Usage-Based vs. Subscription: How AI Founders Are Charging

As an AI SaaS founder, I am constantly evolving our pricing strategy. Right now, we run a hybrid model that is part consumption-based and part seat-based. The challenge is aligning the value we create with the price customers pay.

Consumption metrics alone do not always reflect value, and neither does a flat seat fee. If 10 insights lead to $1M in revenue, that is worth far more than the raw count. On the other hand, 1,000 insights that only help with early pipeline may not justify a high cos

That is why we keep experimenting. AI is disrupting every industry, rewriting what customers value and how they expect to pay for it. There is no “best way” anymore. Founders must stay agile, test, learn, and adapt every day.

In my experience, usage-based pricing is often the most effective way to align price with value. The more you use, the more value you get, and the more you’re willing to pay. It’s fair, scalable, and keeps both sides honest.

That said, selling usage-based pricing as a startup isn’t always easy. For many customers, predictability still overshadows flexibility, especially in procurement-driven B2B environments. If the value isn’t crystal clear or the pricing logic feels murky, you risk scaring people off before they even start.

Make the value obvious, keep the pricing transparent, and help the customer feel in control. When done right, it’s a win-win for everyone.

AI started it, credits/tokens – but it’s affecting all of SaaS. New startups are coming out with “unlimited user” plans, charging per value-metric instead.

At Face2Face, we charge per minute of time spent-on-call. It’s not purely usage-based, there still are cliff-jumps in pricing depending on feature-sets, but the model almost completely scales depending on your talk-time. It just makes sense for everybody.

I believe SaaS pricing will change significantly with the rise of AI. Seat-based pricing will eventually disappear. I’m already seeing a shift towards a pay-per-request model, where you only pay for what you actually use, such as requests or tokens.

With AI agents, I expect fewer real users to be active in the software, which means the traditional seat-based model will fade away.

That’s why I think pricing needs to adapt to this shift. I also see a move from a fixed recurring revenue model to a more “re-occurring” one, where the average customer spend can fluctuate from month to month.

Every SaaS business will pivot to, or at least add some element of usage-based pricing over the next few years.

AI is the big driver, as the costs for providers scale with usage, so you need some way to pass that on flexibly to customers.

It may change slightly as more and better open source models bring costs way down, to the point where they are just like running any other 3rd party API, but that’ll still be a few years.

I think customers will also start to move to the mindset of only wanting to pay for what they use, not like a gym membership you pay for and never use.

Outcome-based usage pricing will also become more popular, although for many that can mean costs increase uncontrollably, it’s one of the reasons we chose to stick to a simple usage based (customer support tickets) pricing model with My AskAI.

There’ll probably also still be some base-level subscription amount charged but it will be a minority of the overall revenue for the customer.

The other model that may become more popular is also the “lifetime discount” offer.

With any movement in behavior or pricing, there will be some people who object and want to dig their heels in and go the other way, and lifetime discounts are probably the best example of that.

They’ll bring their own API keys but be able to use the service for as long or as much as they want.

But I think this will still only be 10% max of users, most will be usage-based.

AI SaaS pricing isn’t just a finance decision—it’s a behavioural trigger. The founders who win in 2025 understand they’re not selling a product, they’re selling the feeling of making the smartest decision in the room.

“Too Stupid Not to Pay”

In 2025, the smartest AI SaaS founders price for psychology: lead with ‘nice to have,’ grow into ‘need to have,’ and land at ‘simply too stupid not to have.’ That’s when your model—subscription, usage, or both, stops being a cost and starts being a brag.

When we’ve been building AI SaaS products, there’s one thing I’ve learned. Pricing is no longer a “set it and forget it” exercise.

We have seen a wave of AI products either go low in margin because they priced too simply, or lose deals because their model didn’t match how customers actually consume value.
AI economics are different. Every API call has a cost, model prices change quarterly, and customer usageis different every month. So the big question is: Subscription, usage-based, performance-based or hybrid?

Here’s how we decide:

Subscription: We still believe subscription model can be strong when predictability matters more than anything else. You also should know the usage patterns of your users VERY well, to avoid heavy users consuming far more than the average. This eats up your margins if there are no boundaries to how much compute they can use.

