AI Feature Pricing Unit Economics: Fix Margins Before They Spiral

AI features didn’t just change what SaaS products can do — they destroyed the unit economics model that made SaaS profitable in the first place. The shift from near-zero cost-to-serve to variable, consumption-driven token costs means every pricing decision you made before 2022 is now a liability. Roee Hartuv, Senior Consultant at Willingness to Pay and a 15-year software sales veteran, lays out exactly how to rebuild pricing around the new cost structure: cost-correlated tiers, usage guardrails, jobs-to-be-done packaging, and iterative pricing cycles that match the pace of LLM model releases.

Key Takeaways

SaaS companies adding AI features face a structural margin problem that flat-fee or seat-based pricing cannot solve. The fix requires aligning pricing directly to consumption costs, segmenting packages around specific customer jobs, and treating pricing as a continuous iteration process rather than an annual decision. Outcome-based pricing works for vertical AI solutions with measurable outputs, but horizontal platforms need hybrid models. Churn caused by pricing misalignment is recoverable through packaging strategy — not just price cuts.

• Your highest-usage customers are now your highest-cost customers — without usage-correlated pricing, engagement is a margin killer, not a health signal
• Pre-AI SaaS margins of 70–95% are gone unless pricing tracks the variable token costs introduced by AI features
• Outcome-based pricing only works for vertical AI — horizontal platforms must use tiered access plus usage guardrails
• Pricing is now an iterative product sprint, not an annual decision — OpenAI and Anthropic are still testing weekly and haven’t finalized their models
• Churn caused by value-price misalignment is fixable through packaging, including downgrade bundles that retain logos at reduced ARR rather than losing accounts entirely
• Sales teams are the least agile function in most organizations and will resist pricing iteration more than product or marketing will
• Companies with 22% churn driven by competitive bundling can recover revenue by creating defensible feature bundles around capabilities the competitor doesn’t offer

Deep Dive

Why AI Broke SaaS Unit Economics — and What It Costs You

AI fundamentally altered the cost structure of software delivery. For the better part of three decades, SaaS operated on a near-zero cost-to-serve model: once software was built, delivering it to one more customer cost almost nothing. That structural reality allowed gross margins to reach 70–95%, and it made usage a purely positive signal — high-engagement customers were your best customers.

Token-based AI changed every one of those assumptions simultaneously.

“Classical SaaS before we had an AI the cost to serve a customer was practically zero. And now we’re in the age of AI and we all know that the age of AI AI doesn’t like it costs real money. So all of a sudden the entire unit economics and our cost structure completely change cuz it’s no longer zero to serve the next customer. It comes with a cost. If we don’t price our products, our products correctly, then we’re just going into that marginal spiral and it can get spiral out of control.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

The marginal spiral Hartuv describes isn’t theoretical. When a customer who generates 10x average token consumption pays the same flat monthly fee as a low-usage customer, you are effectively subsidizing their margin destruction with revenue from lighter users. At scale, the most engaged segment of your user base becomes the most economically dangerous cohort in your portfolio.

The business risk compounds because the old instincts are exactly wrong. Product and customer success teams are trained to celebrate high engagement. Power users get case studies. Heavy usage signals product-market fit. None of that logic applies when every interaction carries a real cost you haven’t priced for.

How to Build a Cost-Correlated Pricing Model for AI Features

Cost-correlated pricing is the foundational fix for AI feature pricing unit economics. Instead of charging flat fees or per-seat rates disconnected from consumption, you build a pricing structure where the fee a customer pays tracks directly with the token usage or interactions they trigger. This protects gross margin by ensuring that as your cost-to-serve increases, revenue scales proportionally.

