The board meeting always follows the same script.

"Where's our AI strategy?" someone asks. "Competitor X just launched AI features. We need something by Q2."

And just like that, teams scramble. Engineers explore LLMs, Product managers draft roadmaps and Marketing prepares launches.

But there's one conversation nobody's having, “How do we price this thing?”

Marcos Rivera has seen this movie play out hundreds of times. As the former head of pricing at Vista Equity Partners and founder of Pricing I/O, he's priced over 500 B2B SaaS products. And in 2025, he's watching a painful pattern repeat.

"Nobody agrees on what they need to do, where it needs to go in the packaging, what the goal is," Marcos says. "They just got to do AI something. The board wants to see AI."

That vague urgency is creating a pricing nightmare. Companies are copying competitor models that don't fit their use case. They're confusing customers with meaningless metrics like "actions." They're bleeding margins because one AI agent might operate at 60% margins while another runs at 10%.

TL;DR

• Most B2B SaaS companies are botching AI pricing because they're copying competitors without understanding their own strategy. After pricing 500+ products, Marcos Rivera identifies three frameworks that actually work:

• Framework #1: Underscore the Core - If AI improves your existing features. Fold it into your product and raise prices 10%+ because you're delivering faster time-to-value.

• Framework #2: Upgrade the More - AI extends your product into premium territory. Launch as an add-on first, collect 2-4 quarters of data, then decide whether to fold in or keep separate.

• Framework #3: Unlock the New - AI creates entirely new capabilities. Price it standalone with fixed + variable structures tied to outcomes (like transactions processed or claims approved). Define the "anti-outcome" to avoid billing disputes.

• Key mistakes to avoid: Vague metrics like "credits" or "actions" that customers can't understand. Jumping from pure subscription to pure usage-based pricing without hybrid transitions. Adding pricing complexity on top of product complexity when users are already climbing the AI learning curve.

• The safe default when uncertain: Launch as an add-on, watch who buys it and how they use it, then decide based on real data rather than assumptions.

• AI is moving too fast for two-year pricing cycles. Review quarterly or bi-quarterly. Strategy clarity comes first you can't price what you can't define.

The old SaaS pricing playbook doesn't work anymore.

But after analyzing hundreds of companies, Marcos has identified three frameworks that actually work for AI pricing. Not theory, what actually drives revenue while protecting margins.

Let's break down why most companies are getting this wrong, and how to fix it.

1. The Current State of AI Pricing Chaos

Walk into any B2B SaaS company right now and ask the product team what their AI strategy is. Then ask sales. Then ask the pricing lead.

You'll get three completely different answers.

This isn't just organizational misalignment, it's creating real financial damage. And Marcos has a front-row seat to the carnage.

2. The "Something with AI" Problem

The pressure is coming from everywhere. Investors want to see AI in the roadmap. Competitors are announcing AI features. Customers are asking about it. So companies rush to ship something, just anything with artificial intelligence baked in.

But what is this AI actually supposed to do?

Marcos is blunt about what he's seeing: Companies can't answer basic questions. 

• Is this AI enhancing your core product? 

• Is it extending into new use cases? 

• Is it targeting a completely different customer?

Without clarity on strategy, pricing becomes impossible. You can't price what you can't define.

3. Three Expensive Mistakes

The chaos is manifesting in predictable patterns:

a. Mistake #1: The Copycat Trap

Someone sees a competitor launch "credits" or charge "per action," and suddenly that's the plan. 

• Never mind that credits mean nothing to customers. 

• Never mind that "action" could mean literally anything. 

It looks like modern pricing, so it gets copied wholesale.

Marcos recently saw a company introduce "per action" pricing. When he asked what an action was, the answer was: "Well, it could mean this and it could mean that."

That's not a pricing model. That's just confusion with a price tag.

b. Mistake #2: Complexity Stacked on Complexity

Your customers are already trying to figure out how to use AI. They're learning new workflows, new interfaces, new ways of thinking about their work. 

And now you're adding pricing models they need a decoder ring to understand?

c. Mistake #3: The Margin Blindspot

Not all AI features are created equal from a cost perspective.

Traditional SaaS could get away with stable, high margins. Build the feature once, serve it to thousands of customers, margins stay intact. 

But with AI agents, one might cost you 10% of revenue to operate. Another might cost 40%. And if you're pricing them the same way, you're bleeding money on every transaction.

