How to price generative AI products

Original source: SenseAI

Image source: Generated by Unbounded AI‌

Everyone seems to be in the exploratory period regarding the pricing strategy for generative AI products. The PLG product growth strategy we previously explained proposed several pricing strategies such as Freemium, while Miaoya Camera charges on the first day it goes online. How should generative AI products be priced? **Ian Clark, who once served as a pricing guru at YC, pointed out the common pricing problems of generative products today and gave a pricing method for generative products. His views are condensed and true. **

Ian Clark has over a decade of experience helping software and internet companies monetize their products. His clients include companies such as Y Combinator, LinkedIn, Eventbrite and Cloudflare. He is also the author of a book on profitability strategies.

It seems like every other day, a new generative AI startup calls and asks the same question: How should they set their prices?

Helping startups with pricing is nothing new to this author—Ian has spent years thinking about this issue in his work helping software and Internet companies monetize their products. However, questions from generative AI startups have increased recently. **Generative AI introduces an issue to software pricing that previous startups didn’t have to face: marginal cost. **

Marginal costs (such as the cost of running large language models) break the normal process of product launches in startups: "acquire users first, think about profitability later." Today's free model is already a dangerous user acquisition model. Subscriptions can result in high losses. But if AI startups follow best practices, pricing their products might not be so scary.

01. Expensively priced software

Let’s get right to the cost issue. How much cost should be considered when setting prices? The answer is basically no at all.

Imagine you run a lemonade stand and are pricing a glass of lemonade. Now, not only are you a lemonade entrepreneur, you're also a mind reader. My superpower is that you know exactly how much your customers are willing to pay for lemonade. There are 10 people waiting in line to buy lemonade: the highest price the first customer is willing to pay is $9, the second customer is $8, and so on, until the last customer is willing to pay $0 for my lemonade. I have a perfect understanding of the market. So, what’s the best price for your lemonade?

Assuming that different customers cannot be charged different amounts, the optimal price is $5. You can do the math yourself, but if for $5, five customers will make a purchase, I'll make $25.

If I told you that lemonade costs $1 per cup, would that change your answer? It shouldn't be. The best price is still $5. What if lemonade costs $4.50 a glass? You might say this is a bad business, but the best price is still $5. Cost has nothing to do with the price I should set.

Careful readers may point out that there is a real difference in whether we are trying to optimize for revenue or profit. Let us take this objection seriously. Our “revenue-optimal price” and “profit-optimal price” only start to diverge when your cost reaches $2 per cup. In other words, costs only start to have an impact when your gross margin is 60% or lower.

(Is your generative AI product’s gross margin 60% or lower? I hope not.)

02. Why willingness to pay is more important than opponent price

A common mistake AI startups make is always looking back to see what other companies are doing. Instead of focusing on the unique value your product offers, you spend all your time searching for competitors’ prices and scraping price pages. Competitive intelligence is almost always insufficient to answer your pricing questions.

Back to the lemonade example. Using the same example - 10 customers with willingness to pay ranging from $0 to $9. If I told you that the other lemonade stands on this street charge between $3 and $7 per glass, how should you price it?

The optimal price is still $5 (remember, the author is a mind reader and already knows each customer's willingness to pay). What if I don’t know my customer’s willingness to pay? Rival prices still don't help. Should I set my lemonade price in the middle of the market? Should I be above or below the market range? I know nothing.

I've heard your complaints. "What if a new competitor comes in and starts taking away all my customers?" Sure, that can happen, but you still need to listen to your customers. If you've been communicating with them, you should see your customers' willingness to pay drop when new competitors enter the market or lower their prices. Don’t lower your prices just because a smaller competitor has slashed theirs.

03. Correct pricing

Understanding willingness to pay is the first step to developing the right pricing strategy. Startups should conduct at least 10 customer interviews per month to assess willingness to pay. Once you understand which customers get value from your product, you can design a monetization strategy that encourages each customer to pay their fair share.

I like to start by choosing a good price indicator. Price metrics are essentially what you charge for: users, API calls, bytes stored, compute. The right metric is highly correlated with willingness to pay. Price metrics don't have to be cost-related, which is basically irrelevant when determining your price. This is a mistake many AI companies are making now.

Next, you want to think about price structure – how your price indicators relate to time and volume. **Pay in advance each month? Are you buying the number of users? Are the first 1,000 API calls included? How are quantity discounts considered? You'll want to adjust your price curve for different quantities to match your customers' willingness to pay. Try not to have an overly complex structure; after all, more complex structures are harder to sell and implement. **

The final step is to set the price, often referred to as the price level. But before you do this, make sure you’ve conducted enough customer interviews. What about surveys or sales tests? Forget about heavy A/B testing or large-scale market research for now. A "good enough" answer can be arrived at using some of the methods described above - and getting the indicators and structure right is more important than polishing price levels, which can be adjusted annually.

04. How to start pricing

I keep talking about measuring customer willingness to pay as if it were something you could measure with a tape measure. In reality, you need to use one of many "pricing methods." One of my favorites is "Van Westendorp," a technique introduced in 1976 by Dutch economist Peter van Westerndorp to help determine consumer price preferences. Ask customers what is a fair or cheap price for your product, as well as what is expensive and exorbitant. By getting three or four data points from each customer (some methods also ask about "too cheap"), you can estimate the demand curve for your entire customer base.

There are many other advanced methods for determining willingness to pay, but most startups can rely on the Van Westendorp method. Having one person lead pricing decisions is helpful because rapid iteration is key in startups.

AI products should be given the same good pricing practices as any other product – regardless of marginal cost or not. Focus on willingness to pay, not cost or competition. Set your price targets, structure and levels by interviewing customers, and consider using simple price studies such as Van Westendorp to determine willingness to pay.

Finally, ditch that “profit later” strategy. Put pricing first, where it should be.