Usage-Based: Pure pay-as-you-go works best when you are basically an API or infrastructure. The value you provide is very predictable and know your cost of goods sold (COGS). The only real downside that we have seen is that revenue is way more volatile than with subscription-based pricing.

Preformance-Based: Only get paid for the results you bring to clients. This approach can only be effective in specific cases where you have full control over the outcome. It is hard to scale, and has some attribution challenges.

Hybrid: Most of our recent AI SaaS projects are using hybrid pricing. It means that we have a base platform fee, including some credits. (predictability for both sides). Heavy users can buy more credits. No seat administration needed. The market is not as used to it as the two models above. At PhaseOne AI, we definitely believe it is the future for the next years.

We call it ‘’personalized pricing’’, which gives a fair price adjusted to the customer’s situation. In the age of AI, we believe that everything will be customized to the client’s needs. Pricing should be no different.

I mostly work with early stage SaaS/Ai founders, so picking the ‘right’ pricing model is always an important part of GTM.

My recommendation is ‘simplicity’ over ‘perfect’ model early on. Start with pricing V1 (brings you the first 10 clients), then iterate to pricing v2 (brings you to 100 clients), then pricing v3,4…

I encourage founders to don’t overcomplicate pricing early on: find the right value metric, build 2-4 tiers, incentivize yearly upfront payments (for cashflow)… your pricing needs to validate demand and be ‘good enough’ to not stop you from growing – but it’s not about ‘the perfect model

There are 4 pricing modalities: Flat fees, Usage based, License based and Credit Systems.

Modalities basically is ‘how’ you price – where the pricing metric is ‘what’ you price (e.g. users, API calls etc.).

Usage and license are the two main modalities in B2B SaaS and they are constantly compared and evaluated against each other – because they are polar opposites on key dimensions.

Usage: the customer has a right to access the product and pays for actual usage at the end of the billing period. This lowers risk for the customer and puts cashflow at the end of the billing period, not the start.

Licenses: the customer has a right to a certain volume of usage – e.g. 5 user seats – and pays this upfront at the beginning of the billing period, regardless of what actual usage then occurs.
This model lowers risk for the SaaS vendor and pulls cashflow to the beginning of the billing period.

For these reasons – risk and cashflow – the two are very different to sell and expansion also is very different.

One is not better than the other – it’s about choosing the model that is right for your product and GTM situation.

In my work with SaaS companies, I have seen how pricing models are shifting. The old per-seat subscription approach gave providers steady revenue and customers predictable costs, but that structure is becoming harder to sustain.

Many products now require resources that vary from one customer to the next, and that is pushing more companies toward usage-based or hybrid pricing where charges reflect actual consumption. I understand the appeal.

Customers feel they pay in proportion to the value they get, and providers can better match revenue with the cost of delivering their service. Some of the fastest-growing SaaS companies I work with are the ones combining a base subscription for stability with extra usage fees for flexibility.

From a finance perspective, though, I see real challenges. Usage-based revenue is inherently variable, and that makes forecasting much harder. In my experience, only a small portion of finance leaders have the systems in place to track consumption accurately, which turns budgeting into a guessing game.

Most businesses build their budgets annually and do not plan for monthly cost swings. This creates friction between teams eager to offer flexible pricing and finance teams trying to protect cash flow. Without measures like usage caps, more predictable billing tools, or clearer communication of value and cost, these models designed to modernize pricing can end up eroding customer trust.

We see the shift towards usage-based pricing, and it totally makes sense. I think it ties back to the very beginning of Value-based Pricing with a mix of market conditions.
To add a bit of context:

• Value-based pricing is simple – you just take a fraction of the value your product delivers to your clients.
  • The more value you contribute to
  • The more tangible it is
  • The bigger fraction you can charge

Why is this important?

Delivering solutions based on usage provides tangible value with a high contribution (higher than just access).

In the past, companies were pushing towards subscription because it was scalable and predictable.

In the world of AI, where building a product is less complex, competition is fierce, and costs play a role, companies are looking for options that create better margins & growth.

Usage → Higher price → More margin → Part of the story behind usage-based pricing trends

Nevertheless, an important part of growth is friction, it’s no different this time.

