Four Perspectives on Credit-Based Pricing for AI Agents — valueIQ

Credit-based pricing has become the dominant monetization model for AI agents. Agents execute complex, multi-step tasks with highly variable cost and duration — traditional per-seat or flat-subscription models simply don't map to that reality. Credits offer a flexible unit of account that absorbs the variability while keeping billing legible to customers.

But the shift to credits creates new problems: unpredictable costs for buyers, communication failures during transitions, and the temptation to engineer opacity into systems that should be transparent.

We recently hosted Brandon Hickie, Senior Director of Monetization Strategy at LinkedIn and Venture Partner at Companyon Ventures, for a deep dive into credit models and the critical questions agent monetization teams are wrestling with right now.

Questions on the agenda

In scope

What exactly is an "agent" for pricing purposes?
What is credit-based pricing, structurally?
Why are companies moving to credits for agents?
Who is executing credit pricing well?
What design risks should teams anticipate?
What can go wrong?
Will credits interoperate across companies?
What might happen next in the agent economy?

Four key perspectives

To build a complete picture, here are four essential reads — each contributing a distinct analytical lens on credit-based pricing for AI agents.

Why Everyone's Switching to AI Credits

Kyle Poyar · Growth Unhinged

Guide to the Design of Credit-Based Pricing Models

valueIQ Pricing Intelligence

2025 Playbook for Pricing AI Agents

Chargebee

2025 Field Report from Leading SaaS Teams

Metronome

Summary across all four reports

Key themes · synthesis of four reports

Theme	Growth Unhinged	valueIQ	Chargebee	Metronome
Credits as bridge strategy	✓ Primary thesis	✓ Transition model	✓ Dominant pattern	✓ Field confirmed
Output-based over tokens	✓ Strong emphasis	✓ Design principle	✓ Buyer preference	~ Mixed findings
Transparency & dashboards	✓ Highlighted	✓ Recommended	✓ Table stakes	✓ Top request
Hybrid multi-axis models	✓ Key finding	✓ Structural rec.	✓ Case validated	✓ Enterprise norm
Sales enablement / value narrative	~ Mentioned	✓ Core focus	✓ GTM section	~ Operational
Baseline credits reduce friction	✓ "Batteries included"	✓ Recommended	✓ All case studies	✓ Enterprise default

Top insights

Credit-based models are spreading rapidly, with most companies treating credits as a bridge strategy before moving to clearer, more value-aligned pricing.
Customers prefer output or outcome-based credits (easy to forecast) over raw consumption units like tokens — tokens are an internal cost metric, not a customer value metric.
Unpredictable costs and lack of transparency are the leading adoption obstacles. Clear dashboards and detailed usage metering are strongly recommended across all four reports.
Successful vendors combine credits with other axes (subscriptions, features) and enable sales teams with strong value narratives that can answer "what did I actually get?"

"The best credit models don't just measure consumption — they confirm value delivered."

Six case studies

Case studies bring the key motivators, opportunities, and challenges into focus. Across these six implementations, a clear pattern emerges: credits succeed when they're legible, predictable, and tied to outcomes. They fail when they obscure cost or surprise customers.

Salesforce Agentforce

Successful pivot

The challenge

Salesforce originally charged $2 per conversation for Agentforce. Customers complained that complex conversations involving multiple actions created unpredictable costs and poor value alignment — a deal-breaker for enterprise procurement.

Key design principles applied

Output-based pricing: Credits consumed only when agents complete actual work — updating records, resolving cases
Transparency: Digital Wallet provides detailed usage analytics and forecasting tools
Baseline inclusion: Enterprise customers receive 100,000 credits free with Salesforce Foundations

Result: In May 2025, Flex Credits launched at $0.10 per action (20 credits), sold in packs of 100,000 for $500. Most use cases now cost 10–30 cents per interaction versus the previous $2 flat rate — reducing barriers to AI adoption while directly tying costs to business outcomes.

Cursor

Cautionary tale

What happened

In June 2025, Cursor transitioned from 500 fast requests per month to $20 worth of API credits — without clear communication. Heavy users ran out of credits within days, triggering surprise overage charges and widespread complaints on social media.

