If you sit anywhere near a budget, Salesforce has probably pitched you AI in the last ninety days. Einstein, Agentforce, Data 360, agents, copilots, credits — the names change fast, the demos are polished, and the pricing pages assume you already know what everything means.
Here's the thing: underneath the branding, there are only three kinds of AI being sold, and they map cleanly to three kinds of business value. Once you can tell them apart, every Salesforce AI conversation — with your account executive, your IT team, or your board — gets dramatically easier. This guide is that map: what each type of AI is, which purchasable Salesforce products deliver it, how the pricing actually works, and the questions worth asking before signing anything.
(This is the executive version. We've also written companions for your team on the ground and your IT leaders — same map, different altitude.)
Predictive AI looks at your historical data and estimates what happens next. Which leads will convert. What the quarter will actually close at. Which customer is about to churn. It's been in Salesforce for nearly a decade under the Einstein brand — it's the most proven and least flashy of the three.
Generative AI produces new content: drafts an email, summarizes a call, writes a knowledge-grounded answer to a customer question. This is the large-language-model wave that arrived with ChatGPT, embedded into the tools your team already uses.
Agentic AI goes one step further: instead of predicting or drafting for a human to act on, an AI agent does the work — reasons through a multi-step task, uses your business systems, and escalates to a human when it hits its limits. This is where Salesforce has bet the company, and it's where most of the sales attention is right now.
Everything on a Salesforce pricing sheet with "AI" on it is one of these three — or the data plumbing underneath them. That's the whole taxonomy. (Machine learning, LLMs, and neural networks are the underlying techniques; you don't need them for a buying decision.)
What it looks like in practice: a score next to every lead and opportunity, a forecast that's based on pattern analysis rather than rep optimism, cases automatically classified and routed, and — on the marketing side — Einstein features inside Account Engagement like behavior scoring and campaign insights that tell you which prospects are genuinely warming up.
What you'd buy: much of this is already included in the editions and products you may own — higher editions of Sales Cloud and Service Cloud bundle a lot of Einstein predictive capability, and Account Engagement's Advanced and Premium tiers include its Einstein features. Some capabilities are per-user add-ons. Before budgeting anything new here, have someone audit what your current contract already includes; in our experience most companies are paying for predictive AI they've never turned on.
The honest caveat: predictions need history. Scoring models trained on a few hundred sloppy records produce confident-looking nonsense. If your CRM data is thin or dirty, fix that first — more on this below, because it's the theme of this entire subject.
What it looks like in practice: sales emails drafted from CRM context, call recordings summarized with action items, service replies suggested from your knowledge base, campaign copy variants generated inside your marketing tools. The pattern across all of it: the AI produces a draft, grounded in your data, and a human approves it.
What you'd buy: generative features ship inside the clouds you already license — Sales Cloud, Service Cloud, Marketing — typically via edition upgrades or AI add-ons, and Prompt Builder lets your team build reusable, grounded prompts into everyday workflows. Salesforce Foundations (a free feature set for Enterprise Edition and up) includes starter AI capability, which makes it a low-risk way to see generative features on your own data before spending anything.
The honest caveat: quality tracks the data it's grounded in. An AI answer drafted from a stale knowledge base is a fast wrong answer. The Einstein Trust Layer handles the security mechanics — keeping your data out of model training, defending against prompt attacks, logging everything for audit — but it can't make bad source content good.
What it looks like in practice: a service agent that resolves routine cases end-to-end, around the clock, and hands the hard ones to your team with full context. A sales development agent that answers inbound leads in minutes, qualifies them, and books meetings. Employee-facing agents that answer policy questions or chase down data in Slack. You define the guardrails — what it may do, what it must never do, when it must hand off — and the agent works within them.
What you'd buy: Agentforce is the product family, and since late 2025 Salesforce has packaged the whole platform as Agentforce 360 — including Agentforce Builder (build agents conversationally), Agentforce Voice (agents on the phone), and Slack as the everyday interface where humans and agents work together.
How the pricing works — and this is genuinely different from anything else you license from Salesforce: agentic AI is mostly usage-based. As of mid-2026 the shapes are: pay per conversation (around $2 each), pay per action via Flex Credits (sold in blocks — roughly $500 per 100,000 credits, where a typical agent action burns 20 credits, so call it ten cents an action), or flat per-user Agentforce 1 editions for predictable budgeting at scale. Enterprise Edition and above currently get 100,000 Flex Credits free through Foundations — enough to run a real pilot without a purchase order. Numbers move; the shapes are the point. Usage-based pricing means an agent that works hard costs real money, so treat consumption like a cloud bill: monitor it, alert on it, and know your unit economics before you scale it.
Underneath all three types sits the same foundation: Data 360 (renamed from Data Cloud), Salesforce's layer for unifying customer data across systems into one profile that AI can actually use. It's priced on consumption, like the agents are.
Here's the plain-English version of why it matters: every AI capability above is only as good as the data it reads. This isn't consultant folklore — in Salesforce's own research, 86% of technical leaders agree AI's outputs are only as good as its data inputs, and roughly six in ten organizations admit they don't have a unified data strategy. Duplicate contacts, half-empty fields, three systems disagreeing about the same customer — predictive models learn the mess, generative drafts repeat the mess, and agents act on the mess, autonomously and at scale. The least exciting line item in an AI budget — data cleanup and unification — is reliably the one that determines whether the exciting line items pay off. If your team can't trust the CRM's reports today, that's your first AI investment, and it isn't optional.
Start from a problem, not a product. "Response time on inbound leads is four hours" is a problem an SDR agent can be measured against. "We need an AI strategy" is how pilots die in month three.
Audit what you already own. Between edition-included Einstein features and the free Foundations tier, most Enterprise-and-up customers have meaningful AI capability they've never enabled. Free is a good price for a first experiment.
Pilot on one metric. One agent, one process, one number that should move — case deflection rate, lead response time, forecast accuracy. Expand on evidence, not enthusiasm.
Know your meter. Per-user, per-conversation, per-action, per-credit — each pricing shape rewards different usage patterns. High-volume simple interactions favor per-action; deep always-on usage eventually favors flat editions. Model it before the renewal, not after.
Budget for the data work. Whatever the AI line item is, put a real number next to it for data quality and integration. That ratio is what separates companies where AI stuck from companies with an expensive demo.
If your team wants the ground-level view, the companion pieces cover what these tools actually do in a rep's or marketer's day and what IT needs to put in place to run them safely.
And if it would help to talk any of this through with someone who isn't carrying a quota on it — we spend our days inside Salesforce and Account Engagement for B2B companies, and "what's actually worth buying" is a conversation we're happy to have with no deck and no pitch. We're easy to find.
Founder of Thompson Technology. Salesforce and Account Engagement consultant for B2B companies.
Connect on LinkedIn →Want an outside read on where your revenue engine is leaking? That’s a 30-minute conversation — no deck, no obligation.
