Salesforce combines a full CRM suite with Agentforce, an agent platform billed in Flex Credits, and the Data 360 foundation. Built for enterprise sales, service and IT teams that need governed AI acting on real customer records.
Used this tool? Rate it
Used this tool? Rate it
Salesforce is a cloud platform for customer relationship management that has repositioned itself around autonomous AI agents. The company's own front-page framing is unambiguous: "Salesforce is the #1 AI CRM, helping companies become Agentic Enterprises where humans and agents drive success together through a unified AI, data, and Customer 360 platform." That single sentence tells you what changed. For two decades Salesforce sold a system of record — a place where deals, cases and contacts lived. It now sells a system of action, where software agents read those records, decide what to do, and execute.
The longer official definition fills in the shape of the product. Salesforce describes itself as bringing "companies and customers together by providing a unified set of applications — powered by agentic AI and data — that help every department, including sales, service, marketing, commerce, and IT, work as one." In practice this means you are not buying one tool. You are buying into a stack with four distinguishable layers: the CRM applications themselves (sales, service, marketing, commerce), the Data 360 foundation that unifies records, the Agentforce layer that builds and runs agents on top of that data, and an integration tier including MuleSoft for systems that do not connect natively.
For a directory of AI tools, the honest question is whether the AI here is the main line or a thin veneer over legacy CRM. The evidence says main line, and the evidence is commercial rather than rhetorical. Three tests matter more than marketing copy: whether the AI is included in standard plans or sold separately, whether it has a dedicated metering unit, and whether pricing was restructured around it. Salesforce fails the first test in a revealing way — the pricing page states plainly that "AI can be added to Enterprise and above," meaning AI is not simply bundled into every tier. It passes the second decisively: Agentforce has its own currency, Flex Credits, priced at "$500 USD/Per 100k Credits." And it passes the third, because an entire top-of-range edition family now exists purely to package agents, starting "from $550 user / month."
That combination — separately sold, separately metered, and with the price ladder rebuilt around it — is the signature of a strategic bet, not an add-on. It also means the AI has a running cost that scales with usage, which is the single most important practical fact for anyone evaluating this platform.
Agentforce is the component that justifies the "agentic" label. Salesforce defines the underlying concept as follows: "Agentic AI refers to advanced artificial intelligence systems that can act autonomously to achieve a specific goal with minimal human intervention. Unlike traditional AI that only generates content or analyzes data, agentic AI can reason, plan multistep actions, and execute tasks." The distinction being drawn is between a copilot that drafts an email for you and an agent that qualifies the lead, updates the opportunity, books the meeting and logs the interaction without a human in each loop.
The tooling around this is more substantial than a prompt box. The product page lists a full development lifecycle: "Agent Builder — Configure Agentforce, Subagents, Actions, and Instructions in a few clicks," alongside Agent Script for deterministic control over agent behaviour, Agentforce Voice for telephony and voice channels, and Agentforce Observability to "Monitor, analyze, and optimize agent performance in near real time." Two entries in that list matter disproportionately for technical buyers: Agentforce MCP Support, which allows connection to external tools through the Model Context Protocol via verified partner servers, and Multi-Agent Orchestration for building teams of agents that hand work between each other.
The build path is deliberately layered so that different skill levels can participate. Salesforce's own description spans the whole range: "You can use Flows to create automations across any system. You can use MuleSoft API connectors to connect Agents to any system. And last, you can use your Apex & Javascript to create the custom code and business logic for your Agent." Admins configure declaratively, integration teams wire up APIs, and developers drop into code for anything the clicks cannot express.
An agent is only as trustworthy as the data it reasons over, which is why Data 360 sits directly beneath Agentforce rather than beside it. Salesforce markets it as the way to "Unlock trapped data with the only data platform native to the world's #1 agentic CRM." Note the product name carefully — this component was called Data Cloud until a rebrand, and material written before that change uses the old name for the same thing. Anything you read about Salesforce data pricing should be checked against the current Data 360 pages rather than trusted from memory.
