AI Usage Limits and Overage: Navigating Salesforce’s AI Consumption Model

AI Usage Limits and Overage: Navigating Salesforce’s AI Consumption Model

Introduction – Why Usage Limits Matter

Salesforce’s AI pricing often hides a “meter” behind each add-on. In the excitement to deploy Einstein GPT or the new Einstein Copilot, it’s easy to overlook that these tools come with usage limits. Without clarity on these limits, enterprises risk incurring unpredictable overage bills that exceed budgets.

For example, generating thousands of AI-driven emails or case summaries can incur costs beyond the base license if you exceed the included allowance. Early adopters must push for transparency and control over these usage metrics. Read our overview of Salesforce AI & Automation Licensing.

The goal is to ensure predictable costs – negotiating terms that prevent surprise bills and give you levers to manage consumption.

A skeptical, strategic approach to Salesforce’s AI consumption model will pay off when it’s time to reconcile usage with costs.

How Salesforce Measures AI Usage

Understanding how Salesforce measures AI usage is the first step to controlling it. Salesforce uses units like “credits,” “API calls,” “interactions,” or “outputs” to quantify AI consumption:

• Credits: Salesforce often defines a credit as the unit of generative AI work. For instance, producing one AI-generated email or case summary might consume a certain number of credits. Credits abstract the complexity behind the scenes (tokens, compute time) into a simple number.
• API Calls: If you’re invoking Einstein GPT via API or using underlying AI services programmatically, those calls count against limits. An “Einstein GPT API call” might be one request to generate or analyze text. These could be metered if you integrate AI into custom code.
• Interactions/Conversations: For chat-based assistants like Einstein Copilot (Salesforce’s AI “copilot” that users converse with), usage might be counted by conversation or interaction. For example, each conversation session with the Copilot (a user prompt and the AI’s response, possibly with follow-up Q&A) could be a billable interaction.
• Outputs: In some contexts, Salesforce might measure outputs (the pieces of content generated). For instance, if an AI feature generates a meeting summary, an email draft, and a report as three outputs, each of these might count individually toward a usage quota.

Per-User vs. Org-Wide Allocation: Salesforce’s licensing historically ties many features to users, but AI usage blurs those lines. Some AI usage allowances are per user (each licensed user gets their own quota of credits or interactions), while others are effectively org-wide pools.

For example, if each Sales Cloud Einstein GPT user gets a set number of monthly credits included, do those combine into one bucket? In practice, Salesforce often treats the credits as an organization-level pool accessible by all licensed users, even if described per user.

This prevents the scenario where one power-user runs out of credits while others go unused. Always clarify this with Salesforce – if not already pooled, negotiate to pool credits org-wide so usage balances out across your team.

Published Limits:

Salesforce does publish some limits in documentation, but you have to dig for them. As of 2025, you can find references to Einstein GPT credit limits and Copilot conversation limits in their pricing FAQs and Salesforce Help articles.

For instance, Einstein GPT for Sales and Service launched with a “limited number of Einstein GPT credits, now known to be on the order of a few thousand generative outputs per month.

Salesforce Einstein Copilot (the “AI assistant” in the UI) has been described in terms of conversations (initially a flat rate per conversation in pilot phases). The key is that these limits exist, even if sales brochures don’t highlight them.

When negotiating, ask Salesforce to spell out the usage limits in writing for any AI add-on you purchase, including what counts against those limits.

Read our tips, How to Negotiate Salesforce AI Add-ons (Einstein/AI Cloud): Tips for Early Adopters.

Einstein GPT Usage Model

Einstein GPT is Salesforce’s generative AI layer across products (Sales GPT, Service GPT, etc.), and it comes with usage entitlements that you need to manage.

