Below are Dolphin Research's notes from Salesforce's FY27 Q2 earnings call. Compiled by Dolphin Research.

I. Core takeaways from the print. Key highlights follow.

1) Shareholder returns: the accelerated $25 bn buyback program remains in force, with management estimating repurchases to date equal to at least 14% of shares outstanding. The Avg. cost is $176 per share.

2) FY27 revenue guide raised by $300 mn at constant FX, but the mix matters.a) The organic lift is only $100 mn, driven by momentum in Agentforce, Data 360 and Slack, offsetting ongoing volatility in license revenue.b) The remaining $200 mn reflects the planned consolidation of Contentful and Fin, with closings targeted in the coming weeks.c) Versus the prior guide, an additional $100 mn FX headwind is now assumed.d) Updated FY27 revenue is $46.1–46.4 bn, with subscription and support up slightly above 12% YoY and constant FX growth just under 12%.

3) Q3 guide and FY margin framework.a) Q3 revenue of $11.42–11.50 bn, implying ~11–12% YoY at constant FX; Q3 CRPO growth of ~14% at constant FX.b) The Q3 guide excludes any incremental contribution from Contentful and Fin, as management will update after close for balance accuracy.c) Full-year Non-GAAP OPM held at ~34.3%, GAAP OPM updated to ~20.1%; operating cash flow and FCF growth still guided to ~4–5% YoY.

4) This quarter's performance and cadence ahead.a) CRPO reached $33.5 bn, +14% at constant FX and 1 pt above guidance, attributed to Slack, Agentforce, Data 360 and near record-low churn.b) Non-GAAP OPM was 34.1% and GAAP OPM 20.5%; FCF was $1.1 bn, up 81% YoY.c) The FY28 financial framework will be provided at the Dreamforce investor day in Sep; IR lead shifts from Mike Spencer to Mark Murphy.

II. Details from the call. Expanded commentary below.

2.1 Management remarks: core updates. Key product layers and go-to-market.

1) Claudeforce and AI Force (a new interface layer).a) Salesforce and Anthropic introduced Claudeforce, placing Claude's reasoning atop Salesforce data, apps, workflows and the permissions model; this quarter it opened internally to operating unit heads.b) Underpinning it is 'AI Force', a new enterprise AI harness debuting at Dreamforce, which also powers the new Slack and Coworker within Lightning.c) The platform is decomposed into four layers: data, apps and semantics, agents, and interface; models are probabilistic, while these layers are deterministic, and value comes from their combination.d) Zero data retention is enforced and customer data does not feed models; this capability has been engineered since 2023.

2) Agentforce and AI monetization.a) Agentforce ARR has reached $1.5 bn, and combined AI + Data ARR is approaching $4.0 bn.b) Bookings for Agentforce 1 Edition and Agentforce for apps more than doubled QoQ; AWU consumption hit 3.2 bn this quarter, +97% QoQ.c) Agentic calls initiated via MCP and CLI rose 6x, with MCP and API usage up significantly.d) Vertical expansion: Agentforce IT Service has 450+ customers, including NYU Langone and McAfee converting from ServiceNow; Agentforce Life Sciences has 140+ customers and covers the top five pharmas.e) Built on Regrello, Agentforce ops converts back-office processes like supply chains into agentic workflows; customers include Kellogg's and Dell, which manages 20,000 suppliers.f) A multi‑bn‑dollar expansion with the U.S. Army was completed, whose HR Command expects up to 55 mn Agentforce conversations per month; Deutsche Telekom, FIFA, NTT Data and Cisco are scaling AI investments.

3) Slack.a) Slack delivered the fastest single-quarter net AOV additions since the acquisition.b) Slack bot reached 1 mn active users five months post launch, up 150%+ QoQ, and higher-tier Slack upgrades have tripled since its debut.c) Slack Code was introduced as a shared space for humans and agents, forming a 'multi‑user IDE' with Slack bot; enterprises on Slack now exceed 1 mn.

4) Customer trends and rebuttal of the 'SaaSpocalypse' narrative.a) Management refuted a 'SaaS endgame': seats are growing YoY, contract terms are improving across segments and in both new and renewal cohorts, and pricing has held.b) Nine of the top ten global AI companies use Salesforce and Slack, with their spend up 435% YoY.c) Use cases: Uber for Business used Agentforce to process previously unattended inbound leads, lifting conversions 60% within two weeks; Replit tripled Agentforce sales seats.d) Internal use: Salesforce help agents have handled 5 mn+ customer conversations, with 64% resolved autonomously; Slack bot delivers an annualized 8.1 mn hours in efficiency gains, more than doubling QoQ.e) Guest David Friedberg (Ohalo) argued standardized vertical SaaS will be most disrupted by AI, while horizontal platforms will not be replaced by vibe coding; his team built out Salesforce in one month using Claude and Cursor.

2.2 Q&A. Selected exchanges and management responses.

Q: How will Claudeforce be taken to frontline sales? How will sales coordinate with Anthropic?

A: GA is slated for Dreamforce in Sep, and customers must upgrade to higher-tier editions to buy. This quarter, it rolled out first to global operating heads, and the most senior sales leader is already using it to demo to key accounts. The plan is to push company-wide now, and every Salesforce rep will demo post-GA, with coordinated selling alongside Anthropic's team.