The architectural point worth understanding is that Data 360 and Agentforce share a billing currency. The data pricing page states that you can "Seamlessly transfer your Flex Credits between Data 360 and Agentforce." One pool of credits funds both data operations and agent actions. This is convenient for procurement and genuinely useful for capacity planning, but it also means a heavy data pipeline and a heavy agent workload compete for the same budget line.
Beneath the AI story sits the mature product most people still think of as Salesforce. Sales Cloud covers lead, account, contact and opportunity management, quoting, forecasting, pipeline management and conversation intelligence. Service Cloud handles case management across channels, knowledge bases, chat and bots. Marketing and Commerce Cloud extend into campaigns and storefronts. Customer 360 is the unifying view across all of them, and the homepage positions it as the way to "Ground every decision in proven business context."
These applications are not legacy baggage — they are the reason the agents have something to act on. An agent that can "update customer records, automate workflows, or resolve cases" needs those records, workflows and cases to exist in a structured form first.
Salesforce is explicit that not everything connects out of the box, and its guidance for the gaps is worth quoting because it has direct cost implications: "For those that don't have built-in integration, like on-premises data sources, you can purchase licenses of MuleSoft, work with our integration partners, or use solutions and apps on AgentExchange." Three paths, and two of them involve additional spend. On-premises systems and older line-of-business applications are precisely where integration budgets tend to overrun.
This section deserves separate treatment because agent pricing is where evaluations most often go wrong, and where the difference between list price and real cost is largest.
There are three parallel commercial models, and a given customer may end up using more than one.
Pay per action, via Flex Credits. The unit price is "$500 USD/Per 100k Credits," and Salesforce's help documentation defines the conversion: "One Agentforce action consumes 20 Flex Credits ($0.10 USD)." An action is a discrete thing the agent does — updating a customer record, running a workflow step, resolving a case. The arithmetic is simple but the forecasting is not, because a single customer conversation typically consumes several actions rather than one. Anyone modelling this should count actions per resolved case, not conversations per month.
Pay per conversation. The alternative model is listed at "$2 USD/Per conversation" for customer-facing agents. This trades granularity for predictability: you stop caring how many internal steps an agent takes and pay a flat rate per interaction. Which model is cheaper depends entirely on how action-dense your workflows are, and that is not knowable before deployment without a pilot.
Pay per seat. For employee-facing use there are flat add-ons — "$125 user / month" for Sales, Service and Field Service, and $150 for Industries clouds — both described as covering "Unmetered Agentforce usage for employees." Unmetered is the operative word: these convert a variable cost into a fixed one. Separately, a lightweight Agentforce User License exists at "$5 USD/User/Month" but the same page notes it "(Requires Flex Credits)," so it grants access rather than usage.
At the top of the range, the Agentforce 1 editions bundle everything: "from $550 user / month," with the "Agentforce add-on included" and "2.5M Flex Credits per org per year." Service Cloud's equivalent tier confirms the same allowance. That included credit pool is the number to anchor on, because it defines where metered overage begins.
The direction of travel here is documented. The Register reported Salesforce moving away from its original approach, quoting CEO Marc Benioff: "When we first started with Agentforce, we were talking about [charging] so much per conversation." The company subsequently introduced an Agentic Enterprise License Agreement offering "a flexible menu of options and reusable credits in a single package, but on seat-based pricing." Buyers should read that as an admission that pure consumption pricing created friction — and should expect the commercial packaging to keep moving.
Agent pricing sits on top of ordinary CRM licensing, and the base ladder is public. For Sales Cloud the published tiers are Starter Suite at "$25 USD/User/Month," Pro Suite at "$100 USD/User/Month," Enterprise at "$175 USD/User/Month," Unlimited at "$350 USD/User/Month," and Agentforce 1 Sales at "$550 USD/User/Month (Billed annually)." Service Cloud mirrors this structure at the same five price points, with Agentforce 1 Service as its top tier.