Typically, when you enable Einstein GPT features:

• Included Credits: Your subscription includes a certain number of Einstein GPT credits for generative AI. For example, a Sales Cloud Einstein GPT add-on might include around 8,000 credits per month for your org. In practical terms, Salesforce has indicated that it might support roughly 8,000 AI-generated sales emails monthly. Similarly, Service GPT might include ~8,000 credits, enough for about 1,000 complex case summaries (since each case resolution could use multiple credits). These numbers are examples – your contract should specify the exact credit allocation.
• Per-User or Org Allocation: While you pay the Einstein GPT fee per user (e.g. $50/user/month for Sales GPT in early pricing), the credits provided are effectively shared across the organization. This is important: if you have 100 users with the add-on, you don’t want 100 separate silos of credits that each user might hit. Ensure the credits pool together org-wide. In negotiations, confirm that a credit is a credit, no matter which user triggers it, so unused capacity by one user can benefit another.
• What Consumes a Credit: Salesforce defines what actions or outputs count as one or multiple credits. Generating an email draft might be one credit, while summarizing a long case with lots of data could eat up several credits. The complexity (and token usage behind the scenes) can affect credit burn. Press Salesforce for examples: e.g., “How many credits for a 500-word AI-generated report?” Having these examples helps you predict consumption.
• Overage Mechanics: What if you need more than the included credits? Salesforce has an overage model for Einstein GPT. After you exhaust your allotted credits, you can purchase more – often in packs or blocks of credits. For instance, Salesforce might charge a certain dollar amount per additional 100 or 1,000 credits used. In 2024–2025, an example rate floated was about $30 per 1,000 credits (i.e. $0.03 per credit), sold in preset bundles. The model could be pay-as-you-go (billed for actual extra use) or require purchasing an expansion pack upfront that provides additional credits. Understand which it is: will the system automatically continue and bill you, or do you need to proactively top up? Never assume it just stops – clarify if there is any hard cap vs. auto-billing when credits run out.
• Negotiation Leverage: Armed with this knowledge, you have a few levers to negotiate on Einstein GPT:
  • Pool Credits Org-Wide: As mentioned, insist that the included credits (and any expansion) apply to the whole org. This prevents one team from hitting a wall while others have spare capacity.
  • Secure Overage Rates: Try to lock in the overage pricing for the term of your contract. If you’re signing a 3-year deal, get a clause that additional Einstein GPT credits are at a fixed rate (or even better, at a discounted rate if you exceed certain volumes). This avoids the scenario of Salesforce raising prices on extra credits next year.
  • Overage Caps or Alerts: Negotiate for controls on overage – for example, an agreement that you will not be charged beyond a certain point without approval. Some customers negotiate a “cap” where, say, after 20% over the allotment, the system should stop or the account team must notify you to approve more spend. At a minimum, get Salesforce to agree to provide usage reports (even real-time dashboards) so you can monitor credit consumption and avoid surprises.
  • Enterprise Expansion Packs Deals: If you anticipate needing many more credits, leverage that in negotiation. For instance, “We expect to use double the included credits; we want a bulk expansion pack at a better per-credit price as part of the deal.” Salesforce may have “enterprise expansion” options that give a chunk of extra credits at a lower unit cost – but they won’t volunteer it unless you ask.

Copilot Usage Model

Salesforce’s Einstein Copilot (sometimes simply referred to as “Copilot”) is an AI assistant that enables users to interact with it in natural language to accomplish tasks across Salesforce.

Its usage model has its own twist, often centered on conversations and actions:

