Salesforce Einstein and Agentforce Pricing
Einstein spans predictive features, generative add-ons, the Einstein 1 edition, and the Agentforce consumption model. This page separates the tiers and shows where buyers overpay.
Einstein 1 Edition lists at $500 per user per month, while the standalone Einstein add-ons sit at $50 to $75 per user per month, and most buyers overpay by bundling Einstein across an entire org when fewer than 30 percent of seats actually use the AI. Salesforce sells AI at four levels that are easy to confuse: predictive Einstein features baked into editions, generative add-ons priced per seat, the all-in Einstein 1 Edition, and the Agentforce consumption model that sits above all of them. Buying the wrong level, or the right level for too many seats, is the most common AI overspend in a Salesforce estate.
The four levels of Einstein pricing
The first level is predictive Einstein, the scoring and forecasting features that come bundled into higher editions at no separate charge. The second is the generative add-ons, Einstein for Sales and Einstein for Service, priced per seat for features like email drafting and case summarization. The third is Einstein 1 Edition, which folds the AI and Data Cloud into the seat price. The fourth is Agentforce, the autonomous agent layer, which is consumption-priced rather than per seat and is covered in our Agentforce pricing guide.
| Level | What it is | List pricing |
|---|---|---|
| Predictive Einstein | Scoring, forecasting in higher editions | Bundled |
| Einstein for Sales (add-on) | Generative drafting, insights | $50 per user per month |
| Einstein for Service (add-on) | Case summaries, reply drafting | $75 per user per month |
| Einstein 1 Edition | AI plus Data Cloud in the seat | $500 per user per month |
| Agentforce | Autonomous agents | Consumption: $2 per conversation |
Standalone add-ons versus the edition
For most orgs the practical choice is between buying generative add-ons on top of Enterprise and moving up to Einstein 1. The add-on path costs the base Enterprise seat plus $50 to $75 for the AI, applied only to the users who need it. The Einstein 1 path costs $500 per seat but folds in Data Cloud and removes the add-on line items. The breakeven depends almost entirely on how many seats genuinely use the AI. Give AI to a third of the org and the targeted add-on path is far cheaper; standardize across nearly everyone and Einstein 1 starts to compete. The edition math is detailed in our Enterprise versus Unlimited comparison.
The everyone-gets-AI trap: The most expensive Einstein decision is applying it org-wide because it is simpler to administer. First-year AI adoption rarely exceeds 30 to 40 percent of licensed seats. Paying $50 to $500 per seat for an entire org when a third use the feature is shelfware at AI prices. License the AI to the user groups that use it, and expand as adoption proves out.
Predictive versus generative cost
The two kinds of Einstein behave differently on cost. Predictive features, scoring and forecasting, run on Salesforce's models at no incremental consumption charge once the edition includes them. Generative features, drafting and summarization, increasingly ground on your unified data, which means they consume Data Cloud credits each time they run. An org that rolls out generative Einstein broadly should expect a parallel Data Cloud bill, modeled in our Data Cloud pricing guide. Treating the per-seat add-on as the only cost understates the real number.
| Capability | Cost behavior | Watch for |
|---|---|---|
| Predictive scoring and forecasting | Bundled, no consumption | Already included, do not pay twice |
| Generative drafting and summaries | Per-seat add-on plus credit burn | Parallel Data Cloud consumption |
| Autonomous agents (Agentforce) | Consumption per conversation | Volume commitment and grounding credits |
Negotiation and right-sizing
The Einstein decisions are mostly about scope, not rate. The largest savings come from licensing AI to the right user count rather than the whole org, holding the per-seat add-on rate for the term, and avoiding the bundled edition until adoption justifies it. Where Agentforce volume is part of the deal, the consumption commitment needs the same true-down protection as any usage product. The clauses that protect these terms are in our contract red flags guide, and the discount context in our discount benchmarks.
Phase the rollout: The cleanest way to control Einstein cost is to license a pilot group, measure real adoption and Data Cloud burn for a quarter, then size the full purchase to proven usage. Salesforce prefers the all-in commitment at signature because it locks in spend before adoption is known. A phased rollout keeps the bargaining power with the buyer.
Why AI adoption lags the license
The gap between licensed AI seats and used AI seats is the defining cost problem of enterprise AI, and Einstein is no exception. Adoption lags for predictable reasons: the features require workflow changes, they need quality data to produce useful output, and users default to the methods they know. First-year adoption of a generative add-on commonly sits between a quarter and 40 percent of licensed seats, even when the rollout is well run. Licensing the feature to the whole org on the assumption that everyone will use it converts an AI investment into shelfware at premium prices.
The remedy is to license to demonstrated demand rather than projected demand. Identify the user groups with a clear, repeated use case, license those, and hold the rest until adoption proves out. This is uncomfortable for a vendor selling an org-wide vision, but it matches spend to value and preserves the budget for expansion once the feature has earned its place. The same right-sizing logic drives our SaaS license optimization service.