Two footnotes on that page carry more weight than the numbers. The first is the AI gate already mentioned — AI is added at Enterprise and above, which means the $25 and $100 tiers are not entry points to the agent platform. The second is contractual: "You can purchase licenses of Salesforce Starter with a monthly or annual contract. … All other subscriptions are generally paid annually in advance," and "Most Salesforce products use annual contracts, but Salesforce subscription terms vary." Annual prepayment on a per-seat product means the commitment is the seat count multiplied by twelve months, agreed before you know your real adoption curve.
There is a free entry path. Salesforce states that "Salesforce Starter helps you hit the ground running for free, providing an all-in-one suite of sales, service, and marketing tools designed specifically for small business needs." Trials exist across the range, but with a restriction that catches teams out: "Trial subscriptions cannot be used in a production capacity and must be used only for non-production purposes." You cannot quietly run a small team on a trial while you decide.
Data 360 carries its own pricing dimension with three models: Flex Credits at "$500 Per 100k Flex Credits," Profiles at "$240 Per 1k Profiles/Year," and Enterprise Profiles at "$420 Per 1k Profiles/Year." Profile-based pricing is aimed at customer data platform use cases where the record count is the natural unit; credit-based pricing suits variable processing loads.
Customer service deflection. The most mature agent use case. Agents answer questions, resolve cases, manage orders and troubleshoot issues on always-on channels, escalating to humans when the problem exceeds their scope. The economics are legible here because the alternative cost — a human handling the same case — is already measured.
Sales development. Agents answer product questions, handle objections and book meetings for reps around the clock, which suits high-volume inbound motions where response latency correlates with conversion.
Employee support. Internal agents handle HR policy questions, leave requests and onboarding tasks. The per-seat unmetered add-on is designed for exactly this pattern, where usage is broad and continuous rather than spiky.
IT service management. Salesforce has pushed agents into help desk territory — ticket resolution, alert handling, access resets — placing the platform in direct competition with dedicated ITSM vendors.
Field service and appointment scheduling. Agents schedule technician visits based on history and context, update work order status, confirm parts availability and manage reminder notifications.
Unified data and analytics. Data 360 consolidates records from across systems so that both humans and agents work from one grounded view rather than departmental copies.
Large enterprises with complex processes. This is the core fit. Organisations with many teams, regulatory obligations, existing integrations and dedicated administrators get value from depth that smaller buyers experience as overhead.
Mid-market companies on a growth path. The tiered ladder allows starting at Pro Suite and stepping up. The caveat is that the AI story genuinely begins at Enterprise, so mid-market buyers specifically wanting agents should budget from $175 per user per month upward, not $25.
Small businesses. Salesforce offers a free starting point and Starter at $25 per user per month, which is a credible small-business CRM. But the platform's centre of gravity — annual contracts, configuration effort, consumption-metered AI — is calibrated for larger organisations. Small teams should be clear-eyed that they are entering an enterprise ecosystem.
Teams who should look elsewhere. If you want AI assistance without governance requirements, without existing CRM data to ground it in, or without an administrator to own configuration, a lighter tool will deliver value faster and cheaper. The independent analyst view is worth weighing here: a KeyBanc CIO survey reported by The Register found that "customers' data is not in order to do meaningful AI work; and 2) Agentforce, as a product, just isn't there," and that "the majority of customers are not willing to pay for AI capabilities through their CRM provider." Data readiness is a genuine prerequisite, not a detail.
Salesforce runs primarily in the browser as a cloud service, with no client installation required for standard use. Mobile access is available through Salesforce's mobile applications for iOS and Android, aimed at field and travelling users who need record access and approvals away from a desk.
For programmatic access the platform exposes a comprehensive API surface — REST, SOAP, Bulk and Connect APIs share the daily allocation described earlier. Development can happen declaratively in the browser or in code through Apex and JavaScript, and sandbox environments support staged testing, with a Full Sandbox carrying its own 5,000,000 daily API allocation.