Founders frequently fear:

• Unpredictability
  • Less conversion
  • More churn

Well, I think it’s not that different from other usage models we can see beyond technology – e.g., telco, taxes, or energy/water bills.

What to Apply?

• Amazing pricing and packeging (tailored to ICP, use-case, data-driven!)
  • True-ups (in case you want to sell usage within subscription limits)
  • Fair trade usage policies
  • Value conversations / Value proofs

So, I think the tech world is coming back to normal, where different models are applied rather than access-only. Usage is not for everyone, but there are definitely more usage-use cases than we’ve seen in the past.

In SaaS, we’ve historically obsessed over acquisition, sign-ups, trials, top-of-funnel numbers. But the real levers have always been activation and retention. The model only works if you can sustain and grow your existing customer base. Everyone knows it’s far easier to retain and upsell than to replace a lost customer.

AI has changed the equation. SaaS is becoming commoditized, which forces founders into two very different strategies:

– Defend with big tickets, build a strong moat and “protect” your product with larger annual subscriptions.
– Go full-on land grab, raise heavily and grow like Uber, as we’re seeing with Lovable, moving fast and aggressively to capture the market.

Pricing debates, usage-based vs. subscription, often miss the point: you can blend them. A “usage-based” plan with a minimum monthly commitment is basically a subscription in disguise. The real trade-off is speed vs. stability. Usage-based can drive rapid expansion, but often with higher churn.

In the end, there’s no universal truth. You have to factor in your ICP’s spend behavior, your unit economics, and broader market trends. What I’m sure about is this: the days of ultra-cheap subscriptions with unlimited AI usage are numbered. They’re a short-term tactic in a market where no one yet knows the long-term profitability and margins of GenAI products.

For as long as I can remember, subscription pricing has been the dream in SaaS. It’s simple. You know what’s coming in. Keep churn low, and the money just keeps rolling, investors love that kind of clarity.

In fact, in 2023 churn dropped to around 5.4 percent, and subscription businesses grew 4.6 times faster than the S&P 500 over the past decade. Recurring revenue really is powerful. Source: Cashfree

But AI has turned that model upside down. Every AI interaction, every token, every API call, costs real money. It’s not seat-based cost any longer; it’s compute-based cost. You can’t just ignore it. Business leaders are noticing, IDC data shows 59 percent of software companies expect usage-based pricing to make up more of their revenue in 2025, up significantly from earlier years. Source: revenera.com

And we’re seeing companies that blend subscriptions and usage-based pricing—what we call hybrid models—growing around 21 percent faster than pure-subscription businesses. Source: Maxio

For us, the answer was obvious: mix both. At UJJI AI, we offer a subscription that gives customers their own AI agent for L&D, an admin portal with analytics and delivery options, and human support. That’s our stable base. Then, when customers want to build video lessons, quizzes, and real-world tasks—they buy AI-credit packs. It’s pay-as-you-go where the compute cost matters.

Yes, the AI credits side brings some volatility. But we’re still anchored: ARR stays visible, and margins hold because usage charges reflect actual cost. Customers don’t get surprise bills either—they see what they’re paying for, and why. That transparency earns trust.

My Key Tips for SaaS Founders

Anchor your revenue with a subscription

Keep ARR visible. Let it serve as the financial backbone of your business.

Layer on usage-based pricing for AI-heavy features

Only charge usage where compute costs hit you, this keeps your unit economics sound.

Be transparent, show usage clearly

Let customers see their spend, ideally in real-time. Transparency lets them manage spend and trust you.

Forecast usage-based revenue actively

Most SaaS companies doing usage-based are forecasting their variable revenue to manage predictability.

Use hybrid models, they perform

Hybrid is no longer trendy, it’s proving itself. Founders see better net retention, higher growth, and more scalable business outcomes when these models are done right.

In our area of collaboration services, I remain personally unconvinced about the suitability of usage based billing models. My experience has been that many decision makers strongly prefer to have full cost transparency from the very beginning of a contractual relationship.

