Einstein licensing, three models wearing one brand
Salesforce markets Einstein as built in intelligence across the platform. The licensing reality is three pricing models stitched together: some intelligence bundled into the edition you already pay, some sold per user as add ons, and the generative and agent layer metered by consumption credits that behave nothing like a seat.
Prepared by Redress Compliance · August 6, 2026 · Salesforce negotiation advisory. Based on 20 to 30 AI engagements benchmarked 2024 to 2026.
Executive summary
Model one: edition bundling. A real layer of Einstein capability ships inside the higher editions, and the same features are sold separately on lower ones, which produces the brand's signature trap: in roughly 1 estate in 3 we benchmarked, buyers paid twice for an Einstein feature, once inside the edition and again as a standalone add on nobody reconciled against the entitlement.
Model two: per user add ons. The named user Einstein products price per seat on top of the edition, and they follow seat logic: scoped populations, adoption evidence, and the standard Salesforce renewal mechanics. They are the predictable third of the bill.
Model three: consumption credits. Agentforce and the generative features meter by credits, and credits are the easiest Salesforce line to overbuy: initial commitments ran 40 to 70 percent ahead of first year actual consumption across our engagements, and unused credits rarely rolled over, converting the oversizing directly into waste at term end.
The recovery is scoping, not austerity: right sizing editions and credits together cut AI add on cost 20 to 35 percent in our benchmarks without starving adoption. The controls are a feature to entitlement reconciliation before any purchase, a usage baseline before any credit commitment, and rollover or true down terms in the order form.
The three models, and which line each behaves like
| Model | How it prices | The failure mode |
|---|---|---|
| Edition bundled | Inside the edition rate you already pay, varying by edition tier | Paying again for features the edition already includes, 1 estate in 3 |
| Per user add ons | Per seat, on top of the edition, scoped populations | Enterprise wide licensing for departmental adoption |
| Consumption credits | Metered burn against committed credit blocks | Commitments sized 40 to 70 percent above real consumption, forfeited at term end |
The double paying trap, editions versus add ons
The edition ladder moved under everyone's feet: features that were premium add ons became edition inclusions at the higher tiers, and the add on SKUs kept selling. The result in 1 of 3 estates was the same capability entitled twice, once inside an Unlimited or high tier edition, once as the legacy add on renewing quietly alongside it.
The reconciliation is mechanical and pays immediately: the entitled feature list per edition, against the add on SKUs on the order form, against actual usage. Every overlap is a removal at renewal, and the same pass scopes the per user add ons to the populations that demonstrably use them, the standard discipline of the wider license optimization practice.
The Salesforce license optimization playbook
The edition entitlement reconciliation, the add on scoping method, the credit commitment sizing worksheet, and the renewal sequence for the AI estate.
Get the white paper →The credit layer, where forecasts meet the meter
Agentforce and the generative features draw Flex Credits per action, the meter the Headless 360 pillar works in detail: roughly 20 credits per action, 30 for Voice, near $500 per 100,000. The credit layer behaves like every consumption meter in this library: the seller forecasts adoption, the buyer signs the forecast, and the meter records reality, which in our engagements sat 40 to 70 percent below the commitment in year one.
The asymmetry is the standard one, sharpened by expiry: undersized commitments simply top up, oversized commitments forfeit at term end because credits rarely roll over. The controls follow: a usage baseline from a scoped pilot before any commitment, commitments sized to the measured floor, and rollover or true down language negotiated while the deal still wants signing. The full credit mechanics and buying models sit in the Agentforce pricing guide.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across AI engagements, 2024 to 2026
Across roughly 20 to 30 Salesforce AI engagements benchmarked between 2024 and 2026, consumption credits were consistently the most mispriced line, and the patterns repeated:
Initial Agentforce commitments against first year actual consumption, signed on adoption forecasts nobody had piloted.
The same Einstein capability entitled in the edition and purchased again as a legacy add on renewing alongside it.
The third pattern was expiry waste: unused credits dying at term end, treated as an adoption failure when it was a sizing failure. The estates that held their AI spend did the unglamorous sequence, reconcile, baseline, then commit, and their 20 to 35 percent recovery came with adoption intact, because none of the savings touched a user who actually used anything.
Your first five moves
- Demand the model per line on every AI quote: bundled, seated, or metered, each with its entitlement basis stated.
- Reconcile edition entitlements against add on SKUs and remove the 1 in 3 overlap at the next renewal.
- Pilot before committing credits: a scoped baseline of real burn, then a commitment sized to the measured floor.
- Negotiate rollover or true down terms on the credit block while the deal still wants signing; expiry is where the oversizing lands.
- Scope the per user add ons to demonstrated populations, and bring the whole AI estate into one renewal conversation. The Salesforce practice runs it with you, on your side of the table.
Frequently asked questions
How is Salesforce Einstein licensed?
Through three models under one brand: capabilities bundled into the higher editions, per user add ons priced per seat on top, and the generative and agent features metered by consumption credits. Every AI quote should state which model each line rides, because the entitlement basis and the negotiation differ per model.
Is Einstein included in our Salesforce edition?
Partly, and that is the trap: a real layer of capability ships inside higher editions while the same features sell separately on lower ones, and the edition ladder has moved over time. In 1 of 3 estates we benchmarked, buyers paid twice for a feature entitled in their edition and purchased again as a legacy add on.
How do Agentforce consumption credits work?
Generative and agent actions draw Flex Credits, roughly 20 per action and 30 for Voice at near $500 per 100,000, against committed credit blocks. Initial commitments ran 40 to 70 percent ahead of first year consumption in our engagements, and unused credits rarely roll over, so sizing from a piloted baseline is the entire game.
What happens to unused Einstein or Agentforce credits?
At standard terms they expire at term end without rollover, which converts commitment oversizing directly into waste. Rollover or true down language, negotiated at signing while leverage exists, is the control, alongside commitments sized to a measured floor rather than an adoption forecast.
How much can we save on Salesforce AI licensing?
Right scoping editions and credits together cut AI add on cost 20 to 35 percent across our benchmarks without starving adoption: the savings came from removing double paid features, scoping add ons to real populations, and sizing credits to measured burn, none of which touched an active user.
Should we commit to Agentforce credits before piloting?
No. The forecast gap in our file, 40 to 70 percent, is the cost of committing on the account team's adoption model instead of a scoped pilot's measured burn. Pilot first, baseline the credits per workflow, then commit to the floor with capped overage, exactly as with every consumption meter on the platform.