• Counting Conversations/Interactions: Copilot usage is typically measured by conversations – each time a user engages the AI assistant with a request. For example, if a sales rep opens Einstein Copilot and asks, “Summarize my open opportunities and draft a follow-up email to Acme Corp,” that could be one conversation (even if the AI performs multiple behind-the-scenes actions to fulfill it). Initially, Salesforce floated a pricing of about $2 per conversation during pilot programs. This flat fee per conversation was meant to simplify costs, though it raised concerns: what if an employee has dozens of quick questions daily? The costs would add up.
• Actions vs. Conversations: In 2025, Salesforce introduced a more granular model using “actions” as the unit. Each AI-driven action (e.g., retrieving data, sending an email, creating a record as directed by Copilot) consumes a set number of Flex Credits. Currently, Salesforce defines one standard Copilot/Agent action as consuming about 20 credits (with some variance if the task is extremely large, e.g., processing an unusually long input). In a conversation, the Copilot might execute several actions. So you have two ways to measure:
  • Flat per conversation: Good for simpler scenarios (caps the cost per session).
  • Per action (Flex Credits): More precise for complex, multi-step sessions (you pay more if more is done).
• Edition Differences: Is Copilot included or extra? As of 2025, Einstein Copilot isn’t just “on” by default – it’s part of the new AI capabilities Salesforce sells. Some top-tier packages might bundle a certain amount of Copilot usage. For example, Unlimited Edition+ customers initially got early access with some free usage credits to experiment. Now, Salesforce has Agentforce add-ons (their AI bundle) that include Copilot capabilities. If you buy an Agentforce unlimited add-on for users, those users get unmetered Copilot usage (internal use). Otherwise, if you stick to consumption, you’ll be paying per conversation or per action as above. Always confirm: Does your license include any Copilot usage allowances? If not, you need to budget for either the pay-per-use model or an add-on.
• Monitoring Challenges: Tracking Copilot usage can be tricky. Unlike traditional API calls, an AI conversation is a higher-level concept. Salesforce is rolling out a “Digital Wallet” and usage dashboards to help admins see how many conversations or credits have been used. In practice, early adopters found it hard to know usage in real-time. You don’t want to find out at month-end that one enthusiastic user racked up hundreds of AI conversations. Therefore, push for transparent reporting: your Salesforce rep should provide a way to monitor Copilot conversations consumed (e.g., a dashboard or at least a weekly usage email). Also, implement internal guidelines: if you notice usage spikes, investigate which teams or features are driving it. Sometimes, a poorly designed prompt or an automated process can cause Copilot to be called repeatedly. Without monitoring, that’s a silent budget killer.
• Negotiation Leverage: When dealing with Copilot’s usage terms, use these tactics:
  • Insist on Reporting: Make it a condition that Salesforce provides detailed usage metrics on Copilot. If the product doesn’t yet have a self-service dashboard, request an arrangement (even if informal) where you can ask for usage stats monthly. This gives you data to manage consumption actively.
  • Org-Level Pool for Conversations: If any “free” conversations or trial credits are included (for example, Salesforce sometimes seeds accounts with several free Copilot conversations to get started), ensure these are org-wide, not per user. You want the freedom to have one power user try 50 conversations while another only uses 5, without hitting individual caps.
  • Negotiating Bulk Conversation Packs: Similar to credits, ask if there’s a discount for committing to a certain volume of Copilot usage. Salesforce might be willing to offer a package of conversations at a flat rate (e.g., X dollars for Y conversations) instead of purely pay-as-you-go. This can give cost certainty.
  • Buffer for Overages: For Copilot, an overage could mean an unexpected invoice if usage is higher than anticipated. Try negotiating a buffer – e.g., the first 10% over your anticipated conversation count is free, to allow for some variance. Alternatively, negotiate a pilot period: for the first 2-3 months of Copilot use, you only pay a fixed amount regardless of usage, until you gauge actual demand. Use that data to adjust terms before a true-up kicks in.
  • Usage Controls: Ask Salesforce what controls exist to limit Copilot usage. If none, consider building your own policies: you might restrict Copilot availability to certain user groups initially (to prevent thousands of employees from hitting it simultaneously). Communicate to users that the AI isn’t “free” to use without purpose – encourage thoughtful use to avoid frivolous consumption. These soft controls internally can support the hard negotiation you do on the contract side.

AI Cloud & Unlimited Options

Salesforce has realized that not every customer wants to count credits or conversations endlessly.

Enter AI Cloud bundles and “unlimited” usage options.

