Overview

You've built AI features. They work. Customers like them. Now the question: should you charge for them?

This is where product strategy meets business model. Get this right and AI features fund your continued investment in AI. Get it wrong and you're subsidizing features customers would happily pay for.

The decision isn't "charge or don't charge." It's "what's the right pricing model for this particular capability, our market position, and our business goals?"

The Core Pricing Models

Model 1: Free (Table Stakes)

The AI capability is included in every plan. It's expected but not differentiating.

When to use: Everyone's competitor has this feature. Customers expect it included. It's hygiene, not differentiation. Or it's a hook to get users into your ecosystem; the ROI comes from upsells and expansion, not from the free feature itself.

Example: Gmail's smart compose. It's included because everyone expects email to be "smart" now. It's not a premium feature. But it keeps you in Gmail instead of switching to Outlook. The value is in lock-in, not in direct revenue.

Risk: You invest in building and maintaining a capability that customers don't pay for. Makes sense only if it reduces churn (they'd leave if you didn't have it) or drives expansion (they upgrade to premium for other reasons). If neither is true, you're burning resources. Measure: "If we removed this AI feature, would 10%+ of customers churn?" If not, charge for it.

Financial model: Free AI feature costs you $5/user/year to run. If it prevents churn worth $200/customer, or drives $300 expansion revenue per customer, then it pays for itself 40-60x over. But if it drives $0 expansion revenue and prevents 0 churn, it's a pure cost center. Be honest about which situation you're in.

Model 2: Freemium (Free for Basic, Premium for Advanced)

Basic AI capability is free. Advanced version costs money.

Example: Free users get AI writing suggestions with 100 uses/month. Premium users ($15/month) get unlimited suggestions plus AI-powered analytics and custom model fine-tuning.

When to use: There's a natural gradient from basic to advanced. You want to acquire users on free tier. Some will pay for premium. Freemium works best when: 1) free tier is genuinely useful (people find value immediately), 2) premium tier has clear upgrade reasons (users hit limits or want advanced features), 3) free tier cost is low enough that many free users can sustain you profitably.

Financial model: If you have 10K free users at $2 cost/year and 500 paid users at $15/month ($180/year), revenue is $90K/year. Free tier costs $20K/year. Net: $70K. Profitability depends on whether other costs (support, infrastructure) allow profitable operation at that scale. Most freemium products need 3-5% conversion rate to the paid tier to work financially. Below that, you're burning cash on free users.

Risk: Freemium is complex. Most users stay on free tier (typical conversion: 2-3%). You need a compelling reason for them to upgrade. And free tier costs you money (hosting, compute, support). Make sure payoff is worth it. Track: "Cost per free user vs. LTV of paid customers." If cost is negative LTV, you're losing money on the model.

Model 3: Premium Add-On**

Base product costs X. AI features cost X + Y.

When to use: Your base product is already paid. AI is a clear upgrade path. Customers already understand paying you.

Example: Slack. Base Slack is $10/user/month. AI features (transcription, summaries) might be $5-10 more per user per month.

Risk: If base product isn't defensible, add-on pricing is hard to justify.

Model 4: Usage-Based (Pay Per Use)**

Customers pay for how much they use the feature. 1,000 API calls? $10. 10,000 calls? $100.

When to use: Usage is variable. Some customers use lightly, others heavily. Pay-per-use aligns incentives.

Example: Claude's pricing is pay-per-token. You only pay for what you use.

Risk: Customers dislike unpredictable bills. Usage-based works if: a) customers understand the model, b) costs scale reasonably, c) there's some cap or predictability.

Model 5: Seat-Based (Per User Per Month)**

Each user who uses the feature is $X/month.

When to use: Usage is tied to individual users. Each user gets value independently.

Example: Copilot for Office is $20/month per user for Microsoft 365 subscribers.

Risk: Can be prohibitively expensive if there are many users.

The Pricing Principle: Charge based on value delivered, not on cost. If an AI feature saves a customer $10,000/year, charge $3,000/year. Don't charge $500 because your cost is $400. That leaves money on the table.

Deciding on Pricing: The Framework

Question 1: Is This Feature Table Stakes or Differentiation?**

Table stakes = everyone has it or expects it. Price as free (or included).

Differentiation = competitors don't have it or you do it much better. Price as premium or add-on.

