👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.

2025 will be remembered as the year when seemingly everybody lost confidence in their pricing. Among the top 500 players in SaaS and AI with transparent pricing, there were more than 1,800 pricing changes in 2025 alone. That’s a staggering 3.6 per company.

I enlisted help from my friend Rob Litterst, co-founder of pricing intelligence platform PricingSaaS, to give readers a tactical rundown of what happened and what (if anything) seems to be working in SaaS pricing. You can go deeper via their new pricing change log, and stay tuned for a full 2025 retrospective later this month.

Today’s edition is free. Subscribe or upgrade to support Growth Unhinged.

Pricing changes used to be big milestones. Every few years, SaaS companies would pony up six-figures on outside consultants, run willingness-to-pay studies, collect usability feedback, and build complicated financial models before rolling out new pricing. 

And the big pricing change? Probably a slight tweak to their Good-Better-Best model with a 10% price increase.

And now: pricing changes have begun to feel like everyday occurrences. 

At PricingSaaS we closely monitor pricing and packaging changes across 500 of the top players in SaaS, and have observed more than 1,800 changes in 2025 alone. 

Take Lovable, the vibecoding darling which recently hit $200M ARR (four months after reaching $100M ARR). They made roughly one meaningful pricing update per month. Among the changes: they launched a Team plan (March), killed the team plan (June), added rollover credits (August), and tweaked starting price and credit limits for Pro and Business (August and September). Even a SaaS behemoth like Salesforce launched AI credits (May), flexible payment options (August), a new user license option (September), and now an agentic ELA (October). 

Of those 1,800 pricing and packaging changes:

Across these changes, we’ve got our eyes on four trends that defined SaaS in 2025, and will continue to shape how AI business models evolve in 2026. For each trend I’ll share real-life examples and guidance to help you evolve your own pricing.

If there's a single pricing trend that defines 2025, it's credits. Out of 500 companies in the PricingSaaS 500 Index, 79 now offer a credit model, up from 35 at the end of 2024 (this is up 126% YoY). Among the new additions are household names like Figma, HubSpot, and Salesforce.

Credits help vendors and customers manage AI economics. They give customers the predictability of a license, while giving vendors a usage component to ensure margins stay intact at scale.

Credits sit in the middle of the spectrum between charging for access and charging for outcomes. They’re more transparent than legacy per-seat licenses (you can see what actions are being taken), but easier to implement and measure than pure outcomes. 

All that said, credits can get complicated quickly. Before implementing a credit model, there are four core questions to think through:

1. What is a credit worth and how does that change at scale?

The simplest approach is dollar-credit parity. Lindy does this cleanly: $50/month gets you 5,000 credits at roughly $1 per 100 credits. It's dead simple for customers to do the math.

From there, most companies introduce volume discounts. Clay offers 2,000 credits for $134 ($.06/credit) versus 50,000 credits for $720 ($.01/credit).

2. What can a credit buy?

Some credit systems are beautifully simple. Take Audible: one credit equals one audiobook. Most SaaS products are more complex.

Clay credits can purchase company revenue lookups, LinkedIn scrapes, email enrichments, and dozens of other actions, each at different credit costs. Lovable credits can make simple design tweaks, add authentication, and build a full landing page, each at different costs as well.

This complexity is a good thing for SaaS vendors, giving you more flexibility and pricing leverage. On the other hand, this can get complex (and annoying) for customers.

3. Where does the credit fall on the value chain?

Credits can map to different points between inputs and outcomes.

Fireworks AI sits at the beginning of the value chain, you're essentially buying LLM infrastructure (tokens). While a company like Reply sits further down. You're paying for contacts actively being nurtured, which sits closer to the business outcome.

Where your credits land on this spectrum affects perceived value and how customers evaluate ROI. The closer to outcomes, the easier the value case, but the more you're on the hook for results. 

4. Is your credit system customer-friendly or vendor-friendly?

This helps shape how your customers use credits. Whether credits rollover, if they pool at the user or org level, and how additional credit purchases are handled can determine how customer-friendly your credit model is. 

A framework for implementing credits

Across the PricingSaaS database, we’re observing a range of credit implementations although most fall into one of the following four buckets. The buckets are based on two variables: (1) how flexible the credit policy is for customers (e.g., rollovers, account-level pooling, top-ups) and (2) whether credits are cost-based (to protect margins) or value-based (to monetize the work delivered by AI). The key driver of cost-based vs value-based is whether the company explicitly charges different prices for different providers, or abstracts the LLM layer and focuses on actions and outcomes. 

Top-left: AI credits as product feature (Value Based, Vendor-Friendly) 

Companies in this quadrant price AI around outcomes and tightly control usage to protect margins. Credits are abstracted away from models, reset frequently, and don’t rollover. This is common among legacy SaaS workflow tools bundling AI into core plans (e.g. Adobe, HubSpot). The goal is to increase product stickiness while avoiding runaway costs. 

