• Insights

How to Develop An AI Pricing Model

Your guide to developing an effective and scalable AI pricing strategy.

July 28, 2025

by Derek Asbun

AI is transforming the world — but many AI companies are still struggling to capture its financial value.

In a recent survey of 614 CFOs, 71% said their company struggles to monetize AI effectively. Another 68% said their current pricing model no longer works. As AI adoption accelerates, a key challenge is emerging: how to price, budget for, and purchase GenAI solutions at scale.

Traditional SaaS pricing — refined over decades to fit enterprise procurement and budgeting cycles — doesn’t map cleanly to the new economics of AI. Where businesses are used to predictable, seat-based pricing, AI costs are driven by factors like usage, compute requirements, and model improvements. This creates a mismatch: customer expenses are less predictable and less cyclical, complicating things for both the startups building AI products and the enterprises buying them.

There’s no “right” or “wrong” when it comes to AI pricing. The playbook is being written in real time — and we’d like to contribute our perspective. After speaking with numerous sales and product leaders across Salesforce and our portfolio, here’s our framework for identifying and implementing an AI pricing strategy for your business.

Breaking Down Emerging Pricing Models

While there’s no consensus on the “best” AI pricing model, startups are actively experimenting with a variety of unique approaches. The following graphic is a summary of the various pricing models for AI products in the market today:

Here’s a breakdown of the AI pricing strategies that have gained the most traction thus far (highlighted in blue above):

User-Based Pricing (Monthly Active Users – MAU)

Concept: With a MAU approach, companies charge a fixed fee for unlimited usage by active users each month. This model aims to simplify billing and provide predictability for customers, particularly for tools where consistent engagement from a set user base is expected. The focus is on the number of people actively using the AI, rather than the volume of their individual interactions.

Best For: This model makes sense when the value derived from the AI primarily comes from the access and collaboration among a defined group of users, rather than the raw compute power or individual API calls. It’s often favored by companies that want to encourage broad adoption within an organization without users constantly worrying about exceeding usage limits. The “unlimited usage” aspect within the fixed fee can be a strong selling point. You’ll typically see productivity and workplace tools (e.g., AI copilots in sales, marketing, or customer support software) leverage this type of pricing model.

Examples:

• OpenAI’s ChatGPT Plus: While OpenAI has a more complex API pricing based on tokens, their consumer-facing ChatGPT Plus offers a flat monthly fee for individual users, providing “unlimited usage” (within reasonable bounds, and with potential for higher priority access to new features or less throttling).
• AI-powered collaboration tools: Imagine an AI assistant integrated into a project management tool. A team might pay a fixed fee per “active seat” on the platform, allowing each team member to leverage the AI for summarization, idea generation, or task automation without per-use charges.
• AI-driven design or content creation platforms: If a platform offers AI tools for graphic design or content generation, they might charge per “designer seat” or “creator seat” that actively uses the AI features within a month.

Token/Usage-Based Pricing (API Calls)

Concept: API Call-Based Pricing entails pricing based on the number of times a service is “called” or used, or how many conversations the user is driving through it. This model often extends to “token-based pricing” in the context of large language models, where a “token” is a unit of text (roughly a word or part of a word).

Best For: This is a “pay-as-you-go” model, directly tying cost to consumption. It’s highly transparent and fair for users with fluctuating or unpredictable usage. For large language models (LLMs), token-based pricing is desirable because computational cost is directly related to the amount of data processed (input tokens) and generated (output tokens). The more complex the query or the longer the response, the more tokens are consumed, and thus, the higher the cost.

Examples:

• OpenAI API (GPT-3.5, GPT-4, etc.): OpenAI charges for its various language models based on the number of tokens used for both input (prompts) and output (generated text). Different models and capabilities (e.g., higher quality, longer context window) have different per-token rates.
• Google Cloud AI Platform: Services like Google’s Natural Language API or Vision AI often charge per API call, or per unit of data processed (e.g., per 1,000 characters for text analysis, per image for image recognition).
• AI voice assistants or conversational AI platforms: These services might charge per minute of conversation, per API call for a specific intent recognition, or per message exchanged.
• AI image generation APIs: Providers might charge per image generated, with higher resolution or more complex styles incurring a higher per-image fee.