In 2025, Salesforce introduced new SKUs under the “Agentforce” and AI Cloud branding that give more predictable pricing for AI:

• AI Cloud / Agentforce Bundles: These are packaging offers where you pay a premium to get a large bundle of AI capacity. For example, an “AI Cloud” or Agentforce 1 Edition license might include not just Einstein GPT for one cloud, but a whole set of AI features plus a huge pool of credits. One such bundle was noted to include 1 million Flex Credits per org per year, along with Data Cloud storage credits, etc. The idea is to package a lot of usage upfront. This appeals to organizations that plan to heavily use AI across the board – instead of buying a bit for sales, a bit for service, and risking overage on each, you commit to an all-in-one AI-enabled Salesforce with big limits.
• “Unlimited” Internal Usage (Per-User): The clearest offering is the new Agentforce Add-On (starting around $125 per user/month), which provides unmetered generative AI usage for that user. Essentially, if you attach this to, say, your Sales Cloud user license, that user can use Einstein GPT and Copilot features without worrying about credit consumption. It’s truly unlimited for internal use by that user (subject to fair use policies, of course – you can’t script a single user to hammer the system 24/7, but normal heavy use is fine). This is ideal for power users or roles that rely heavily on AI assistance throughout the day. There’s a hefty premium, but it buys peace of mind and cost predictability. No matter how many emails Joe Salesrep has the AI draft or how many cases Jane Agent gets auto-summarized, your cost doesn’t increase beyond the fixed fee.
• Premium vs. Pay-as-You-Go: The trade-off between these unlimited/premium options and pay-as-you-go is a classic one between predictability and cost efficiency. If your AI usage will be high (or you simply cannot afford any service disruption from hitting a cap), the unlimited model is worth it. You’ll pay significantly more per user upfront, but you’ve transferred the risk to Salesforce – they have to accommodate your heavy usage. On the other hand, if usage is uncertain or low, paying per credit or per conversation can be far cheaper. Many organizations will end up with a mix: perhaps a core group of users on unlimited licenses (e.g., your support team leads or sales power users) and the rest on a pooled credit model that you monitor.
• Testing the Waters: Negotiation can help here too. We recommend negotiating a pilot period or benchmark test with these models. For example, ask Salesforce for a one-month trial of unlimited AI usage for a subset of users (or a temporary oversize credit allotment). Use that to measure how many credits you would have consumed. This data lets you calculate the break-even point for unlimited. Maybe you discover that one sales user would have consumed $300 worth of credits in a month – giving them a $125 unlimited license would actually save money. Or maybe they only would have used $10 worth – in which case unlimited would be overkill. Use pilot results to choose the right model.
• Negotiation Angles for Unlimited/AI Cloud SKUs: When discussing these high-end options:
  • Volume Discounts: If you want to roll out unlimited AI to, say, 500 users, Salesforce may give a discount off that $125 list price. Leverage the commitment – “We’re willing to invest in 500 unlimited AI licenses if the price per user comes down to $100.” Large enterprise deals often have such flexibility.
  • Swap Rights: Salesforce’s new Flex Agreement concept allows swapping between user licenses and credit consumption. Ensure your contract lets you switch some users from unlimited to metered or vice versa after a period if needed. This way, you’re not locked in if usage patterns change. For example, get a clause that says after Year 1, you can convert some of your unused unlimited licenses into a pool of flex credits instead (or convert high-consuming credit users into unlimited licenses) without penalty.
  • All-You-Can-Eat Org Deals: In rare cases, with extremely large customers, Salesforce may consider an org-wide unlimited AI usage deal for an overall price. This is essentially a flat fee for unlimited use across the entire company. It’s not standard, but if your usage is going to be massive (and you can prove it would cost far more via credits), you can propose this as a custom deal. At the very least, it might push Salesforce to offer a higher credit bundle that meets your needs for a set cost.
  • Comparing Scenarios: Show Salesforce that you’ve done your homework. Come to the table with scenarios: “In Scenario A (pay-per-use), if we have 100k AI outputs a month, we’d pay $X. In Scenario B (unlimited for 50 users), we pay $Y. We need a better middle ground or we’ll have to severely limit adoption.” This signals that you won’t blindly accept a blank check model – you need a fair, predictable outcome. Often, Salesforce will work with you to adjust limits or pricing once they see you’re rigorously analyzing it.

