.

At $247.55 our central case earns 14.2% a year to the exit, above the 10.14% cost of equity. The price equals our forecast at a 23.5x exit multiple, between the 22.3x and 24.2x rows of the table in section 4: the market pays for about our forecast at the multiple software companies growing 5-10% fetched in 2016-2025, not the 30.9x we take from the peers' anchor. The bear's multiple alone on our central forecast gives $235 (section 6), so the price is close to the de-rating path.

The Street's revenue path is ours to FY29: $46.3B, $50.9B and $55.7B for FY27-FY29 against our $46.4B, $51.0B and $55.2B. It then runs ahead: $61.4B in FY30 (16 analysts; ours $59.1B, 4% lower) and $71.5B in FY31 (17 analysts; ours $63.1B, 12% lower), close to management's $63B+ FY30 target carried on. So the gap is not near-term growth; the Street is more optimistic than we are on the later years and the market pays less than we do for the exit. Its average 12-month target, $274.18, discounted a year at our cost of equity is $248.94, about the price; our value is 1.17 times it. The targets are falling: the one set in the last month is $250, against $272.13 over the year. The Q3 FY27 print (expected 2026-12-02) settles part of it: data-group organic growth above 10% supports the bull's retention, below 5% the bear's.

Alpha's view, 2026-10-26. Alpha, the quantitative model, scores Salesforce 86 of 100 (tier Strong), with expectations A, momentum B, valuation B, growth C and quality D; it gives no rank today and flags its data as limited, because goodwill and intangibles rose with acquisitions and organic growth is unverified (alpha ticker analyze CRM). Its top drivers are earnings surprises, the six-month price move and the free-cash-flow yield; its top drag is the return on capital. We agree on the return on capital: our fade gives acquired growth no value after FY32 for the same reason. We differ on the surprises: Alpha reads them as momentum, we read a company that guides close to its results (a median 0.51% beat) and whose growth is increasingly bought.

The chart, as timing only. A technical quick read on 2026-10-26 grades the chart B: an advancing trend after the 2026-08-27 results gap ($206 to $252), above a rising 30-week average ($196) and the 50-day ($235), in a range of $220-269 since; the October low ($220.29, 2026-10-08) held the September low ($221.18), and the shares rose 43% over three months against 3% for QQQ. It is 1.6 average daily ranges above the 21-day average; the range top, $264-269, is the resistance. Results are about 26 sessions away (2026-12-02, not confirmed by the company). It falls to C on a close below $220. The grade says when, never what Salesforce is worth, and it moved no number here.

6. Why the other side could be right

For a lower value. The bear is a coherent world, not a stress test: Claude, ChatGPT and Gemini become where work is done (Salesforce itself has put its CRM inside Claude); customers buy agents from the labs and their partners, priced per outcome; Service seats fall by a third; Salesforce sells access to the record per call at prices it has not yet set (CFO, 2026-08-26); MuleSoft, Tableau and Informatica lose to agents that read data directly; and the margin falls because the cost base was built for a seat business. Each part has a dated sign today; none has shown in reported revenue. Salesforce's record on acquisitions (a 5.8% median return on added capital, below its cost of capital) also argues that the buying that props up growth in our central destroys value; our fade already gives that growth no value after FY32, but the five forecast years still pay for it.

De-rating path. If our central forecast holds but buyers in 2032 pay the bear's 21.5x for it, as they might for a mature seat business under threat, the stock is worth $235 today (threat_bridge.txt). What would keep the multiple there: Service and Sales seat counts falling in Salesforce's own disclosures, headless access staying unpriced, and data-group growth below 5%.

For a higher value. Seats are growing today and attrition is near a low; premium-edition bookings more than doubled quarter on quarter (Q2 FY27); nine of the ten largest AI companies run on Salesforce and Slack (Investor Day, a management claim); management's FY30 plan implies about 11% growth. The Anthropic stake ($5.1B carrying) could be marked well above that at a listing. If the bull lands, the stock is worth $572.

7. Next results preview: Q3 FY27

Salesforce had not announced the date on 2026-10-26 (its investor-relations event and news feeds); Q3 FY26 was released 2025-12-03. We expect early December 2026.

Each $1B of annual revenue above our path, held, is worth about $7.0 a share. The print that would move us: data-group growth without Informatica (about 8% in H1) and any disclosed seat count or headless pricing.

8. Catalysts and falsifiers

• Q3 FY27 results, early December 2026 (date not announced). Data-group organic growth below 5% points to the bear's retention; above 10% to the bull's.
• Headless and Claudeforce pricing. A per-call or capacity price for agent access (the Max edition's promised Headless 360 allocation) is the first evidence of whether the record is paid for when the interface moves.
• FY28 guidance, late February 2027. Organic growth guided at 7.5% or less, or a non-GAAP margin below 34%, would lower the central's growth and margin drivers (each point of FY32 core margin is about $12 a share).
• Braze close (FY28) and any further acquisition above $5B.
• Anthropic's listing. A mark on Salesforce's stake.
• Service seat disclosures. Any statement that Service seats fell year on year.