Why this matters more than ever

Sam Altman and other LLM CEOs are predicting 2025-26 as the "year of AI agents." Marcos thinks they're being optimistic, 2027-28 is more realistic. Humans need time to change processes, rethink workflows, actually adopt these tools.

And that timeline gives you breathing room to get pricing right before the agent revolution fully hits. Companies that figure this out now will dominate when agents become table stakes.

The ones still fumbling with vague "credit" systems and copied pricing models will be stuck explaining to their board why margins collapsed and customers churned.

And there's a better way. And it starts with understanding which of three frameworks actually fits what you're building.

Framework #1: Underscore the Core

Let's start with the simplest and often most overlooked approach to AI pricing.

Underscore the Core

Marcos breaks it down simply: "Underscore the core’ means that your strategy is to enhance what you already do, your search, workflows, decision-making, and core functionality.

If AI makes those things better, faster, or smarter, then you should fold it into the product itself, and reflect that improvement in your pricing.”

This is the unglamorous truth about most AI implementations. You're not building Jarvis from Iron Man. You're making your existing search faster. Your workflows are smoother. Your analytics sharper. 

The fundamental job-to-be-done hasn't changed, you're just helping customers complete it with less friction and more speed.

And here's what makes this powerful: some of this AI is visible to users, some runs invisibly in the background. 

A customer might see AI-suggested actions in their dashboard, but not see the AI pre-processing data, optimizing queries, or predicting their next need. Both types deliver value. Both justify higher pricing.

When this framework actually fits

The mistake companies make is thinking every AI feature needs its own SKU, its own add-on, its own pricing tier. Sometimes the smartest move is integration, not separation.

You're in Underscore territory when:

• Your AI gets users to value faster without changing what that value is

• The enhancement feels inevitable, like "of course this should work this way"

• Users would struggle to separate "the AI part" from "the product part"

• The cost structure remains relatively stable and predictable

Think about spell-check in Microsoft Word. Nobody pays extra for it. Nobody thinks of it as a separate feature. It's just... how Word works now. That's what Underscore the Core AI should feel like.

Fold it In, then raise 10%+

Here's where Marcos pushes back against the 3-5% increases most companies consider.

"I’m not talking about small increases like 5%. I mean taking prices up by 10% or more especially when users can clearly see the impact of the AI, and even when some of that impact is invisible.

Either way, the AI is getting users to value faster. It’s helping them accomplish their jobs and use cases more quickly, more effectively, or more deeply."

Why does 10%+ work when 5% feels aggressive?

Because you're not just adding a feature. You're compressing time. You're eliminating frustration. You're making your product do what customers always wished it would do.

Consider the math from a customer's perspective: If your AI cuts the time to complete a workflow from 30 minutes to 10 minutes, you've just given them back 20 minutes per workflow. 

If they run that workflow 50 times a month, that's 16+ hours saved. What's that worth? Way more than 10%.

The trick is making sure customers feel that time compression. If the improvement is invisible or incremental, you'll face resistance. If it's obvious and immediate, the price increase becomes a no-brainer.

Getting this right

Don't lead with "We added AI." Nobody wakes up wanting AI. They wake up wanting their work to be easier.

Your announcement isn't "Introducing AI-powered search!" It's "Find what you need 70% faster with intelligent search that understands context."

The AI is the how. The time saved is the what. The outcome is the why.

Timing the increase: Marcos is clear on this: "Fold it in and raise your prices right? I'm not talking 5% raises."

But when matters as much as how much. Roll the price increase when you launch the AI features, not after. The value jump and the price jump need to happen simultaneously.

Because customers understand paying more for getting more when it's obvious they're getting more. Launch AI in January, raise prices in June, and you're asking them to remember what changed six months ago. Launch both together, and the connection is immediate.

The grandfathering decision: Every company struggles with this. Do existing customers get the new AI at their old price? Or does everyone move to new pricing?

There's no universal answer, but here's Marcos's framework: if the AI delivers meaningfully more value and we're only talking about Underscore the Core, if it does then existing customers are getting that value too.

You can soften the blow with timing (6-month notice, grandfather until next renewal) or with transitional pricing (halfway between old and new for the first year). 

But ultimately, if you've truly enhanced your core product, your existing customers are benefiting. They should expect to pay for that benefit.

The companies that get in trouble are the ones who under-communicate. They assume customers will notice the improvements and understand the value. Wrong. 