Question 2: What's the Customer Willingness to Pay?**

Do customers see this as valuable? Would they pay for it standalone? Or is it just "nice to have"?

Easiest way to find out: ask them or test. "If we charged $50/month for this feature, would you pay?" If 30%+ of customers say yes, it's viable.

Question 3: What Does Your Competition Charge?**

If your competitor charges $50/month for similar AI, you probably shouldn't charge $500/month unless you're dramatically better.

Use competitive pricing as a reference point, not a rule. You might charge more (better feature), less (penetration pricing), or differently (different model).

Question 4: What's Your Business Goal?**

Maximize revenue? Maximize adoption? Build moat? These lead to different pricing.

Max adoption: free or freemium. Max revenue: premium add-on. Build moat: premium or usage-based with volume discounts (lock in big customers).

Question 5: What's Your Market Position?**

Are you a premium product or cost leader?

Premium (Figma, Notion): price AI features as premium add-ons. Customers expect to pay for quality.

Cost leader (Canva): price lower or include free. Customers expect good features at low cost.

Packaging: How to Structure AI Features

Don't Charge For Each Feature**

Avoid: "Writing assistance costs $5/month. Code generation costs $10/month. Summarization costs $3/month." This is confusing and creates decision paralysis.

Instead: Package related capabilities together. "AI assistant pack: all writing and analysis capabilities for $15/month."

Create Clear Tiers**

If you're doing freemium:

Free: Basic AI features, limited usage (100 operations/month)

Pro: Advanced AI features, unlimited usage, priority support ($30/month)

Enterprise: Custom models, dedicated support, highest priority (custom pricing)

Clear tiers make it obvious what you get at each level.

Use Limits as Levers**

Free tier: 100 AI operations/month, 5-minute response time

Pro: 10,000 operations/month, real-time responses

Limits drive conversion without creating resentment. People understand why free has limits.

Combine AI With Other Value**

Don't sell AI alone. Sell "AI + features around it." "Advanced customer support" (includes AI, knowledge base, routing). This is easier to price and justify.

When This Goes Wrong: Pricing Failure Modes

The Pricing Inversion**

You launch with usage-based pricing at $0.10 per transaction. Early customers love it and scale aggressively. One customer does 100K transactions/month and gets a $10K bill they didn't expect. They churn, leave a 1-star review, and tell their friends. Now people are scared of your product. Solution: cap bills. "Maximum $500/month charge even if you exceed limits." Or switch to tiered pricing above a threshold. Make surprises impossible.

The Freemium Trap**

You launch freemium with the goal of 5% conversion rate. You get 100K free users but only 0.5% convert. You're spending $100K/year on free users and making $5K/month from paid. Unsustainable. Solution: before launching freemium, validate that paid conversion will work. Talk to 20 people on free tier. "Would you pay $19/month if you hit the usage limit?" If less than 5 say yes, freemium won't work. Do something else.

The Enterprise Price Ceiling**

You price your AI add-on at $50/month. You sell 500 customers at that price, maxing out to $300K/year ARR. Growth stalls because you've hit the price ceiling. Enterprises would pay $200/month but startups can't. Customers in middle are fine at $50 but you're leaving revenue on the table. Solution: segment pricing. Startup: $30/month. Mid-market: $75/month. Enterprise: $300+/month. This increases your market and revenue significantly.

The Competitor Price War**

You price AI features at $50/month. A competitor launches at $20/month. You panic and drop to $15/month. Now you're losing money. Solution: understand their business model. Are they selling cheaper because they built cheaper infrastructure? Or are they burning cash to gain market share? If they're burning cash, wait them out. If they're actually cheaper, you might need to reduce your costs, not just your prices. Competing on price alone is a race to zero.

Common Monetization Mistakes

Mistake 1: Charging Too High

You build a feature. You think it's worth $50/month. You price it there. Adoption is 5%. Revenue: $2,500/month. You'd have made more money at $20/month with 30% adoption ($18,000/month).

Start low. Increase prices over time as value becomes obvious. Price elasticity works in your favor when you start low. Going high then dropping looks desperate. Going low then raising looks like you've earned the increase.

Mistake 2: Charging Too Low or Free When You Should Charge

You build an AI feature customers love. You include it free. Now you're making $0 on something customers would happily pay $20/month for. You've subsidized innovation that should be revenue-generating.