What went wrong

No transparency: Users discovered the change only after hitting unexpected usage limits
No predictability tools: No way to estimate credit consumption before taking actions
Poor change management: Conflicting messaging across blog posts and pricing pages

Recovery: CEO Michael Truell issued a public apology and offered refunds, acknowledging "we didn't handle this pricing rollout well." This case illustrates the importance of transparent communication and predictability tools that all four reports emphasize.

Clay

Multi-axis model

The model

Clay combines subscription tiers with credit-based usage. Credits cost $16–75 per 1,000 depending on plan level — creating natural upgrade economics where heavier users get better per-credit pricing.

Design features

Baseline credits (2,000 to 50,000+) included in each tier
Credit rollover to prevent "use it or lose it" frustration
Clear credit costs for each action (1–25 credits per data enrichment)
Multi-axis: subscription features layered with usage credits

Result: Clay scales from small teams to enterprise while maintaining predictable costs. The multi-axis approach prevents commoditization — customers aren't just buying credits, they're buying capabilities that justify higher tiers.

HubSpot Breeze

Expanding access

Strategic approach

HubSpot introduced credit-based pricing for Breeze Customer Agent in June 2025, expanding access from Service Hub to all Pro/Enterprise customers — using credits as an adoption vehicle rather than a revenue gate.

Design elements

3,000 credits/month for Pro; 5,000 for Enterprise
Credits tied to specific customer interactions and enrichment actions
Additional credits at $10 per 1,000 (~$0.008 per credit)

HubSpot AI credits per month · by plan Accessed Sept. 10, 2025 ↗

Monthly included credit allowances, plus add-on pack

Starter

500 / mo

Pro

3,000 / mo

Enterprise

5,000 / mo

Add-on pack

100,000 credits · $800

$800

025k50k75k100k

Note: the price per credit is essentially flat across all packages at approximately $0.008/credit ($8,700 per million credits). Compare this to OpenAI's $1.25/million input tokens and $10.00/million output tokens. HubSpot must generate several orders of magnitude of value above raw API costs — which is precisely the bet that outcome-based credit pricing makes.

Replit Agent

Checkpoint innovation

Unique approach

Replit pioneered "checkpoint-based" pricing, where credits are consumed only when agents complete meaningful work milestones — not for thinking or processing time. Originally $0.25 per checkpoint, Replit evolved to effort-based pricing where task complexity determines cost.

Key features

Outcome-based: users pay only for completed functionality, not processing time
"High power" and "extended thinking" modes with explicit cost implications
Core subscriptions include $25/month in credits (~100 checkpoints)

Why it matters: Checkpoint-based pricing is the purest expression of outcome-based billing. It raises the bar for what "output-based" actually means across the market.

OpenAI Enterprise

Credit migration

Strategic shift

OpenAI moved from fixed enterprise pricing to credit-based models for ChatGPT Enterprise, enabling businesses to scale spending based on actual usage across different models and services.

Implementation

Credits usable across different OpenAI services and models
Long-term contracts receive 10–20% volume discounts on credit purchases
Compute-heavier models consume more credits — built-in model differentiation

Market effect: OpenAI and Salesforce are doing the market education heavy lifting on credit-based pricing, making it easier for smaller companies to follow. Early movers should expect longer sales cycles and more customer education overhead.

Five themes across the case studies

Transparency is critical

Cursor's failure demonstrates the importance of usage visibility that all four reports emphasize.

Output-based pricing works better

Salesforce's success with action-based credits validates the shift away from conversation-based models.

Hybrid models dominate

Clay, HubSpot, and others combine credits with other pricing axes — confirming the hybrid approach all four reports recommend.

Baseline inclusion reduces friction

Every successful implementation includes baseline credits to enable habit formation before customers must buy more.

Predictability trumps precision

Companies with clear usage forecasting and spend controls see better adoption than those optimizing for cost-plus precision. Customers will accept slightly higher costs in exchange for budget certainty.

Ten critical lessons

Communication and transparency are make-or-break

The Cursor debacle is the most instructive failure. Pricing changes in credit models are disproportionately risky — power users are both the most exposed and the most vocal.