You need to actively message what changed, why it matters, and how it makes their work better.

The margin math that makes or break this

Here's where the rubber meets the road. Underscore the Core only works if your AI costs are relatively stable and predictable.

Marcos is watching companies miss this crucial detail. They fold AI into base pricing, then realize customer usage varies wildly. 

One customer's AI enhancement costs them $2/month. Another costs $47/month. Same price point, completely different margins.

"The margins are high and stable, and as you keep building and extending the platform, bolting on modules, adding integrations, the margins kind of stay intact. But not in AI, not with agents. 

You can have one agent with a 60% margin and another one with a 10% margin. It just depends on the cost, how much tokens it's consuming and how much users are using it."

This is why Underscore the Core works best for AI that enhances existing features in relatively uniform ways. 

If your AI is optimizing database queries in the background, or improving search relevance, or suggesting next actions and these costs don't spike dramatically based on user behavior, flat pricing makes sense.

But if your AI costs are driven by consumption, if heavy users genuinely cost you 10x what light users cost you need to look at Framework #2 or #3. 

Trying to force Underscore the Core pricing onto variable-cost AI is how you accidentally build a company that loses money as customers adopt your product.

Which is, you know, the opposite of the goal.

Framework #2: Upgrade the More

Marcos describes this as taking your product "a few steps further" with AI. Your product used to end here. But now, with AI, you can go several steps beyond that endpoint.

"Some companies where it's 'I think we're just taking the things a bit further. We could—our products ended here, but now we can go a few steps further with AI.' And this gets into more complex territory. You want to upgrade the more. This is more jobs, more use cases or more extension of the product."

This isn't about making existing features better. It's about unlocking capabilities that were previously impossible or impractical. You're expanding the scope of what your product can do.

The key distinction from Underscore the Core: not every customer needs this. In fact, many won't. And that's perfectly fine, it's what makes this a premium or add-on play.

When does this framework make sense

You're in Upgrade the More territory when:

• Your AI enables new use cases that extend beyond your original value proposition

• The capabilities feel "advanced" or "power user" rather than fundamental

• Different customer segments have dramatically different appetite for these features

• The enhancement is substantial enough that users can clearly distinguish it from the base product

Think of it this way: if someone could use your product effectively without this AI feature, but power users or larger customers would jump at the chance to have it, you're looking at Upgrade the More.

The pricing model: Premium tiers or add-ons

Marcos offers two paths here, and they're not mutually exclusive.

Path 1: Premium Plan Placement Put these AI capabilities in your higher-tier plans. Your Enterprise plan gets AI-powered forecasting. Your Professional plan gets advanced automation. Your Basic plan works just fine without them.

This creates natural segmentation and gives customers a reason to upgrade beyond just "more seats" or "higher usage limits."

Path 2: The Add-On Strategy Offer the AI enhancement as a standalone add-on that any tier can purchase. This works particularly well when you're unsure about demand or still figuring out who values this most.

And here's the safety net Marcos recommends: "If you're not sure, if you're like, 'Well, Marcos, it kind of is all over,' like, that probably means you're not clear with what you're trying to do, but the safest path would be add-on first and then decide if you want to fold it in or keep it separate based on the usage patterns, who's buying it, and what they're doing."

Marcos points to a clear pattern among major players: "Gemini did it, Slack did it, Notion did it, they had the add-on for a while. I think it was like 10 bucks a user, and then they ended up folding it in after that."

The strategy here is straightforward: launch as an add-on, watch what happens, then decide whether to integrate it more deeply into your pricing structure.

These companies didn't immediately try to build the perfect tiered structure. They didn't agonize over whether AI belonged in Professional vs. Enterprise. They launched it as an add-on at around $10 per user, collected data on adoption and usage, and made informed decisions later about integration.

That's the pattern worth following.

The add-on testing strategy: Collecting the right data

If you go the add-on route and Marcos strongly suggests you should when uncertain—you need to know what you're measuring.

Who's buying? Is it your largest customers? Your most sophisticated users? Specific industries or use cases? If only Enterprise customers are buying, that's a signal it belongs in your top tier. If it's spread across all segments, maybe it should be in the core product with

Framework #1.

Usage patterns matter Are customers who buy the add-on using it daily? Weekly? Or did they buy it and barely touch it? High engagement says keep it separate and premium. Low engagement says you might have a product-market fit problem, not a pricing problem.