HubSpot uses a credit-model to charge for AI features. Core plans come with a bank of credits that reset every month, and more credits can be purchased as needed. HubSpot credit values cost between 10 and 100 credits depending on how much work is being done on the user’s behalf. This gives HubSpot a growth lever outside of seats and contacts without disrupting its existing model.

Adobe falls in this quadrant for Firefly, its in-house AI offering, but they also allow customers to access third-party providers. In that scenario, they act as a marketplace, and pass along the costs from each provider. 

Top-right: AI credits as a creative sandbox (Value-Based, Customer-Friendly)

These companies aggressively abstract cost and maximize flexibility to drive usage, experimentation, and habit formation. Credits are pooled, loosely governed, and framed around what users can build, not what it costs.

This quadrant is dominated by AI-native or prosumer platforms (e.g. Lovable, Replit) optimizing for growth over margin discipline. The bet: higher engagement and retention will outweigh cost volatility. This works best when AI is the primary product value, not a supporting feature.

Bottom-left: AI credits as enterprise infra (Cost-Based, Vendor-Friendly)

Here, credits closely track underlying costs and are tightly governed through caps, add-ons, and admin controls. These companies generally pass through costs from the LLMs, and utilize governance to protect margins and control usage. This quadrant is a fit for developer tools and Enterprise platforms that face high variance usage. 

Cursor credits are charged at API rates as part of the monthly seat price. The seat-based model is easy to digest, and the aggressive pricing on credits means Cursor is removing friction to drive usage. While Cursor allows users to enable PAYG billing if they exceed their usage limits, they don’t allow credit rollovers, which has been a source of frustration for customers. 

Bottom-right: AI credits as a pass-through utility (Cost-Based, Customer-Friendly)

In this quadrant, credits often map to vendors, actions, or data sources, making pricing transparent but unforgiving. These companies accept usage variability, but push cost awareness onto customers. 

PostHog uses credits for PostHog AI, and offers $20 of free credits every month for customers to explore PostHog’s AI features, then offer both PAYG and pre-paid plans for larger credit commitments. They’re transparent about charging a 20% markup on LLM costs, and allow customers with a pre-paid plan to rollover half of their remaining credits to the next contract.

Clay’s entire pricing model is based around credits, with credit value tightly correlated with costs from the underlying data provider. They offer credit rollovers, account-level pooling, and top-ups when customers need more usage.

The Bottom Line

The more credit models flood the marketplace, the more customers will want to return to simplicity. In 2025 the pendulum swung toward credits. In 2026 it'll likely swing back toward simplicity and predictability.

Credit-based pricing was the main headline of last year, but it also helped propel a quieter AI pricing trend: the great re-bundling of AI. 

Some of the biggest names in SaaS launched their first AI product as an add-on, and have since bundled that functionality into their core pricing model. We’re observing companies execute this on two tracks. 

Track A: Bundle AI into existing plans as a feature, often with a price increase. 

Track B: Bundle AI into existing plans using credit limits.

On the first track are Notion, Slack, and Loom. Each had charged between $4 to $10 per user for an AI add-on. Now they're bundling AI and raising prices by between $2.50 and $5 per user.

Notion: Bundled Notion AI and raised prices on the Business tier

Notion started with an $8/user/mo AI add-on, then bundled it into the Business plan, raising the price from $15/user/mo to $20/user/mo. They've also introduced custom agents with nascent monetization (reportedly driving $2-5 per user according to Vendr data). 

Slack: Added AI features to all plans and raised prices on the Business tier

Slack started with a $10/user/mo AI add-on that was only available for Business users. Since, they have bundled it into the core product, offering differentiated features in each plan. The AI value ladder gives Slack a stronger AI narrative, and gives customers a natural upgrade path as they need more capabilities. 

Loom: Introduced a Business+AI tier and raised prices. 

Loom started with a $4/user/mo AI add-on before creating a dedicated plan called Business+AI, which they priced higher than the combined cost of the original Business plan and the add-on. This gives Loom an intermediate plan between Business and Enterprise, and a clear narrative around which users get the most value out of Loom AI.

On the second track are Airtable and Atlassian. Each had charged between $6 to $20 per user for an AI add-on. Now they're bundling a limited amount of AI credits while holding prices the same, opening up the opportunity for consumption revenue down the road.

Airtable: Bundled Airtable AI into its core plans at no additional cost. 

Airtable started with a $6/user/mo AI add-on then bundled it into their core plans at no additional cost. Each plan has a defined credit allowance, but users can purchase “Credit Packs” when they run out. While they didn’t raise prices, Airtable AI credits give them a consumption pricing element on top of their existing license model. 

Atlassian: Bundled Rovo into its product suite with unenforced credit limits.

Atlassian started with a $20/month AI add-on called Rovo, but ended up bundling it into its product suite as an AI feature at no additional cost. Each tier of Jira and Confluence comes with a Rovo credit allotment, but interestingly, Atlassian is not currently enforcing the credit limits. They have made it clear that won’t be the case forever, and that they will give 90-days notice before enforcing quotas.