The model has five operational components, as Hartuv describes it:

1. Track real-time cost per customer interaction or token consumption — you cannot price what you cannot measure; instrument usage at the customer level before setting tiers
2. Set usage guardrails to cap high-consumption outliers before they trigger margin collapse
3. Build pricing tiers that correlate directly to your cost structure, not just to perceived value or competitive benchmarks
4. Communicate cost correlation transparently to customers — position it as fairness (you pay for what you use) rather than punishment for engagement
5. Iterate monthly as underlying LLM costs shift — token costs are themselves volatile, and a pricing model calibrated in Q1 may be mispriced by Q3

“So there’s always like that that hockey stick effect. We want to cap the the very strong users or the the users that use this the most. We need to make sure that our pricing correlates to the cost, right? It’s no longer about classical SAS. Doesn’t matter how much you charge, cost is zero. But you need to make sure that the more usage, the more cost, the more the price increases.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

Usage guardrails and fair-use policies are not optional. They are non-negotiable mechanisms for protecting margin in any AI-heavy product. Without them, you’ve essentially written a blank check to your most engaged customers.

Outcome-Based vs. Usage-Based Pricing: Which Model Fits Your AI Product

The right pricing model depends entirely on whether your AI solves a specific, measurable outcome or enables a broad, diverse range of outcomes across different customer types.

The distinction matters because outcome-based pricing requires you to define and measure the outcome reliably. For a company like Fin AI, where the product closes support tickets, the unit of value is unambiguous: a closed ticket. You can charge per ticket, and customers understand exactly what they’re paying for.

“If you’re very vertical niche focused for example Fin AI they do ticket sales support for them it’s clear like you close a ticket the outcome back to outcome based pricing the outcome is clear you can charge on that. OpenAI ChatGPT they’re a horizontal solution. The value is completely different.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

For horizontal platforms, the customer using your AI to write marketing copy and the customer using it to analyze financial data are extracting fundamentally different value from the same product. Charging both identically per “outcome” is impossible because the outcomes aren’t comparable. Tiered feature access combined with usage guardrails is the practical path for horizontal AI — accept that you won’t perfectly capture value, and optimize instead for retention and predictable margin.

Why Churn Is a Pricing Problem, Not Just a Product Problem

Retention failures are frequently misdiagnosed as product problems. The real driver, in a majority of cases, is a mismatch between what customers perceive they’re getting and what they’re paying for it. Pricing and packaging strategy can fix this directly — without requiring a single line of new product code.

Hartuv draws this distinction sharply based on his client work: “When companies churn, it’s because the value that they’re getting does not match the price that they’re paying for it. With the right pricing and packaging, you can fix retention as well.”

The tactical application is the Churn Recovery Through Packaging Downgrade framework. When a customer is at risk of churning — whether due to price sensitivity or a competitor offering a bundled alternative — the reflex move is to discount or match the competitor’s price. Hartuv’s approach is different: identify which features in your product the competitor doesn’t offer, create a lighter package bundling only those defensible capabilities, and offer that as a downgrade option rather than a cancellation.

One of Hartuv’s current clients is experiencing a 22% churn rate driven entirely by a competitor gaining market share through aggressive bundling. The recovery playbook isn’t to out-bundle the competitor — it’s to find the white space:

“There are components that the big competitor does not deliver. So maybe we can retain the logo, create a new bundle, a new package that serves only those things that they do not offer and it’s not a logo journey. It might be a downgrade, but we’re still keeping that’s better than losing them.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

A downgrade from $X ARR to $0.6X ARR is a better outcome than a cancellation at $0 ARR. Logo retention matters beyond the immediate revenue — it preserves the expansion path and protects reference-ability.

Treating Pricing as an Iterative Product Sprint

The single biggest structural shift Hartuv advocates is organizational: stop treating pricing as a periodic decision and start running it like an agile product function.

In most B2B SaaS companies today, pricing is reviewed annually — if it’s reviewed at all. Packaging decisions made at Series A are still in market at Series C. Meanwhile, the LLM ecosystem producing cost changes is evolving weekly. A pricing model calibrated against GPT-4 token costs may be significantly mispriced against whatever replaces it six months later.