They want to know exactly what they will be paying over the term of the agreement, without the uncertainty of fluctuating invoices based on actual consumption. This preference is deeply rooted and has been reinforced over many years, making it unlikely to shift quickly.

A change in perspective would require long-term education and repeated exposure to the potential benefits of usage-based approaches.

This tendency is even more pronounced in our work with the public sector. In this environment, annual budgets are typically established well in advance, sometimes covering Multi year periods.

Procurement processes often span several years and are tied to specific budget cycles, which means that financial commitments must be planned with great precision. Consequently, proposals in this sector must always state the total final costs in clear, unambiguous terms.

Under these conditions, introducing a pricing model with variable or unpredictable charges would face significant resistance and, in my view, is unlikely to gain meaningful acceptance at this time.

Seat-based pricing is dead. Over the last decade in software, selling, buying, and especially as a VP of Sales acquiring countless point solutions, I have noticed it is no longer about seats. The problem with seat-based pricing is simple: you never get 100 percent utilization. Employees go on holiday, take sick leave, get fired, or change roles, so there are always unused seats. This means you are overspending due to low utilization.

A much better model, which we have implemented at Salesforce from day one, is usage-based or consumption-based pricing. The idea is to align the value customers get with what they pay. If they use the product more, such as reaching more prospects or using AI credits, they are getting more value, so they pay more. We follow a “land and expand” approach: start with a relatively low fee, then let customers add usage-based features as they see value. The more they scale, the more we charge. This applies across all our products, and we have fully eliminated seats from every subscription.

In sales technology, where results can be achieved with fewer employees, per-seat pricing makes little sense. Our mission is to help companies generate more pipeline with minimal headcount, so charging per seat would contradict our value proposition. Companies today are shifting budget from salaries and commissions toward software, driven by AI and automation. This means consumption is increasing, so charging for usage is the logical choice.

Some companies experiment with outcome-based pricing, charging only when specific results are achieved, but that only works when outcomes are predictable. If not, usage-based pricing is safer and more predictable. We use usage buckets rather than pure consumption to avoid volatility. Customers purchase credit bundles, giving us predictable revenue while still tying pricing to usage.

Our next step is to bundle our subscriptions so customers buy more products together. The incentive will not be discounts. We prefer offering more credits instead. This approach keeps pricing tied to value while encouraging customers to expand their use of our platform.

Pure usage-based pricing sounds logical until you realise you’re nickel-and-diming customers for iterations. If Midjourney charges per image but half of them aren’t quite right, you feel ripped off. Smart pricing gives customers predictable tiers – ‘I’ll probably generate 100-500 images this month, here’s my flat fee’ – rather than watching a meter tick up whilst you iterate towards the right outcome.

The bigger shift everyone’s missing? Freemium has become non-negotiable. Yes, massive free user bases that don’t convert are economically painful, but market expectations have shifted permanently. More importantly, AI discoverability is reshaping everything. If your product sits behind a demo request form, AI agents simply won’t recommend it because they can’t interface with it. You can’t build for an AI-driven world whilst maintaining pre-AI friction.

What makes sense now is hybrid models that reflect how work actually happens. At ScreenCloud, our core monitoring stays subscription-based because we’re watching screens 24/7, but AI content generation could be tiered usage because not everyone needs it. Roles are being fragmented between humans and AI, so pricing should reflect that reality.

Personally, I’d never go pure usage-based – the revenue volatility would keep me up at night and make planning impossible. But kudos to the founders with the guts to embrace that uncertainty because they’re often the ones growing fastest. You have to adapt to where the world is going, not where it’s been.

Usage-based pricing sounds fair in theory, but it can be hard to define and even harder to manage. You need proper backend tracking, Stripe integration, and constant tweaking. Without that, it’s a headache.

We mostly use subscription pricing because our model combines software with a lot of hands-on service. We tested usage-based pricing for a few customers, but they switched to subscription within months. They preferred knowing exactly what they would pay each month.

Usage-based models can surprise customers when usage spikes. Subscriptions keep billing predictable, which helps both sides.

If you are early-stage and still figuring out product-market fit, keep it simple with a subscription. Once you scale, a hybrid model with a base subscription plus usage fees can work well. That is exactly what we are doing now as we roll out our PLG side.