Miguel Milano noted they previously hand-wired MCP servers and in-house skills, then turned it into a business 'cockpit': a Claude-based agent acts as a 'deputy CRO' for each line, inspecting pipelines and flagging in‑month or in‑quarter risks. Before visiting a country, he has the agent draft ~50 personalized Slack messages to local AEs, which he reviews and sends; teams are primed ahead of arrival and deals close faster. He expects a 'new wave of demand'; Slack and Claudeforce are complementary interfaces, since not all customers use Slack.

Q: How do clients view open source/open‑weight vs. frontier models?

A: No customer has asked for open source vs. frontier; enterprises will consume AI through packaged software. Benioff said customers see three core platforms: Salesforce for the front office, SAP for the back office, and Microsoft for productivity. Add a hyperscale cloud (three to four in Europe depending on country), one data lake provider, and one collaboration provider (Slack or Teams), and the vendor list is about six.

He does not expect enterprises to build, maintain and train models themselves, but to consume AI via packaged apps, akin to not caring which chip or router is underneath, with value accruing at the application layer. He also noted a dynamic market with room for both frontier and open models, citing NVIDIA Nemotron 1.3 fine‑tuning; Salesforce has built models since 2015, runs its platform on its own models, and models can be found on Hugging Face. He attributed the 'SaaS doomsday' narrative to podcasts and short sellers' agendas.

Q: How large is the monetization opportunity for Claudeforce and Headless? When will it show up in growth?

A: No monetization metrics or timing were provided. Benioff focused on the open vs. frontier model debate and did not address the monetization half of the question. In response to Goldman Sachs later, Robin Washington acknowledged Headless monetization is 'still in progress', deferring quantification to the Dreamforce investor day and the FY28 outlook. The confirmed path is pricing: both Claudeforce and Headless require upgrades to higher-tier editions.

Q: What share of Agentforce is paid production vs. pilots? Are cycles lengthening from pilot to paid and to top‑ups?

A: No mix given; 2,000 new paid production customers were added and bookings doubled YoY. Miguel emphasized bookings rather than cumulative ARR: bookings doubled YoY this quarter, with about 50% from existing customers topping up after burning through flex credits, a short-cycle motion. The 2,000 new paid production customers represent +70% QoQ, and several hundred new 'continuous‑consumption' agents were added.

He cited a large digital platform: a 'reactivation agent' launched six months ago completes 1,500 daily interactions to re‑engage dormant merchants, with AWU consumption reaching 45 mn, up 14x. The customer contributes about $40 mn per year and is discussing either outcome‑based rev share or an unlimited agreement. Benioff added that pricing now spans per user, per agent, by consumption, by usage or by outcomes; management did not directly address whether evaluation cycles are lengthening.

Q: Are the unit economics of Agentforce, Headless and Claudeforce comparable? Can ACV converge at steady state?

A: No steady‑state ACV conclusion; Headless monetization is ongoing, with upgrades as the primary path. Robin said customers value predictability and flexibility, and higher-tier upgrades deliver innovation in real time; relative to the installed base this is 'still early', and upgrading the base is the opportunity ahead. Miguel framed AI in two layers: employee productivity is packaged in higher-tier editions (with limited included usage), while customer‑facing use cases scale via Flex Credits that customers can buy on demand, pay overages for, or prepay in bundles.

Three quantitative signals: customers starting the agentic journey have Avg. AOV already 2x, with potential to reach 3–4x; only 5% of sales and service knowledge workers have upgraded to higher-tier editions; and higher-tier editions carry a 60–80% premium. Benioff conceded the company is 'still trapped in legacy per‑user pricing', leaving ample room for more aggressive models, with fuller details at Dreamforce.

Q: Net new AOV hit a four‑year high and churn is near record lows; what changed under the hood?

A: H1 net new AOV growth significantly outpaced overall AOV growth, and H2 acceleration is now 'arithmetic'. Miguel recapped that H2 last year was the best half for net new AOV growth; to re‑accelerate organic revenue in H2 this year, the entire org, including comp plans, was aligned to a single target: net new AOV must outgrow AOV. H1 delivered, hence 'Q3 is arithmetic and Q4 is close to arithmetic.' Demand pipelines are at historical highs, with additional monetization opportunities not yet piped; on supply, AEs number 15,000 and another 1,200 will be hired by year‑end.

Robin added that sales and service seats are showing 'sustainable growth'. Benioff stressed a 'third leg' of added value: AI can dynamically generate complex apps that previously took months or years, with Salesforce handling shared models, business rules and security beneath. He also aims to evolve from outcomes‑based pricing toward revenue‑share, e.g., 'we help you make $20 more, you pay us $2', with FY28 details slated for Dreamforce.

Q: The stock reaction suggests the market doubts the thesis and worries about fundamentals; thoughts?

A: Management argued undervaluation using Slack and Anthropic stakes and thus pursued heavy buybacks. Benioff said bidders call daily to buy Slack, and together with the company's Anthropic stake, those holdings represent a large portion of the firm's value, albeit without a quantified metric. Adding cash flow and the customer base only underscores the discount, which drove his directive to the CFO to repurchase as much as possible at attractive prices.

The company did not disclose the carrying value or valuation basis for the Anthropic stake. Management emphasized they acquired 'a lot of stock at good prices.'

<End of text>

Risk disclosure and statements:Dolphin Research Disclaimer and General Disclosures