9. Sources, dated

• Salesforce 10-K FY26, 10-Qs for Q1 and Q2 FY27, quarterly releases Q3 FY25 to Q2 FY27, Investor Day deck (8-K 2026-09-16), Braze 8-K and investor webinar (2026-10-20), all SEC EDGAR, masked for price terms; saved in data/.
• Q1 and Q2 FY27 earnings call transcripts (TradingView/Quartr, captured 2026-10-15; secondhand transcripts).
• Salesforce releases: Claudeforce (2026-08-26), new editions (2026-09-03), pricing updates (2023, 2025); IR event and news feeds read 2026-10-26 (data/web-captures.md).
• Anthropic, PBC, Form S-1, 2026-10-26, pp. 10, 56, 182, 193, 237 (../2026-10-26-anthropic-s1/data/s1p.txt).
• Gartner press releases 2025-03-05 and 2026-08-04 (summaries); Fortune 2026-02-10, VentureBeat and Dealroom on the labs' enterprise moves (press); press reports of the CEO's support-headcount remarks (September 2025).
• Market data annual statements for Salesforce, its peers and Braze; fiscal-year-end EV/EBIT history (data/exit-anchor.json); the survivor-free exit-anchor study's software starts (data/exit-outcomes-rows.json).
• Damodaran implied equity risk premium and the 5.29% risk-free rate it is set against, 2026-10-01 (the method's dated pair).

This work was done with Claude, a model made by Anthropic, a company this report assesses as a competitor and partner of Salesforce, and in which Salesforce holds a stake: a conflict of interest.

10. What the two reviewers found, and our answers

Two outside reviewers read the committed draft with the market in hand: a model auditor (Codex) and an opposite-side advocate (Opus), who argued for Avoid. Every figure below is rerun by review_answers.py (review_answers.txt) on the final model unless marked. Net effect on the central: $287 to $291 (+$4.8): +$12.15 from the debt basis, -$7.34 from Braze's margin. The call stays Buy.

Flags the automatic checks raised on the final model, answered: flag 4 (low), the outside view 179% above the fade, is driven by the mature return (section 4). Flag 23 (high): the odds' 90th percentile ($560) is 2% below the bull ($572); the simulation reaches the bull in 9.4% of draws, so it can reach our optimistic case, but our ranges are narrow at both ends (3.9% at or below the bear). We tried widening the data-group retention and core margin ranges both ways from history; the tails barely moved (90th $559), because the exit-price draw and the worlds set them, so the ranges were left as registered.

What held up, in the advocate's words: the seats and Agentforce treatment, and the widened ranges (P 54.0% to 53.7%). The auditor confirmed all four scripts, both generated JSON files, the odds and the 62 data pins reproduce exactly, and that acquisition charges, Informatica, debt, stock pay, tax and shares reconcile.

Revision history

• 2026-10-15, $239 (pessimistic $122, optimistic $331). First report: guidance, then 8.5% organic growth fading to 7%, valued at a judged 16.5x FY33 NOPAT set equal to its own fade. Never restated: its only reinvestment was working capital, which the shared method cannot fade.
• 2026-10-26, $291 (+$52). Mostly a method change, then judgment, new data and two review fixes. How the $239 becomes $291:
  • Method (shared rules). The exit is now the mean of the long fade and an outside view (anchor (d) rules of 2026-10-16 and 2026-10-18, case spread of 2026-10-24) instead of a judged multiple equal to the fade. On this report's own forecast the fade alone gives $183, the old report's basis; adding the outside view at 30.9x gives $291. The outside view is the main source of the rise, about +$108 on our forecast. The mature-return rule H (2026-10-26) and its fix the same day (the cost-of-capital floor wins over a lower history cap) set the fade's return at 9.28% against the old 20%; inside the fade alone that lowers value.
  • Judgment. The forecast is rebuilt line by line with the AI threat as drivers. The central path ends close to the old one (FY32 revenue $67.7B against $68.3B), with a lower GAAP margin (23.8% after amortization against the old 27.7% before it) and acquisitions charged as reinvestment (3.5% of revenue a year, none in the old model). The old model's fade-based value was $239; this model's fade alone is $183. That $56 gap mixes the forecast, the acquisition charge and the fade's mature return (9.28% against 20%) and is not split here.
  • Data. Braze ($5.66B, FY28), Fin closed, Claudeforce and the new editions, the 9.0% stock-pay guide carried forward.
  • Error fixed after review, made after the price was seen (section 10): debt and interest put on the same equity basis as the payouts (+$12.15); Braze at its own margin from its 2026-09-08 release (-$7.34). The blind value was $277; the rule H fix took it to $287 before the market pass.
  • Not separated: the two models differ in structure (owner earnings before amortization against GAAP after it; payouts at year end against mid-year), so the steps above are an ordered reading, not a runnable bridge; the residual is not allocated.
  • Exposure. The table above, with the 2026-10-15 price, was read after this report's model was committed and pushed (commit 9b3cec05e); the old report's headings, read before, showed its $239 and that the price was then close to it. This revision is partly unblinded.