“Pricing and packaging is becoming like like products. It is never static. Especially if you have AI and you know with all the rapid changes every week a new LLM comes out. You need to be prepared to iterate test improve like test iterate improve learn and vice versa.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

The Iterative Pricing as Product Development framework operationalizes this with five steps: establish a cross-functional pricing steering committee (product, ops, finance, and sales leadership); run weekly experiments testing messaging, tier positioning, and add-ons with customer cohorts; measure impact on conversion rate, ACV, churn rate, and margin per customer; maintain a shared pricing playbook documenting what works; and push changes incrementally rather than waiting for a “complete” solution.

The resistance point in most organizations is sales. Product teams run sprints. Marketing teams A/B test. Sales teams do neither — they operate on quarterly playbooks and resist pricing iteration because it disrupts their established pitches and objection-handling scripts.

“I don’t feel like sales knows this yet. But when you think of it, pricing and packaging is not only a sales problem. It touches all the disciplines, right? But you’re right. I think the the sales team is the least agile team that we have in our organization right now.”

— Roee Hartuv, Senior Consultant at Willingness to Pay

The implication for GTM leaders is structural: pricing iteration cannot be run through the sales team alone. It requires a cross-functional steering committee with executive authority to push changes without waiting for sales team consensus.

The reassurance for companies that feel behind: even OpenAI and Anthropic haven’t solved this. “If the best companies in the world are testing iterating etc. They haven’t figured it out and we’re also trying to figure out and learn what this new technology and capabilities can give.” The absence of a proven industry template is not a failure — it’s the current reality. The companies that build pricing iteration into their operating rhythm now will compound advantage over those waiting for the industry to produce a definitive playbook.

When to Standardize Pricing — and When to Stay Loose

One of the most counterproductive moves early-stage companies make is trying to optimize pricing before they have product-market fit. Hartuv is explicit: if your product is still pivoting, your customer segments are still shifting, and your go-to-market motion isn’t repeatable yet, there is nothing to standardize.

The right trigger for pricing standardization is repeatable GTM fit — the point where you have consistent signals about who buys, why they buy, and what value they’re extracting. Before that trigger, test freely. After it, treat pricing as the highest-ROI lever in your entire commercial system.

“The biggest growth lever we have as commercial organization is getting the pricing and packaging right. I’ve seen companies that no matter how much we build a fantastic sales process, if the packaging doesn’t work, it adds friction and yeah, you can’t close deal or you’re leaving money on the table. But if you do it correctly, then yeah, renewals get easier, sales process gets faster, ACV goes up.”

That sequence — faster sales cycles, higher ACV, easier renewals — represents compounding commercial leverage that no other single GTM investment delivers at equivalent ROI.

Who This Is NOT For

This approach requires operational prerequisites that not every company has in place. Be honest about where Hartuv’s frameworks do and don’t apply before investing in a pricing overhaul.

• Pre-product-market-fit companies should not prioritize pricing standardization. If your product is still pivoting and your customer segments aren’t stable, standardizing pricing now adds process overhead without commercial return. Hartuv is explicit: test freely until you have repeatable GTM signals, then standardize.

• Companies without usage instrumentation cannot implement cost-correlated pricing. The entire Cost-Correlated Pricing Model depends on tracking real-time cost per customer interaction. If your infrastructure doesn’t measure token consumption or usage at the customer level, you cannot build the pricing tiers required. The instrumentation work comes first.

• Outcome-based pricing fails for horizontal AI platforms with diverse customer segments. If your AI enables fundamentally different jobs across different buyer types, you cannot define a universal outcome unit to charge against. Forcing outcome-based pricing onto a horizontal product will produce revenue unpredictability and customer confusion.

• Sales-led organizations with long deal cycles face significant internal resistance. Hartuv identifies sales teams as the least agile function in most organizations. If your revenue motion is primarily enterprise sales with 6–12 month cycles, iterating pricing weekly will create internal friction with sales reps whose pitches depend on pricing stability. Executive sponsorship of the pricing committee is non-negotiable in this context.

• This is not a discounting playbook. The packaging downgrade approach for churn recovery is distinct from offering price reductions. It requires identifying genuine white-space capabilities that competitors don’t offer. If your product has no defensible differentiation from the bundled competitor, a downgrade package won’t retain logos — it will just delay cancellation.