Credit-Based Pricing (Pricing Bands with Capabilities)

Concept: With Credit-Based Pricing, pricing bands are tiered, with customers paying different rates for different levels of usage or feature access (plus additional fees for overages). Customers can select a specific tier based on their anticipated level of usage and needs so that the user and company can enjoy the benefit of predictable costs / revenue.

Best For: This model provides a balance between predictability and flexibility. Customers can choose a tier that best fits their anticipated usage, often receiving a better per-unit rate at higher tiers (volume discounts). Credits act as a currency within the system, which users “spend” on various AI capabilities. The tiered structure can also be used to gate advanced features or higher performance levels to more expensive bands. Overages allow for continued service even if a customer exceeds their allocated credits, albeit at a higher per-unit rate. For these reasons, this model is often preferable to large enterprises that require predictable budgeting.

Examples:

• Cloud computing platforms (e.g., AWS, Azure, Google Cloud) for AI services: These platforms often offer “reserved instances” or “committed use discounts” for AI compute. Customers commit to a certain level of usage (e.g., a specific GPU instance for a year) and pay a discounted rate. Beyond that, they might revert to on-demand pricing, which acts as the overage.
• AI-powered analytics platforms: A company might offer different tiers based on the volume of data processed, the complexity of AI models that can be run, or the number of advanced analytical features available. Each tier could come with a certain number of “credits” for AI processing, and additional credits can be purchased if needed.
• AI content generation suites: A platform might have a “Basic” tier with 10,000 credits per month for text generation, a “Pro” tier with 50,000 credits and access to image generation, and an “Enterprise” tier with unlimited credits and dedicated support. Overages could be charged per 1,000 tokens or per image if the monthly credits are exceeded.
• Airtable or similar platforms with AI add-ons: These platforms often provide a certain number of AI “credits” included in their higher-tier plans, which can be used for AI features like data categorization or content summarization. Users can then purchase additional credit packs if they exhaust their allowance.

Outcome-Based Pricing

Concept: When pricing is outcome-based, companies charge based on the results achieved, rather than the resources consumed.

Best For: This is perhaps the most value-aligned pricing model for AI. Instead of paying for access or usage, customers only pay when the AI delivers a measurable, predefined business outcome. This shifts the risk from the customer to the AI provider, as the provider only earns revenue if their AI successfully achieves the desired result. Outcome-Based Pricing requires strong alignment between the AI provider and the customer on what constitutes a “successful outcome” and how it will be measured. Mature AI solutions tied to clear business KPIs (e.g., fraud detection, revenue lift, cost savings) may leverage this model.

Examples:

• AI-powered customer support bots: Instead of charging per conversation or per API call, a company might charge for every customer support ticket successfully resolved by the AI without human intervention. Zendesk and Intercom have explored this approach by charging per “successful resolution” by their AI chatbots.
• AI fraud detection systems: A financial institution might pay an AI provider a percentage of the fraudulent transactions successfully prevented by the AI, or a fixed fee per confirmed fraudulent activity detected.
• AI-driven sales lead qualification: A sales team might pay for every qualified lead generated or every sales opportunity converted by the AI, rather than paying for the number of leads processed by the AI.
• AI in healthcare (e.g., diagnostic tools): A hospital might pay for each accurate diagnosis provided by an AI system, or for each reduction in misdiagnosis rates attributable to the AI.
• AI marketing campaign optimization: A marketing agency might pay a percentage of the increased conversion rate or revenue generated by an AI-optimized advertising campaign.

These emerging models reflect a growing sophistication in how AI value is perceived and monetized, moving from simply providing access to aligning costs with actual value.

Migrating Customers From Traditional SaaS Pricing to AI-Native Pricing Models

Finding the right pricing model is only the first step — convincing your customers to migrate to your new AI pricing model is a challenge in and of itself.  This is because AI adoption isn’t just a pricing challenge — it’s a change management challenge. Unlocking AI’s promised efficiencies often requires companies to reimagine workflows, reconfigure team structures, and overcome internal resistance.