The Bottom Line

Both tracks show how SaaS companies are evolving their pricing models to absorb AI features, but they also show how much the structure of these models remain the same. Each example above fits AI features within a seat-based model, a pricing strategy many pricing pundits pronounced dead years ago.

I must have read a hundred obituaries for seat-based pricing in 2025, but when I look at our data – I still see ‘em. As companies introduce more complex pricing models, seats look customer-friendly by comparison. In conversations with pricing leaders, I’m already hearing that customers are actively asking to pay for seats instead of another credit model. 

But seats are under pressure for real reasons. As companies rely less on headcount for growth, seat-based models offer less expansion opportunity. Perhaps the bigger reason is that seats are an input, and AI has made it easier to charge for outcomes.

That said, familiarity matters. Customers expect to pay per user for software. Their budgets are already tied to seats, and the unpredictability of new models is still scary. For many companies, the answer isn't to abandon seats. It's to make them work harder.

Companies doing this well are layering additional value dimensions on top of the seat. Instead of a seat being just permission to log in, a seat becomes a gateway to more product capability, more automation, more consumption, more outcomes, or more flexibility. This maintains the predictability buyers love while creating compelling upgrade paths tied to actual value delivered.

Here are five ways we’re seeing it done:

1. Seats unlock product power (OpenAI)

OpenAI charges for seats, but each tier provides fundamentally more capability. Their differentiation includes model quality, response times, context windows, and features like deep research. With each seat upgrade, users gain more power at their disposal. This works especially well when your product has a wide range of use cases and user sophistication. A casual user and a power user can both have seats, but they're getting meaningfully different products.

2. Seats unlock product access (Figma)

Figma introduced different types of seats, with each seat granting access to different products:

These seat types work across any of Figma's plans (Professional, Organization, Enterprise), with scaling costs for each seat type within each plan. This maps pricing to actual job function, and considering Figma’s mission to make design accessible to all, it makes sense.

3. Seats unlock scale (Monday)

Monday weaves automation and integration limits into their seat-based tiers. With each tier upgrade, users get more automation and integration capacity. This is the classic hybrid model where seat licenses are blended with usage metrics, but leaning closer to outcomes. It also creates a natural expansion trigger. As customers automate more workflows and connect more tools, they hit limits that push them toward higher tiers. 

4. Seats unlock add-ons (Chili Piper)

Chili Piper offers "in-plan add-ons" within their seat-based model. Customers pay a per-user fee for access to Chili Piper, but buying add-ons like Concierge Live and Handoff Live enables Chili Piper to do jobs on their behalf.

This strategy increases average contract value while pushing Chili Piper's monetization further down the value chain from inputs to outcomes. You're no longer just paying for the tool. You're paying for work delivered.

5. Seats can be traded for AI credits (Salesforce)

Back in May, Salesforce announced Flex Credits, and a new Flex Agreement that allows customers to exchange unused seat licenses for AI credits, and vice versa.

This strategy gives customers a choice and allows them to move at their own pace. Salesforce is acknowledging that buyers want flexibility to rebalance spend between human and digital labor as the value of AI becomes clearer. Optionality itself is becoming a feature, especially in categories where long-term commitments once left customers little room to adapt.

The demand for optionality is reshaping how SaaS companies think about pricing more broadly. For the past decade, vendors held most of the leverage: long-term contracts, rigid license models, and extractive pricing practices were staples of the legacy playbook. 

AI breaks that model. It doesn’t fit cleanly into licenses or pure consumption, and while credits are rising in popularity, they increasingly feel transitory. The popular thought is that AI should be charged based on outcomes, but the reality is that outcome-based pricing is easier said than done. The result? No one really knows what to do. Most are just trying to keep customers happy while they figure it out. 

That means customers have options. And they know it. Smart companies are realizing this, and offering multiple buying models to accommodate different customer preferences.

The best place to see this dynamic playing out is in the Customer Support category, which is probably the furthest along the road to outcome-based pricing.

Take Salesforce. They might be the most dominant company in enterprise SaaS, but have made a concerted effort to introduce more pricing flexibility to meet buyers where they are. When a company with that much market power has to bend, the rest of us should take notice. Rather than forcing everyone into a single model for Agentforce, they're offering a range of options:

By offering all five of these options (and more), Salesforce removes a major objection from nearly every buying conversation. The recent additions of license-based pricing and the Unlimited ELA suggest many Salesforce customers were looking for a more familiar way to get started with Agentforce.

Even further down this path is Decagon, which exclusively offers outcome-based pricing, but lets customers choose the outcome. Customers can choose between paying per resolution or per conversation.

Neither model is inherently better. Customers can self-select into the model that matches their confidence level and use case, which reduces friction and speeds up the buying process.

The Bottom Line

In a world where AI is changing what's possible every few months, customers want optionality.