In short, Salesforce’s new AI Cloud pricing gives options beyond pure metered usage. Take advantage of those to craft a plan that fits your risk appetite. If cost predictability is king, lean towards unlimited or large bundled plans (with negotiation to reduce the premium).

If cost efficiency is more important and you can closely manage usage, a metered model with safeguards might suffice. Just avoid the worst case: an unlimited appetite for AI on a strictly metered plan with no safety nets.

Monitoring & Managing Usage

Regardless of the model you choose, monitoring and managing ongoing usage is crucial.

You don’t want to “set it and forget it” with AI features – treat AI usage like a utility that needs oversight:

• Salesforce Tools for Tracking: Salesforce is rolling out tools to help track AI consumption. The Digital Wallet in Salesforce setup (for those with the latest AI features) is designed to display your credit balances and usage rates. Make sure your admins know how to access and read these dashboards. If you have Einstein GPT enabled, look for any “Generative AI Usage” section in Salesforce Setup or Trust dashboards. Additionally, Salesforce may provide usage APIs or reports – ask about these. For example, “Is there an API to pull how many Einstein GPT credits we’ve used this month?” If yes, integrate that into your monitoring systems.
• Internal Usage Audits: Don’t rely solely on Salesforce’s tools. Implement internal logging if possible. Many AI actions might be recorded in Salesforce’s logs (for instance, prompt and response logs for audit purposes). Even if you can’t see “credits” in those, you can see how often users trigger AI features. Use that to spot trends: maybe one department is using 80% of all AI outputs. That might be okay if expected, but it might also indicate you need to buy more capacity for them or train them to use it more efficiently. Identify high-volume users – are they doing something repetitive that could be optimized? Sometimes, user education can reduce wasted usage (such as teaching users to refine a prompt rather than repeatedly clicking “generate”).
• Alerting: Aim to set up alerts. For example, if you have 10,000 credits a month, have an alert when 8,000 are used to trigger an email to IT or the admin team. If the Salesforce platform doesn’t provide this, do it manually: just check weekly usage and project forward. It’s better to proactively call your Salesforce rep to discuss an oncoming overage than to be blindsided by the bill.
• Throttling and Policies: Consider implementing internal throttling policies if possible. Salesforce might not have a native way for you to set a hard limit (because they’d prefer you simply pay for what you use), but you can establish policies. For instance, you might decide that the Copilot feature will only be enabled for certain user profiles initially. Or if you notice a certain AI feature isn’t mission-critical but is eating credits, you could disable it temporarily. Another idea is to rotate usage – e.g., turn on heavy AI features only for specific teams in a given quarter as a trial, then evaluate and expand. This phased rollout prevents a free-for-all that overruns your allotment.
• Monthly Usage Reports from Salesforce: As part of your contract or at least your relationship, ask Salesforce to provide a formal usage report monthly or quarterly. This report should detail your entitled usage vs. actual usage, any overages incurred, and trend lines. Not only does this keep Salesforce accountable, it also creates a paper trail in case of any disputes (“we were never told we exceeded our credits in May…”). Having regular reports means no news is good news – if a report is missing or late, you know to chase it. And if usage is trending up sharply, you have the data to make a case internally for more budget or to Salesforce for a better pricing tier.
• Best Practices to Prevent Overages: Encourage best practices among your users:
  • Train end-users on efficient AI usage. For example, if a seller is using AI to draft emails, teach them to use the feature when it truly saves time, not just to play around.
  • For developers integrating AI via APIs, enforce rate limits in code. Don’t let a buggy script call Einstein GPT 1000 times in a loop.
  • Keep an eye on new Salesforce releases – sometimes new AI features might be enabled by default and start consuming credits. Make sure you review release notes and control the enablement of any beta/pilot AI features in your org.

By actively monitoring and managing, you turn usage-based pricing into a more controllable expense rather than a runaway train. The theme is: visibility and proactive control. You want no mysteries when the bill arrives.