Ecosystem extension runs through AppExchange for applications and AgentExchange for agent components and verified MCP servers. Agentforce Voice extends agents onto telephony channels, and Slack serves as a conversational surface where work already happens. MuleSoft covers integration with systems lacking native connectors, including on-premises data sources.
Microsoft Dynamics 365 with Copilot is the closest like-for-like competitor for organisations already invested in Microsoft 365 and Azure, where identity and productivity integration is a decisive advantage.
HubSpot competes hard in mid-market, generally winning on time-to-value, transparent pricing and lower administrative burden, while offering less depth for complex enterprise processes.
ServiceNow overlaps increasingly in service and IT workflow automation. As Salesforce pushes agents into ITSM, the two platforms meet directly in the help desk.
Zoho CRM and similar cost-focused suites serve small and mid-sized businesses that want broad functionality at materially lower per-seat cost, accepting a smaller partner ecosystem.
Zendesk remains a focused alternative for customer service specifically, without the surrounding CRM platform commitment.
Purpose-built agent frameworks are the choice for teams wanting to build AI agents without adopting a CRM at all — appropriate when the system of record is not Salesforce and grounding data lives elsewhere.
Cost scales with success. Consumption pricing means an agent that works well and handles more volume costs more. That is defensible economically but it inverts the familiar SaaS assumption that heavier usage of a paid seat is free.
Data readiness is the binding constraint. Independent analyst commentary points at this repeatedly. Agents grounded in incomplete or contradictory records produce confident wrong answers, and no amount of prompt engineering fixes a fragmented customer view.
Adoption is not universal. The Register reported that "more CIOs expect to deprioritize Salesforce in their IT budgets than increase investment over the next 12 months" in the KeyBanc survey, alongside concerns about "aggressive price increases." Forrester separately warned that "the disconnect between the inflated promises of AI vendors and the value created for enterprises will force a market correction." These are analyst opinions rather than settled fact, but they belong in an honest assessment.
Vendor performance claims are self-reported. Figures Salesforce publishes about ROI, adoption or agent effectiveness — including commissioned analyst reports promoted on its own site — are vendor-reported and not independently audited. Treat them as marketing input, not evidence.
Configuration complexity and dependency. Deep customisation creates capability and lock-in simultaneously. Migration cost rises with every custom object, workflow and integration, and the resulting switching cost is a real factor in renewal negotiations.
Product naming shifts. Data Cloud became Data 360, and editions have been repackaged around Agentforce. Documentation, third-party guides and even internal runbooks go stale quickly; verify names and prices against current official pages rather than prior knowledge.
Pricing is explicitly provisional. Salesforce states on its pricing pages that they are "provided for information purposes only and subject to change." Every figure here is a list price and a starting point for negotiation, not a quote.
For a CRM the decisive privacy question is what happens to customer data when it is sent to a language model. Salesforce's answer is the Einstein Trust Layer, and the official developer documentation states its core commitment directly: "Salesforce has agreements in place with LLM providers, such as OpenAI. These agreements include commitments for zero data retention." Data sent to an external model is not retained by that provider after the response returns.
Alongside retention, the Trust Layer documents two further protections: masking of sensitive data such as social security numbers before prompts leave the platform, and toxicity detection applied to model generations before they reach a user.
One boundary deserves stating precisely rather than glossed. The zero data retention commitment described in that documentation concerns third-party LLM providers. It is a statement about what external model vendors may do with data passed to them. Salesforce's public developer documentation on that page does not make an equally explicit statement about every internal use of customer data across its own AI programme. Administrators evaluating this should review the data-use settings in their own org and the contractual terms attached to their specific products rather than assuming a single blanket policy covers all cases. Salesforce publishes its contractual agreements by product, which is itself an indication that terms vary across the portfolio.
On the governance side, Agentforce Observability provides near real-time monitoring of agent behaviour, which matters for audit and for catching drift. Sandbox environments allow changes to be validated before reaching production data. Salesforce also publishes a trust site presenting platform status and compliance certifications, and maintains published privacy documentation covering data subject rights, including reporting on requests received from California residents during 2025.