Willingness to pay indicators How many customers ask about it during sales calls? How often do existing customers inquire about adding it? Are they price-sensitive, or do they buy without hesitation? These signals tell you how much value the market perceives.

The timeline question Marcos suggests giving it 2-4 quarters to see real patterns. One quarter isn't enough you'll catch early adopters but miss the mainstream. A full year might be too long if you're in a fast-moving market.

The sweet spot? Two quarters minimum. Four quarters if you want high confidence in your data.

Sometimes the data tells you something unexpected: different customer segments value the same AI feature for completely different reasons.

A project management tool might find that small teams use AI for automated task assignment (they lack dedicated project managers), while large enterprises use it for cross-project resource optimization (they have too many projects to manually coordinate).

Same AI. Different jobs-to-be-done. Different willingness to pay.

When you see this, you might need hybrid approaches: include basic AI functionality in the core, but gate advanced capabilities as premium features or add-ons. Framework #1 plus

Framework #2.

The companies that struggle are the ones that force a single strategy across diverse customer needs. The ones that win are comfortable with nuance.

The Margin Consideration Redux

Remember Marcos's warning about margin variability? It applies here too, but differently.

With Upgrade the More, you're often dealing with more compute-intensive AI. Advanced forecasting models. Complex automation chains. Deep analytical processing. These features genuinely cost more to deliver.

If your margin on these features is tight, that's actually an argument for keeping them separate. Charge appropriately for the value and the cost. Don't subsidize power users by spreading costs across your entire customer base.

But if margins are healthy and stable even on advanced features, and adoption is broad, that's a signal to fold them into tiered pricing rather than maintaining them as perpetual add-ons.

The key question: Does separating this feature give you pricing flexibility you need, or is it just creating complexity that frustrates customers?

Only your data can answer that.

Framework #3: Unlock the New

Sometimes AI doesn't just enhance or extend your product. Sometimes it breaks open entirely new possibilities—new users, new use cases, new markets you couldn't touch before.

That's when you need the third framework.

Unlock the New: When AI Changes the Game

Marcos describes this as the most transformative approach: "Unlock the new—whole new use case, maybe even a different user, a different TAM, something different. You're able to isolate that, charge it as a standalone, and then you can even bundle it with your core and your platform as well."

This isn't about making your product better or taking it further. This is about doing something fundamentally different. Different enough that it could stand alone. Different enough that it might attract customers who would never buy your core product.

The clearest signal you're in Unlock the New territory: you could spin this into a separate company, and it would make sense.

When This Framework Applies

You're looking at Framework #3 when:

• Your AI enables completely new use cases that weren't possible before

• You're targeting a different user persona or buying center

• The capability could attract a different TAM or market segment entirely

• This AI could displace another product category

• The technology creates genuine competitive differentiation, not just incremental improvement

This is the realm of agents doing work autonomously. Of AI that doesn't assist humans but handles entire workflows independently. Of capabilities that make customers rethink their entire process, not just optimize it.

The Pricing Model: Standalone with Bundle Options

Marcos is clear on the approach here: price it as a separate product, but create strategic bundle opportunities with your core platform.

Why standalone? Because the value proposition is distinct enough that it deserves its own positioning, its own pricing logic, its own go-to-market motion. Forcing it into your existing pricing structure would confuse customers and likely leave money on the table.

Why bundle options? Because if you already have customer relationships and a platform they trust, offering this new capability as part of a broader solution creates competitive moats. It's harder for a startup to displace you when customers are getting multiple products that work together seamlessly.

"It also gives you some competitive advantages to be able to use it to penetrate new markets," Marcos notes. This is your wedge into territories where your core product never gained traction.

Real-World Examples: The Agent Revolution

This is where the agent companies are playing, and they're writing the playbook in real-time.

Marcos calls out several: "Zendesk and Intercom do it, ChargeFlo does it, DecaGon, all these different companies that have some agentic solution."

What unites them? They're not positioning AI as a feature enhancement. They're positioning autonomous systems that handle outcomes.

The pattern emerging: These companies are using a fixed + variable pricing structure. The fixed component covers configuration, training, onboarding—the setup work. The variable component is tied to a unit of work the agent actually completes.

Marcos breaks down the specifics: "Maybe that unit of work is transactions like authorizations approved like I've seen in health tech or legal discovery packages summarized like in legal tech."

Notice what's happening here: the pricing isn't about usage of the AI. It's about outcomes the AI delivers.