Full Disclosure: We are angel investors in Clay and we have helped them with pricing in the past. We did not work with them on this latest pricing change.

If you’re newer to Silicon Valley, you may not have heard of this awesome company Clay that helps companies go to market with spreadsheets that have been filled in by AI. Imagine you give Clay a bunch of emails, and the AI goes to ~150 data partners and fills in all the rest of the information: job title, address, favorite ice cream flavor¹, etc. They also connect that data to other parts of the GTM stack; so maybe a lead came in and you want to send them an automated email with the name you just found - you can connect Clay to your email marketing system and voila.

It’s more complicated than that, but that’ll do for today’s discussion.

So what does this have to do with pricing? Last week, Clay announced their new pricing model with significant fanfare, including a YouTube Explainer with the founders, a very long internal memo detailing the change, and a partnership blog announcement through Kyle Poyar’s Growth Unhinged.

First off: A+ on the announcement. These guys nailed the playbook for launching new pricing by explaining the value it had to customers, telegraphing their product playbook, and having crystal clear language around pricing mechanics.

But this post isn’t about the execution, it’s about whether Clay’s new pricing could work for you, and whether we can generalize their strategy to other companies.

In Clay’s earlier pricing, customers purchased “data credits” at a markup over the data providers’ rates. That markup then paid for all of the platform value that Clay was providing. Now you pay for the data + what you DO with the data separately. If before you were paying “per data credit,” now you are paying “per data credit + per action credit”.

Data credits and AI flat fees used to cover the “platform value” that Clay was adding. Now that platform value is priced separately through “action credits”. Rather than absorbing the data and AI cost, Clay has unbundled it and passed that cost onto the customer. The prices for data and AI tokens are more or less what a customer could purchase from the providers directly.

Clay used to have 5 tiers, now they have 4. Features in the most advanced tiers have been pushed to lower tiers alongside a price decrease.

Clay says that 85% of customers will get a price decrease and the net revenue impact will be -10%.

We at Crescendo don’t like to “armchair quarterback” other companies’ pricing, especially ones with which we have a relationship (like Clay). What we will do instead is provide a checklist of Things That Have to Be Right for you, dear reader, to make a similar change work for your business.

Whenever you launch, change, or introduce a new metric, you need to ensure that the price metric passes the 4 tests of good pricing theory. One of those tests is “Communicability” - customers need to be able to understand the metric and not perceive it as “unfair”. Part of understanding is being able to connect the price to the value; in other words, how much am I going to get for my $100?

Clay has had a lot of trouble with this in the past. Customers complain that credit-based pricing is opaque. One customer hired a consultant³ to help determine their Clay budget. The consultant themself cost more than the entire Clay budget!

So, are “Actions” more communicable than “Credits”? Almost certainly not. We at Crescendo are very pessimistic of any token or credit-based pricing model, because it places a barrier between price and value and often is just another way to do “cost-plus pricing”, which is very bad.

Having two different price metrics requires that the two metrics do not correlate to one another. Otherwise, you are adding pricing complexity while not capturing different market segments.

Imagine that I sell apples. I price my apples at $3 per pound. Then I decide that $3 per pound didn’t cover all corner cases (big apples, small apples)⁴, so I decide to change my prices to $1.50 per pound of apples + $0.50 per apple.

3 apples weigh roughly 1 pound, so in the first pricing model I pay $3 for 3 apples, and in the new model I pay $1.50 (weight price) + $1.50 (apple price). But if I bought 3 tiny apples, I would pay more under the new model than under the old one.

But how often does this actually happen? Do we need all of that headache or are number of apples and total weight of apples correlated?

You probably guessed it, but they’re correlated. According to “the internet”, the standard deviation of apple weight is 10%. I don’t want to bore you with the statistics, but that means 95% of all single apple purchases using the complex model above would vary only ±$0.15.

We covered this in depth in our series on data analysis for price metric

So are “Actions” and “Data” correlated on Clay? Clay itself says that they are. Each Data Credit comes with 4-5 Action Credits, and in the team’s own words…

If 90% of customers are unaffected by a price metric, you probably don’t need to charge for it. Ok, but what about the last 10%? What about the customers who had 0 data (because they brought their own) and only used actions. For those customers, there are 2 options:

Which brings me to my 3rd test…

Imagine I have a developer tool that lets people build and manage their own apps….

On second thought, let’s make this “more 2026”…imagine I have a tool that helps you vibe code your own app. We’ll call it “Klaus”.

Some people build apps that do a lot of stuff, like process text or generate graphs. Other people build apps that store a lot of stuff, like a big database of customer orders. Here’s the question…should I charge for doing (aka compute) or storage (aka gigabytes) or both?

I would argue that in today’s world, storage of gigabytes is highly commoditized, and it is unlikely that Klaus can command any marginal customer willingness-to-pay with its storage solution. If Klaus charges for storage, customers can easily hook their app up to a different database provider, causing Klaus to lose that revenue. Klaus should therefore charge only for workload.

Now, what if I tell you that workload comprises two different actions: reading information and writing information. Knowing nothing else about Klaus’ customers, we might assume that some customers make apps that read a lot and others make apps that write a lot. Klaus would be equally valuable to both customer sets and the ratio between reading and writing workload probably does not predict willingness-to-pay. In that case, Klaus should charge for both reading and writing workload.

So, what should Clay do?

Again, we are not going to play Monday Morning Quarterback. Instead, we have to assume that Clay believes that both of the following customer types exist.

…and both customers have similarly high willingness-to-pay. To summarize in a consultant’s 2×2:

Probably not. It is highly unlikely that willingness-to-pay is separately correlated with two different metrics. e.g.

If you want to learn more about price metric - check out our library, especially the posts on finding the right metric with customer interviews and using data to assess price metric viability.

Crescendo works with medium-sized software companies to improve their pricing, packaging, and promotion strategies. If you’d like to book a quick consult, reach out at [email protected] or schedule time via the button below.