For regulated industries the practical checklist is: confirm data residency for your region, confirm which models are in scope for your deployment, confirm masking configuration for your sensitive fields, and get the retention and training terms for your specific contract in writing. The platform provides the mechanisms; the specifics depend on what you sign.
There is a free entry point. Salesforce states that "Salesforce Starter helps you hit the ground running for free, providing an all-in-one suite of sales, service, and marketing tools designed specifically for small business needs." Paid plans begin at $25 per user per month for Starter Suite. Free and trial access is genuinely usable for evaluation, but note that "Trial subscriptions cannot be used in a production capacity and must be used only for non-production purposes."
It depends which of three models you use. Pay-per-action runs through Flex Credits at "$500 USD/Per 100k Credits," where "One Agentforce action consumes 20 Flex Credits ($0.10 USD)." Pay-per-conversation is listed at "$2 USD/Per conversation." Per-seat add-ons are $125 per user per month for Sales, Service and Field Service, or $150 for Industries clouds, both covering unmetered employee usage. Top-tier Agentforce 1 editions start "from $550 user / month" and include 2.5M Flex Credits per org per year.
No. The Sales Cloud pricing page states that "AI can be added to Enterprise and above," which places the entry point for agent capability at $175 per user per month rather than the $25 Starter tier. Even at Enterprise level, agent usage is metered separately unless you buy an unmetered add-on or an Agentforce 1 edition.
Salesforce's developer documentation states that "Salesforce has agreements in place with LLM providers, such as OpenAI. These agreements include commitments for zero data retention," meaning third-party model providers do not retain data sent to them. The Trust Layer also applies data masking and toxicity detection. That documented commitment specifically covers external LLM providers; for the full picture on internal data use you should review your org's data settings and the contractual terms for your specific products, since Salesforce publishes agreements per product rather than as a single universal policy.
Salesforce Starter can be bought monthly or annually. Beyond that, "All other subscriptions are generally paid annually in advance," and "Most Salesforce products use annual contracts, but Salesforce subscription terms vary." Budget for a twelve-month commitment on seat count, and confirm renewal terms before signing.
The daily allocation is "100,000 + (number of licenses x calls per license type)," with Enterprise and Professional licences contributing 1,000 calls each and Unlimited and Performance licences contributing 5,000 each. Developer Edition is capped at 15,000 per day and a Full Sandbox at 5,000,000. Long-running calls of 20 seconds or more are capped at 25 concurrent requests in production orgs and sandboxes.
Yes, with caveats about cost. Salesforce advises that for systems "that don't have built-in integration, like on-premises data sources, you can purchase licenses of MuleSoft, work with our integration partners, or use solutions and apps on AgentExchange." Each of those routes carries additional spend or effort beyond the CRM subscription.
It can be, with realistic expectations. The free tier and $25 Starter Suite are legitimate small-business options, and the platform can scale as you grow. However, the ecosystem is built around annual contracts, configuration work and consumption-metered AI, so a small team without administrative capacity may find a lighter CRM delivers value sooner at lower total cost.
Track consumption rather than seats. Bundled Flex Credits are an annual allowance, so divide it into quarterly budgets and monitor against it using the Digital Wallet tooling Salesforce provides for consumption-based products. During a pilot, measure actions consumed per resolved interaction so you can forecast overage before it appears on an invoice.
The evidence is mixed and worth reading from both sides. Salesforce describes Agentforce as the fastest-growing product in its history and publishes customer examples and commissioned reports — all vendor-reported and not independently audited. Independent analysis is more cautious: a KeyBanc CIO survey reported by The Register concluded that "Agentforce, as a product, just isn't there" and that most customers are unwilling to pay for AI through their CRM provider. The reasonable conclusion is that outcomes depend heavily on data readiness and use case selection, which argues for a measured pilot before a broad commitment.