Negotiation Strategies for Usage-Based Pricing

Finally, let’s zoom out and summarize the key negotiation strategies when dealing with usage-based pricing in Salesforce’s AI offerings.

These apply whether you’re buying Einstein GPT, Copilot, or any AI Cloud SKU with consumption metrics:

1. Pooling Credits Across Users: Push to convert any per-user usage entitlements into a shared pool. Vendors often start with per-user limits (“each user can do 100 AI requests a month”). This almost always leads to wasted capacity and overage simultaneously – some users won’t use their share, while others will blow past theirs. Instead, negotiate language like: “The included AI credits from all licensed users will be aggregated and usable by the Customer across any authorized user.” This way, it’s the total usage that matters. Pooling ensures you actually utilize what you’re paying for and gives a cushion for heavy users. Salesforce has precedent for org-wide quotas (they often talk about org-level credits), so they may accept this if asked.
2. Overage Rate Caps: Lock in how overages will be charged and cap the rate. If today additional credits are $0.03 each, insist that this rate (or a discounted rate you negotiate) is fixed for the duration of your agreement. Additionally, consider negotiating a cap on total overage spend. For example, “We won’t be charged more than 10% of our annual license cost in any quarter for AI overages; anything beyond that triggers a true-up discussion.” This turns catastrophic overuse into a joint problem to solve, rather than a blank check. The goal is to eliminate blank-check scenarios where usage multiplies and so does your bill. Set a ceiling or at least a predictable formula.
3. Flat-Rate Alternatives: Always ask Salesforce what flat-rate or unlimited options are available, even if you initially plan on metered usage. They may have internal flexibility – e.g., “unlimited org usage for $X” or the per-user unlimited add-on. Even if you don’t opt for it, having that option quoted gives you leverage. It provides a comparison: “If we go with consumption, worst case we might pay more than the flat $X, so give us assurances we won’t, or we might as well buy the flat rate.” In some cases, you might use a hybrid: negotiate a flat fee for a certain high volume (effectively a bulk pre-pay with discount) so you reduce variable cost risk.
4. Grace Buffers and Rollover: Try to build in a grace buffer for usage. This could mean requesting that Salesforce include, say, 10-15% additional credits at no charge as a contingency. It’s a negotiating chip: “We’ll commit to this AI add-on, but we want 10% headroom on the credits so small overages don’t immediately cost us.” Another approach is negotiating rollover of unused credits (if you’re on an annual allotment). Salesforce’s standard is that unused credits expire at the end of the term, but you might get an agreement that if you under-utilize in year 1, those extras can roll into year 2’s pool. Even a limited rollover or buffer can save you money and anxiety.
5. Benchmarking and Review Clauses: Because AI usage is new and hard to predict, include a consumption review clause in your contract. For example, at the 6-month or 12-month mark, you and Salesforce will review actual usage vs. assumptions. If you are over-using or under-using, you have an opportunity to rebalance the arrangement. This could involve reallocating licenses, adjusting the model, or securing better rates on additional credits. The idea is to avoid being locked into a bad fit for multiple years. If you overestimated usage and overpaid, you should be able to scale down commitment in the next period. If you underestimated and blew through limits, Salesforce should work with you on a more cost-effective plan for the future. Baking this flexibility into the contract protects you as the buyer. It also puts Salesforce on notice that you expect partnership, not “gotcha” charges, as your actual needs unfold.

Using these strategies, you can enter an AI licensing deal with eyes wide open. Remember, Salesforce’s sales reps are prepared to negotiate on these points – usage-based pricing is new territory for many customers, and pushback is not uncommon.

By emphasizing predictability, fairness, and shared understanding of usage, you set the stage for a better long-term relationship (and fewer invoice shocks).

Visual Table – Usage Models & Levers

Below is a summary of Salesforce AI usage models and ways to negotiate each:

Table: Salesforce AI usage models and negotiation levers for each. Einstein GPT is credit-based, Copilot can be conversation-based, and AI Cloud offers unlimited usage at a premium. Each demands different tactics to manage cost.