Be intentional. Ask: "If we charged for this, would adoption drop below the threshold where it's worth maintaining?" If yes, keep it free. If no, charge.

Mistake 3: Complexity

You price "per token sent" and "per API call completed" and "per seat per month" and "minimum $50 per month but only on paid plans, not free, unless it's Q4." Customers don't understand. Salespeople can't explain it. You lose deals to competitors with simpler pricing.

Simpler pricing models convert 15-30% better, even if they're slightly less economically optimal. Simplicity is worth the revenue loss from imperfect price discrimination.

Mistake 4: Unexpected Bills

You have usage-based pricing and a customer gets a $500 bill they didn't expect. They churn instantly and leave a bad review on G2. Other customers see this and get nervous.

If you use usage-based pricing, have clear limits and warnings. Set monthly caps. Send alerts at 50%, 75%, 90% of their limit. Pause service at 110% instead of overcharging. Make surprises impossible.

Mistake 5: Ignoring Cost Structure

It costs you $0.01 in Claude API calls to serve a customer request. You charge $0.005 per request. At scale (1M requests/month), you're losing $5K/month. At 10M requests/month, you're losing $50K/month.

Know your unit economics. If Claude's API is $3 per 1M input tokens and your feature uses an average of 500 input tokens per request, that's $0.0015 in cost. If you charge $0.0010 per request, you're upside down. Price must be 3-5x your cost to maintain healthy margins.

Case Study: Document Analysis Platform

A Series A startup built AI-powered document summarization. They had 500 customers. Decision: monetize the AI feature.

First Attempt (Mistake: Freemium Wrong): They launched freemium. Free: 50 summaries/month. Pro: $20/month unlimited. They expected 5% conversion. They got 0.8%. Revenue: $800/month. Cost of free users: $2K/month. They were losing money.

What They Got Wrong: Free tier was generous (50 summaries is actually a lot for most users). Customers had no reason to upgrade. The value proposition wasn't clear: "Why would I pay if I get 50 free?"

What They Did: Instead of pivoting completely, they segmented. Free: 5 summaries/month (not generous). Pro ($15/month): 500 summaries/month. Enterprise: custom limits. New conversion rate: 8%. Revenue doubled to $1,600/month. They dropped free limit 10x but conversion went up 10x. Revenue per user went up, and free user costs came down.

Second Adjustment:** After 3 months, they had data. Power users needed more than 500/month and were bumping limits. They added "Pro+" ($50/month, 5000 summaries/month). Now 5% of customers upgraded to Pro+, adding $1,250/month. Total revenue: $2,850/month.

Third Adjustment:** Customers using it heavily were enterprises. They started direct sales. Custom enterprise pricing: $500-3000/month depending on volume. By month 12: 480 free users (low cost), 20 Pro users ($15/month = $300/month), 10 Pro+ users ($50/month = $500/month), 1 enterprise customer ($2K/month). Total: $2,800/month from 511 customers. More importantly: unit economics were profitable.

Key Lesson:** Freemium failed because of poor tier design. The fix wasn't copying competitors. It was understanding their own user behavior and pricing accordingly.

Testing and Adjusting Pricing

Start Conservative

Price lower than you think the market will bear. Build adoption. Build love. Then raise prices over time as value becomes obvious. It's much easier to raise prices than to lower them without appearing desperate.

Segment Customers

Different customers have different willingness to pay. Startups ($0-50K ARR): price lower, maybe freemium. Mid-market ($50K-2M ARR): standard tier pricing. Enterprise (>$2M ARR): custom pricing. You might have 80% of customers in startup segment but 60% of revenue from 5% of customers in enterprise segment. Pricing should reflect this.

Measure Elasticity

When you change price, does demand change? Elasticity = % change in quantity / % change in price. If you raise price 10% and lose 5% of customers, elasticity is 0.5 (inelastic, good, raise prices more). If you lose 20%, elasticity is 2.0 (elastic, bad, prices are too high). Most SaaS features have elasticity 0.5-1.5. Elasticity >1.5 means demand is very price-sensitive and you're probably overpriced.

Test Pricing Tiers**

Before launching new tiers, talk to 10 customers. "We're thinking about a $50/month tier with X features. Would you use it?" If 3+ say yes, build it. If none say yes, don't. This saves months of development on wrong pricing.

