How to commercialize AI applications? 40 leading AI products tell you

The commercialization of AI applications is gradually developing a subscription-based model, among which free versions and pricing strategies based on the number of users are popular, but at the same time, it also provides broad space for innovative pricing models.

Last week, Crow shared two mainstream monetization methods for AI applications and analyzed the pros and cons of the two options.

Today, with the help of another article by Kyle Poyar and Palle Broe, I will continue to focus on the topic of commercialization of AI applications and introduce to you the latest trends in the commercialization of current leading AI products.

In this article, Kyle Poyar and Palle Broe studied the pricing models of 40 native AI application products, covering public information on pricing models, value indicators, public publicity, free versions, and pricing transparency, and summarized the pricing trends of some leading AI applications.

The criteria for selecting products are financing and industry influence, such as Forbes AI 50 list and Sequoia's Generative AI Market Map. The 40 companies include marketing tools (e.g. Jasper, Copy.ai), productivity applications (e.g. Tome, Glean), vertical-specific products (e.g. Harvey, Co:Helm), and other companies (e.g. Synthesia, HeyGen).

Through this report, we will be able to have a clearer understanding of the pricing trends of AI products.

1. Pricing trends of 40 leading AI applications

We found the following five points worth noting:

1. Limited Pricing Innovation – 70% of companies adopt a subscription model, and very few companies offer pure usage-based or pay-per-use pricing models.
2. Most companies charge based on the number of users — consistent with the concept of AI applications acting as “co-pilots” (assisting people) rather than digital “workers.”
3. Free versions were very popular in initial adoption – half offered a free plan, and another fifth offered a free trial.
4. There is a “good-better-best” paradigm when it comes to packages/tiers.
5. Pricing transparency varies – two-thirds of providers have open pricing.

You can explore the full pricing data yourself:

2. Lack of pricing innovation provides opportunities for the second wave of AI applications

Software companies have historically favored subscription and per-user models (although there are signs of a shift toward usage-based and hybrid models). This remains the case with the first wave of breakthrough AI applications.

We are seeing signs of very innovative pricing structures emerging among the second wave of AI companies. These pricing models can accelerate customer adoption while capturing more overall revenue, and even Microsoft is testing innovative pay-as-you-go pricing for its new AI Copilot for Security.

Fin (Intercom), EvenUp, Chargeflow (an OpenView portfolio company), and 11x.ai (previously covered in Growth Unhinged) are examples of companies that have implemented success-based (or outcome-based) pricing models, where customers only pay for successful outcomes.

Chargeflow's pricing is based on the success rate of refunds in the interests of the seller.

This payment method is attractive to customers because it creates a win-win partnership. Only when customers succeed can suppliers succeed. This payment model increases customers' willingness to pay.

This is in stark contrast to many existing SaaS providers, where customers often end up purchasing far more resources than they actually need.

That is, customers are paying a premium for what they get, so we expect that as pay-for-results models for AI products gain popularity, pressure on traditional subscription models will increase.

Finding 1: Limited Pricing Innovation

Of the AI ​​applications we studied, the vast majority (71%) use a traditional SaaS subscription pricing model. Ten companies (26%) use a hybrid pricing model that combines subscription fees with usage fees. PolyAI is the only company (3%) that uses a pure usage-based model.

While the infrastructure that supports these applications is priced almost entirely based on usage (LLM and infrastructure), this pricing model is not reflected in AI applications.

We think there are several reasons:

1. Keep it simple: The core goal of most AI applications is to get users to use the product. Users are more familiar with the traditional SaaS subscription pricing model.
2. Usage-based pricing is hard to implement. Usage-based pricing is not easy to implement. For example, it requires a strong data foundation, which many startups do not have in their early stages.
3. It’s still hard to quantify value: Some AI applications may well create more value in this way than the price they charge. But for many AI companies, it’s still hard to understand exactly how much value they’re creating and how best to capture it.
4. Don’t want to limit adoption: Pricing innovations can limit product adoption. Currently, startups want to get as much usage as possible when developing their products.
5. No focus on profitability: While the ZIR era is over, the goal for many AI startups is not profitability but rather proving they can make money and grow with their customers.

We do see some companies (especially in the marketing, video, and voice generation space) adopting usage-based pricing models, such as word count, video character minutes. Copy.ai is a good example.

Copy.ai’s example applies both a monthly subscription fee and a usage-based component applied in the form of credits

Finding 2: Most companies charge based on the number of users

The main value metric for AI applications is still user-centric. This is a very well-known value metric in the SaaS field and one of the most direct ways to buy and sell software, with high predictability for buyers.

There are about a dozen companies using payment models based on each user and their usage, or purely usage-based models using the following value indicators: credits, roles, video minutes, subtitles, or runtime.

Since AI will eventually replace human labor, a per-user pricing model may be counterproductive as the number of users will decrease over time. This creates a disruptive opportunity for the second wave of AI applications.

Finding 3: Free versions are popular during initial adoption

About 70% of the AI ​​applications we studied had a freemium model. The three types of freemium models we saw were:

1. Free version (47%): A “permanently free” version that usually has limited functionality but allows users to try out its features/product.
2. Free version with limited usage (3%): Free version, but the usage of the product is limited.
3. Free limited-time trial (16%): A free version with basic functionality but limited to 7 or 14 days.

AI applications often deliver value to new users quickly. As these companies continue to iterate on their products, freemium products can help drive early adoption and usage.

Freemium models are not so common in enterprise-oriented applications. These products usually require implementation fees and platform fees to use. It is understood that the enterprise freemium model seems to be a free trial, where customers can try the product for a period of time (usually 3 months) and then make a purchase decision.

Finding 4: There is a “good-better-best” paradigm in terms of packages/tiers

When we talk to early-stage startups, we often recommend starting with a variation of the good-better-best product structure.

It allows companies to differentiate their products based on their customers and create clear upsell paths. The number of tiers varies by company, ranging from two to five (including freemium and enterprise editions). In most cases, the differences between tiers are based on product features and usage.

Marketing strategies are usually formed gradually as the product matures and develops its features. In the early days, you usually don’t have much to market because you don’t know who your customers are or how to segment your products.

Browse AI, for example, offers five different tiers and a combination of subscription and usage-based pricing.

Finding 5: Different levels of pricing transparency

Currently, about two-thirds of companies publish prices on their websites. Transparent pricing tends to be the norm for apps targeting personal or professional consumers, rather than for apps targeting businesses.

Most enterprise AI applications don’t reveal any pricing details. They may do so for several reasons:

1. It’s a competitive space: competition is getting fierce, and companies don’t want to offer anything to potential competitors or allow them to compete based on price.
2. Customization: They customize their prices to each client’s specific circumstances and determine the upper limit of the price they can charge.
3. Greater flexibility: Many of these categories are still in their infancy and prices have yet to be determined. Hiding prices provides greater flexibility for future changes in suppliers.

With the rise of price benchmarking providers like Vendr or Tropic, it’s likely that this pricing information will become public over time.

3. Final words

We are in the early days of AI adoption. Many companies are still seeking product-market fit (even if they have raised significant funding) and want to prove market demand. Innovation in pricing models is hard and, understandably, was not a core focus initially.

The rules of the game seem to be: (1) make pricing predictable, and (2) don’t let pricing become a barrier to using your product. Here’s a framework for figuring out where to start.