FAQs

1. What is an “AI credit” in Salesforce?
An AI credit is essentially a unit of measure for Salesforce’s generative AI usage. Think of it as a token or point that gets consumed when you use Einstein GPT or related features. For example, if you ask Einstein to generate some text (an email draft, a chat reply, a code snippet, etc.), it will cost a certain number of credits. Salesforce uses credits so they can have a consistent way to charge for AI across different products (since one “AI action” might use varying computing resources).

In short, a credit abstracts the underlying AI workload into a simple count. One AI credit might correspond to, say, one short output or part of a larger output, depending on complexity.

When Salesforce sells you Einstein GPT capabilities, they’ll include X credits in your license – that’s your prepaid usage. Consuming credits beyond that allotment could incur extra cost. Always get clarity on what actions cost how many credits, so you know what you’re spending your credits on.

2. Do unused credits roll over?
Generally, unused AI credits do not roll over indefinitely. Salesforce’s policy (as of 2025) is typically use-them-or-lose-them within the term. If your contract or licensing term is annual, any unused credits at the end of the year expire. If they’re metered monthly, unused monthly credits likely reset each month (with the exception that some contracts treat it as a lump sum for the year).

This is why negotiating an org-wide pool or an annual pool is beneficial: it gives you more flexibility to utilize credits where needed before expiration. In some negotiations, customers have asked for a partial rollover – for example, allowing up to 10% of unused credits to carry into the next term – but that’s not standard. The safe assumption is no rollover by default.

Thus, size your purchase of credits wisely. You don’t want to vastly over-purchase credits and then watch them expire. Better to start a bit conservative and then top up if needed (unless the deal heavily favors buying in bulk). And if Salesforce is including a big chunk of “free” credits to get you started, note when those expire as well.

3. Can credits be pooled across users?
They should be – and you should make sure they are. Pooled credits mean all your included AI credits sit in one bucket that any user can draw from. Salesforce’s intent with Einstein GPT credits has been to provide them per organization (especially since only Unlimited edition customers initially received them, which was initially org-wide). However, if there’s any language indicating per-user limits (for instance, “each user gets up to N outputs”), clarify that in practice it’s an org-wide cumulative limit. In negotiation, explicitly ask: “Are the Einstein GPT credits globally shared for our org?” and get that in writing.

Pooled credits prevent a scenario where, say, user A runs out of their allotment and gets blocked or charged, while user B has unused capacity. Practically, Salesforce tracks usage at the org level for billing purposes, so pooling is usually the case. Where you might run into non-pooled situations is if you have different license types – e.g., you gave Einstein GPT access to 50 users and not to others.

Only those 50 users’ actions draw from the pool, obviously. But among those 50, it’s shared. The key is to avoid any segmentation that strands resources. If Salesforce somehow has separate pools (like separate pools for Sales GPT vs Service GPT initially), you could even ask if those can be unified at the contract level. The more fungible the credits, the better you can optimize usage.

4. How are overages charged?
Overages are typically charged in one of two ways: automatically per use, or via pre-purchased blocks. In either case, it translates to a cost per unit beyond your included amount. For example, Salesforce might set a rate such as $0.03 per additional AI credit used.

If you go 1,000 credits over, that’s $30 extra on your bill. This could show up as a line item in your next invoice (“1,000 Einstein GPT extra credits”). Some contracts require you to consciously buy an expansion pack once you approach the limit – say, you buy an extra pack of 5,000 credits for a set fee, and you can use those as needed. The trend, though, is toward more fluid consumption-based billing, meaning if you exceed limits, they’ll just tally it up and charge you at pre-agreed rates.

For Copilot conversations, overage might be charged per conversation (e.g., $2 each beyond any free quota) – you’d see a count of conversations and the total charge. Importantly, ask about throttling: Does Salesforce ever throttle or cut off the service when you hit the limit, or will it just keep serving and billing?