Get Feedback**

"We're thinking about pricing this feature at $20/month. What do you think?" Not "Would you pay?" (people say yes to everything). Instead: "At what price would this feature be obviously too cheap? At what price would it be obviously too expensive? What price would be just right?" Willingness-to-pay questions are more revealing than yes/no questions.

What to Do Monday Morning

• Define whether each AI feature is table stakes or differentiation.
  - List what competitors charge for similar features. Understand the market.
  - Survey or interview customers: "Would you pay for this feature? At what price?"
  - Calculate your cost: How much does it cost to deliver this feature per customer?
  - Set a price: Based on value, competitive benchmarks, and costs, what's the right price?
  - Create packaging: How do you bundle and present this feature? Free, premium add-on, separate tier?
  - Design limits if free: If you're offering a free tier, what's the limit (usage, time, features)?
  - Test and iterate: Launch at one price. Measure adoption and sentiment. Adjust.

FAQ

Q: Should we offer a free trial or freemium model?

A: Depends on your use case. Freemium works if: feature creates habit-forming value, free users can become low-cost advocates, you have healthy unit economics on free tier. Trial works if: feature is easy to evaluate (shows value in days), your sales process is short. Most AI features benefit from 14-day trial OR generous freemium (real usage limits, not nagware). Don't do both (confuses customers).

Q: How do we handle cannibalization? If we offer basic AI free, will that kill premium AI sales?

A: It might, but no free tier will also prevent adoption. The right answer: make free tier useful with real limits that drive some to upgrade. Example: 50 operations/month free, 5000/month for $20. Most users stay on free (50 ops is plenty). But heavy users hit the limit and upgrade. This is healthy cannibalization. You're converting free users who found genuine value. That's better than converting no one because you had no free tier.

Q: How often should we change prices?

A: Annual review cycle is standard. Don't change prices frequently (confuses and angers customers). But do review annually: "Have our costs changed? Has customer willingness to pay changed? Has competition changed?" Raise prices 5-15% annually if costs or value justify it. Most customers expect 5-10% annual increases and don't object.

Q: Should different tiers get different quality of AI?

A: You can but be careful. "Free tier uses Claude 3.5 Haiku, Pro tier uses Claude 3.5 Sonnet" means free tier is slower (Haiku is 5x faster but less capable). This can work IF free tier quality is good enough that people don't feel cheated. If free tier quality is bad, people never upgrade. They just leave. Recommendation: same model quality for all tiers. Differentiate on limits (operations/month) not on quality.

Q: What if a competitor undercuts us on price?

A: Don't immediately match. Understand why they undercut. Are they burning VC cash? Have they built cheaper infrastructure? Are they targeting different customer segments? If they're burning cash, wait them out. If they've genuinely built cheaper tech, you might need to do the same (reduce costs, not prices). Competing on price alone is a race to zero. Compete on value: better features, better support, better UX, faster performance. Then charge accordingly.

Q: Should we use seat-based, usage-based, or flat pricing?

A: Depends on what drives value. Seat-based: "Each user who accesses this costs $10/month." Good when value is per-user (licenses, concurrent seats). Usage-based: "Each API call costs $0.001." Good when customers use differently (some 1K calls/month, others 1M). Flat: "$49/month unlimited." Good for simplicity and predictability. Most SaaS started flat, moved to usage-based as they scaled. Usage-based is most "fair" but creates billing surprises. Hybrid is common: "Base $50/month includes 10K calls, then $0.001 per extra call."

Pricing AI features is about understanding what creates value for customers and capturing some of that value in revenue. Decide: table stakes (free) or differentiation (charge). Understand customer willingness to pay. Set pricing based on value, not cost. Create simple, clear pricing packages. Test and iterate. The goal: maximize lifetime value of customers while maintaining adoption. Most teams leave money on the table by being too cheap. Conversely, pricing too high kills adoption. The sweet spot: customers feel like they're getting value for money AND you're making good margins.

On This Page

Watch the Lecture
Core Pricing Models
Decision Framework
Packaging Strategy
Failure Modes
Common Mistakes
Case Study
Testing and Adjusting
Monday Morning Action
FAQ

Chapter Details

Part ofChapter 6