Despite growing interest, most enterprises still seek low-risk entry points and hands-on guidance from vendors. Startups must design pricing models and sales motions that enable phased adoption, align with existing enterprise frameworks, and focus on tangible customer success.

With this in mind, here’s our five-step plan for companies trying to move their customers to a new AI-first pricing model. 

1. Audit Your Current Pricing and Customer Contracts

Start by mapping out all your existing pricing structures, contract terms, and renewal timelines. Identify which customers are on long-term plans, and what percentage of revenue is tied to these longer-term contracts that may require special handling. Customers on shorter-term contracts could be candidates for early migration.

2. Stress-Test Your AI Pricing Model

Run internal financial modeling to ensure your new pricing structure aligns with your revenue and margin goals. Then pilot with select customers (i.e., those who are already power users of your AI features) on shorter-term engagements to collect feedback. Also ensure your internal systems (billing, CRM, finance) are ready to support your new model. Token- and credit-based models often require bespoke tooling.

3. Build Comms and Change Management Plans

When moving to a new pricing model, transparency is key. You need to help your customers justify their AI investment to colleagues. Prepare clear messaging: why you’re changing pricing, how it benefits them, and how they’ll be affected. It’s important to position the change as aligning with value delivered, rather than simply as a price increase. Provide comprehensive details (i.e., an FAQ page, blog post, webinar, etc.) as well as a direct contact high-profile accounts can reach out to for more information.

4. Offer Transition Paths

Customers won’t respond well to an abrupt change in pricing. Offer to grandfather customers into your new plan once their current plan expires, or provide them with incentives (e.g., discounts, additional services) for early migration. For key accounts, you may need to offer a bespoke pricing package for a time to ease them into the new model. Minimize friction by showing customers how your new AI tooling will fit into their existing workflows and IT budgets.

5. Monitor Customer Behavior and Revenue Impact Closely

Once customers have migrated to your new pricing model, it’s important to track key performance metrics closely: revenue growth, churn, and margin impact. Based on these metrics, teams should be ready to iterate quickly as data and customer feedback rolls in. Startups should also plan for how their value metrics will evolve with customers’ AI maturity. Some businesses may seek to establish a special team within pricing, sales, or customer success to manage migration and surface learnings quickly.

A thoughtful pricing policy should evolve with your customers and prioritize their success. Including professional services in contracts, offering flexible credit models, or implementing more forgiving overage terms can make it easier for customers to scale as their AI adoption deepens. Some companies also take a land-and-expand approach — intentionally underpricing early to gain market share, then transitioning customers to a more sustainable model as value is proven.

Planning For Long-Term Success

When it comes to AI pricing, playing the long game is essential. Ask yourself the tough questions: Are you charging for something customers will eventually expect as table stakes? Are you factoring in the support needed to help them succeed and scale?

While the playbook for AI pricing is being written in real time, there are still a handful of tried and true best practices that all startups migrating to AI pricing should keep in mind:

1. Account for real costs, including development, infrastructure, and ongoing model operations.
2. Be simple and transparent — pricing should be consistently communicated across your website, marketing, and sales materials.
3. Balance customer acquisition with long-term margins, ensuring early adoption doesn’t undercut sustainability.
4. Map to enterprise budgeting frameworks, so buyers can justify and plan for their investment.
5. Stay flexible, evolving as costs (e.g., compute) and AI capabilities change.
6. Emphasize domain-specific value, especially in verticals where specialized knowledge creates durable differentiation.

There’s no one-size-fits-all model, and few examples of teams with years of scaled success to draw from. But the factors explored in this article offer a useful foundation for any company shaping its AI pricing strategy. The “right” model for your business will balance development costs, perceived customer value, and enterprise buying behavior, while staying adaptable in a fast-changing market.

_

Salesforce Ventures regularly hosts workshops and discussions for startup founders and teams. Sign up for our email newsletter to learn of upcoming events in your area.