In most cases, they will keep the service running (they don’t want to degrade your user experience) and simply bill the excess. So, it’s up to you to monitor

5. Is unlimited AI usage available?
Yes, Salesforce now offers unlimited AI usage options – but at a premium. The main path to “unlimited” is via the Agentforce unlimited add-on license. By paying a higher fixed price per user, you get unmetered generative AI for those users. In practical terms, that means those users can use Einstein GPT, Copilot, etc., as much as they need (within normal use), and you won’t incur extra costs for their usage.

This is great for heavy users or mission-critical roles where you never want to worry about hitting a quota. There isn’t, as of 2025, a one-size-fits-all “unlimited for the whole org for free” – you do pay for the privilege. The Agentforce add-on (approximately $125/user/month, in addition to your base license) is the way to go for unlimited internal usage. It also often bundles other AI capabilities (predictive AI, pre-built agents, etc.), so you’re buying a suite of AI, not just raw unlimited GPT calls.

Outside of internal use: if you’re asking “Can we get unlimited usage for a public-facing AI (like an AI chatbot for customers)?” – that’s usually not unlimited, that would use the consumption model (e.g., flex credits or conversation charges). In those scenarios, you have to carefully estimate volume or negotiate a high-volume package.

But for your employees using Salesforce AI, you absolutely can choose an unlimited plan for predictability. Just run the math: how many credits would they likely use, and is the unlimited cost per user justified? In many cases, for power users, it will be. Also, “unlimited” is subject to Salesforce’s fair use policy, which is usually not an issue unless someone tries to intentionally abuse it (like hitting the API millions of times via automation). For normal business usage, it truly feels unlimited.

Five Expert Recommendations

To wrap up, here are five expert tips to ensure success when navigating Salesforce’s AI consumption model:

1. Demand Clear Definitions: Never accept opaque definitions of a “credit” or “request.” Insist that Salesforce clearly define what counts as a credit, a conversation, an action – whatever unit you’re billed on. Get examples and documentation. This prevents misunderstandings later. Don’t sign a deal if you’re even slightly unsure what you’re actually paying for when it comes to AI usage.
2. Maximize Pooled Resources: Pool credits org-wide to prevent waste. Ensure that all included AI capacity can be utilized by those who need it. Pooled usage cushions the usage of heavy users with the slack of lighter users. Siloed per-user limits are a recipe for some users hitting walls and others leaving value on the table. Negotiate pooling and manage it centrally for efficiency.
3. Cap and Fix Overage Costs: Cap overage rates for at least the term of the contract. If you do have to pay overages, treat it like a utility rate – lock it in. Also consider an overall cap: e.g., you won’t be charged more than 10-15% above your planned spend without renegotiation. This forces both parties to address high usage collaboratively, rather than you just getting a huge bill. Predictability is key; cap what you can.
4. Insist on Real-Time Visibility: Require real-time (or near real-time) usage reporting. Do not accept a black box. If Salesforce can’t provide self-serve dashboards yet, get a commitment for monthly reports or an API feed. The sooner you see a usage spike, the sooner you can react (whether that’s optimizing usage or buying more capacity on your terms). Surprises in data usage equal painful surprises in cost.
5. Evaluate Unlimited vs. Pay-Go Rigorously: Always model out unlimited vs. consumption before signing. Don’t go with one by default. Use your pilot data or best estimates to simulate costs in both scenarios. Often, a hybrid ends up optimal. And remember, you can revisit this – structure your agreement to allow changing course if your modeling turns out wrong. The bottom line: make an informed, numbers-driven decision on which model saves you money and headaches.

By following these recommendations, you’ll approach Salesforce’s AI offerings as a savvy negotiator and manager, ready to harness Einstein GPT and Copilot’s benefits without losing control of your budget. Salesforce wants you to adopt these AI features, and with a smart strategy, you can do so while maintaining cost predictability and leverage.

Read about our Salesforce Negotiation Services

Do you want to talk to us about our Salesforce Negotiation Services?

Please enable JavaScript in your browser to complete this form.

Name *

Email *

Company

Message *

Submit