The hidden Data Cloud cost of generative Einstein
Generative Einstein features increasingly ground their output on your unified customer data, which means each generation can consume Data Cloud credits in addition to the per-seat add-on fee. An organization that budgets only for the $50 to $75 per-seat line and ignores the credit draw will see a second, variable bill it did not plan for. The two costs have to be modeled together, because a broad generative rollout that looks affordable on the per-seat line alone can become expensive once the credit consumption is added.
This interaction is why the AI products cannot be priced in isolation. A realistic Einstein business case includes the per-seat add-on, the expected Data Cloud credit burn from grounding, and, where agents are involved, the Agentforce consumption meter. The full credit model is in our Data Cloud pricing guide, and the agent layer in our Agentforce pricing guide.
| Rollout scope | Per-seat cost | Hidden credit cost |
|---|---|---|
| Targeted pilot group | Bounded, easy to justify | Small, measurable |
| Department-wide | Moderate | Material if grounding is heavy |
| Org-wide generative | High, much unused | Largest, often unbudgeted |
Measuring AI value before scaling
The discipline that controls Einstein cost is the same one that controls any consumption-adjacent product: measure before you scale. A pilot quantifies the time saved per user, the quality of the generated output, and the credit consumption per active seat. Those numbers turn an AI rollout from an act of faith into a business case with a return that can be defended to finance. They also reset the negotiation, because the buyer sizing the full purchase to measured adoption is negotiating from evidence rather than from the vendor's projection.
Scaling without measurement is how AI budgets overrun. The pressure to deploy quickly is real, but a 90-day pilot costs little against a year of org-wide licenses for a feature that a third of users adopt. The buyers who get AI economics right are the ones who treat the first phase as a measurement exercise, then expand into proven demand, protected by the clauses in our contract red flags guide.
Common Einstein pricing questions
Is Einstein included in Enterprise edition?
Some predictive Einstein features are bundled into the higher editions, but the generative add-ons, Einstein for Sales and Einstein for Service, are separate per-seat purchases on top of Enterprise. Check exactly which capabilities the edition includes before paying for an add-on that duplicates a bundled feature.
Do Einstein add-ons require Data Cloud?
Increasingly, generative features ground on unified data, which means they draw Data Cloud credits when they run. The per-seat add-on fee and the credit consumption are separate lines, and both belong in the business case for a generative rollout.
How does Einstein 1 differ from buying add-ons?
Einstein 1 Edition folds the AI and a Data Cloud allotment into a single $500 per-seat price, while the add-on path keeps the base seat and layers AI only on the users who need it. Einstein 1 wins only when nearly all seats use the AI; otherwise the targeted add-on path is cheaper.
Can Einstein licenses be reduced at renewal?
Only if a true-down right is negotiated. Standard terms make AI seats as hard to reduce as any other license, so an over-bought AI footprint becomes a recurring cost unless the contract allows resetting to actual usage.
Building the AI business case
A defensible Einstein business case rests on three numbers, and a pilot is the only reliable way to get them. The first is adoption: what share of licensed users actually use the feature in a typical week. The second is value per active user: the measured time saved or quality gained, expressed in terms finance recognizes. The third is the parallel consumption cost, the Data Cloud credits a generative feature draws when it grounds on unified data. Without all three, the case is a projection, and projections are how AI budgets overrun.
The pilot produces these numbers under real conditions rather than demo conditions. A 90-day run on a defined user group shows the true adoption curve, which almost always climbs more slowly than the vendor forecast, and the true credit burn per active seat, which is usually higher than the per-seat add-on suggests in isolation. Armed with the measured figures, the buyer can size the full purchase to demonstrated demand and defend the spend line by line.
The same numbers reset the negotiation. An account team selling an org-wide rollout is selling the aspirational adoption curve; a buyer holding measured adoption data is negotiating against evidence. The buyer who can say that the pilot showed 35 percent weekly adoption and a specific credit cost per seat is no longer accepting the vendor's framing of value, which is the foundation of the right-sizing work in our SaaS license optimization service.
The discipline is uncomfortable because it slows the rollout, and the pressure to deploy AI quickly is real. But a phased approach that licenses a pilot, measures, then expands into proven demand costs far less than a year of org-wide licenses for a feature a third of users adopt. The business case built on measured numbers is the one that survives the next budget review.
Where this fits
Einstein is the per-seat layer of Salesforce AI; Agentforce is the consumption layer and Data Cloud is the data layer underneath both. Start with the complete Salesforce licensing guide, then read the Agentforce pricing guide and the Data Cloud pricing guide. For help sizing the AI footprint to real adoption, see our Salesforce advisory practice, our SaaS license optimization service, and the software licensing advisory team.