Provide 30–60 days advance notice with clear migration paths
Build transparent usage dashboards before any pricing shift goes live
"Credit-based pricing sounds great on paper... but the execution is almost always messy, unpredictable, and — if you're not careful — a trust killer"

Predictability beats precision

Salesforce's success came from clear action-based pricing ($0.10 per completed task) rather than complex conversation models. Clay's hybrid approach works because users know exactly what each action costs before they take it. Customers prefer slightly higher but predictable costs over variable pricing that creates budget uncertainty.

Baseline credits reduce adoption friction

The "batteries included" approach prevents users from feeling nickeled-and-dimed in their first month. Every successful implementation includes meaningful credits in base plans:

HubSpot: 3,000–5,000 credits/month in Pro/Enterprise
Salesforce: 100,000 credits free with Salesforce Foundations
Clay: 2,000–50,000+ credits depending on tier

Output-based pricing outperforms cost-plus models

Tying credits to successful outcomes shifts the framing from "cost center" to "investment with measurable return." Replit's checkpoint model charges only for completed work milestones — not thinking time. Salesforce Agentforce credits are consumed only when agents complete actual business actions.

Multi-axis pricing prevents commoditization

Pure credit models risk margin compression. Clay combines subscription tiers with credits, preventing direct price comparison. HubSpot uses credits as expansion revenue on top of existing subscription relationships. Layer credits onto other value drivers — features, seats, enterprise capabilities — rather than making credits the sole pricing axis.

Real-time usage visibility is table stakes

B2B customers increasingly demand transparent usage tracking. Salesforce Digital Wallet provides detailed consumption analytics and forecasting. Clay shows clear credit costs before actions are taken. Cursor's failure partially stemmed from users having zero visibility into consumption patterns before hitting limits.

Credit rollover and flexibility drive retention

Unused credit policies significantly impact customer behavior. Credit rollover reduces hoarding and increases actual product engagement. Annual vs. monthly limits give customers flexibility to adapt to seasonal usage patterns. Punitive expiration policies create negative experiences and reduce platform engagement.

Sales enablement requires new playbooks

Credit-based models create unique GTM challenges. Sales teams must help customers forecast credit consumption and map it to business value. Common objections: "how do we budget for this?" and "what happens if we run out mid-month?" Successful implementations require extensive training, customer success playbooks, and clear value narrative frameworks.

Infrastructure complexity is often underestimated

Operational challenges extend far beyond pricing strategy. Real-time metering requires robust data pipelines handling billions of usage events. Credit rollover, volume discounts, and custom enterprise terms create intricate billing logic. Plan for 6–12 month implementation timelines for enterprise-ready credit systems.

Market education takes time

OpenAI and Salesforce are doing the market education heavy lifting, making it easier for smaller companies to adopt similar models. But customer procurement teams remain skeptical of unpredictable spending models. Early movers should expect longer sales cycles and design their pricing communications accordingly.

A framework for success

The most successful companies treat credit-based pricing as a bridge strategy rather than an end state. Here's what that looks like in practice.

1
Start with output-based credits
Define your credit unit around a clear customer outcome — a record enriched, a case resolved, a document generated. Avoid tokens or compute-hours. Customers care about results, not your infrastructure costs.
2
Invest in transparency
Build robust usage dashboards and forecasting tools before launch, not after. Customers need to see credit burn, projected depletion, and historical patterns — not just a number counting down.
3
Layer value — combine credits with other axes
Credits alone are too easy to commoditize. Combine them with subscription tiers, feature access, or support levels. The full package creates durable willingness to pay; pure credits do not.
4
Plan for evolution from day one
AI infrastructure costs will continue to decline. Your credit model needs to absorb those changes without constant customer-facing disruption. Design pricing systems that can adapt quietly.
5
Tie credit consumption to outcomes — then prove it
The most defensible model is one where customers can clearly trace what they got for what they spent. Reporting that closes the value loop is a competitive moat, not a nice-to-have.

Four Perspectives on Credit Based Pricing for AI Agents