About Roee Hartuv

Roee Hartuv is a Senior Consultant at Willingness to Pay, a firm specializing in pricing and packaging strategy for B2B software companies. His perspective on AI feature pricing unit economics is grounded in 15 years of direct software sales experience — not theoretical frameworks — followed by four years running a general go-to-market strategy consulting practice covering sales process design, outbound motions, and marketing integration. After four years of broad GTM work, Hartuv concluded that pricing and packaging was consistently the highest-ROI growth lever available to commercial organizations, and narrowed his practice to focus exclusively on it. He currently works with clients navigating competitive bundling threats, AI-driven cost structure shifts, and churn driven by value-price misalignment — including a client managing a 22% churn rate caused by a competitor’s aggressive market consolidation.

Ready to Rebuild Your Pricing Before the Margin Spiral Hits?

The window between “AI features are live” and “AI features are destroying our margins” is shorter than most founders expect. Roee Hartuv’s core argument — that pricing must now be treated as an iterative, cross-functional product sprint rather than a periodic commercial decision — is a structural shift that touches sales velocity, ACV, renewal rates, and gross margin simultaneously. If you’re a founder or GTM leader at a B2B SaaS company with AI features in market and pricing models that haven’t been updated to account for variable token costs, the cost of inaction compounds every week.

Talk to a Growth Strategist →

Frequently Asked Questions

How do you implement usage guardrails without angering customers or leaving money on the table?

Position guardrails as fairness mechanisms, not punishment. Communicate transparently that your pricing is consumption-based — customers who use less pay less, and heavy users pay proportionally more. Build tiers with clear usage thresholds communicated at purchase, not disclosed only when customers hit limits. Hartuv’s principle is that guardrails must cap the highest-consumption outliers who trigger margin collapse, while giving average users headroom they never hit. Surprises cause anger; transparent caps cause negotiation, which is recoverable.

How often should SaaS companies update their pricing model in the AI era?

Monthly at minimum for the cost-correlation calibration (as LLM token costs shift), and weekly for packaging experiments and messaging tests. Hartuv uses the product development analogy deliberately: pricing iteration should follow the same agile cadence as sprint cycles. OpenAI and Anthropic — the best-resourced AI companies in the world — are still testing and iterating their pricing weekly. Any B2B SaaS company treating pricing as an annual decision is operating on a cadence that is structurally incompatible with the pace of AI cost changes.

Can you compete with larger bundled competitors by changing your pricing and packaging strategy?

Yes — but not by matching the competitor’s bundle. The strategy is to identify capabilities your product offers that the bundled competitor does not deliver, then create a defensible lite package around only those features. Hartuv is working with a client at 22% churn from exactly this competitive pressure. The goal is logo retention at a lower ARR rather than full account loss: “It might be a downgrade, but we’re still keeping that — that’s better than losing them.” The retained logo preserves the expansion path once the competitor’s bundled offer loses momentum.

What’s the difference between pricing for horizontal AI tools vs. vertical AI solutions?

Vertical AI solves a specific, measurable outcome for a defined customer type — a support ticket AI that closes tickets can charge per closed ticket because the unit of value is unambiguous and consistent across all buyers. Horizontal AI enables different outcomes for different users, making per-outcome pricing impossible. A horizontal platform serving marketing, finance, and operations buyers simultaneously cannot define one outcome unit that works for all three. Horizontal products require hybrid models: tiered feature access combined with usage guardrails, accepting that pricing will never perfectly capture every customer’s value.

How do you get your sales team to accept and iterate on new pricing models?

You don’t route pricing iteration through the sales team — you run it above it. Hartuv identifies sales as the least agile function in most organizations, more resistant to change than product or marketing teams who already operate in sprint cycles. The solution is a cross-functional pricing steering committee with authority from product, operations, finance, and sales leadership — not sales reps. Changes are pushed incrementally with clear measurement frameworks (conversion rate, ACV, churn rate, margin per customer) so data, not internal preference, drives decisions. Sales buy-in follows results, not persuasion.