Monday Price Point:​
Here are the three main questions I ask myself when thinking about pricing for any AI product:

​

1) Does the solution generate a single, repeatable outcome?​
​(or many different outcomes?)

2) Is the path to value clear to the customer?

(or does the solution require explorative usage to reveal the true value?)

3) Is there significant cost involved in powering the AI solution?

(or does the deployment cost about as much as regular SaaS?)

​

Each question is designed to clearly

​

Let's do an example:

OpenAI is a NO-NO-YES or NNY for short (whether it's ChatGPT or one of the API interfaces):

1 = NO: OpenAI has many different outcomes across a vastly different pool of customers.

2 = NO: The value of OpenAI is not immediately clear and has to be explored by each user.

3 = YES: there is significant infrastructure cost of running OpenAI.

​

Now let's do another example:

Intercoms 'FIN 2' AI customer support chatbot is YYY:

1 = YES: FIN answers support tickets instead of a human agent.

2 = YES: Support tickets are well defined with clear unit economics. FIN charges $0.99 per resolution against the average of about $5/resolution for a human.

3 = YES: Intercom has real token costs running FIN. My estimate is it costs ~ 20 cents to run a ticket, all in.

​

​

OK, so what does this tell us?

​

Question 1 tells us whether we can have an outcome based pricing or not. If the solution generates many outcomes we have to price the inputs.

​

This is what OpenAI does when it prices the tokens - a form of general 'currency' to measure the compute weight required to run the solution.

​

Intercom, on the other hand, can clearly price against the outcome of a resolved ticket.

​

The main difference is that OpenAI customers does not inherently *want* more tokens - they want whatever outcome they are after. Tokens are just a cost of doing business. But a cost they'd like to minimise as much as possible.

​

This is different for Intercom and FIN as customers actually *want* as many resolved tickets as possible (until Intercome invents a way to not have the tickets occur in the first place - an even better outcome).

​

Users will try to reduce tokens - byt tro to increase resolved tickets.

​

(If you're familiar with my taxonomy and previous work: question 1 tells us if the pricing metric should be placed at the beginning or the end of the customer value chain.)

​

Question 2: tells us whether customers can be sold full volumes up front or if you have to build significant expansion paths into your model.

​

OpenAI prices monthly in arrears on a usage-based pricing modality.

Intercom prices the same way - but with a 50-resolution minimum license.

​

The usage-based model allows users to more freely explore the value with low upfront cost and risk, which promotes adoption.

​

But since the value of FIN is so clear, it would be possible for Intercom (as indeed it happens in enterprise agreements) to charge on a license based modality (e.g. upfront prepay for 200K resolutions), which is essentially what they already are doing with the 50-resolution minimum.

​

This comes into play especially with larger accounts, who want a discount. If I bring you 1million tickets a year I'm not going to pay $0.99 for each.

​

So the play for intercom is to switch to a license model and charge, say, $350K/year upfront for the full committed volume.

​

This license modality becomes an option when volume expansion and value exploration is no longer really an issue because the outcome and volumes are so clearly known. Intercom can now switch the risk to the customer and collect cash upfront.

​

And crucially: this is how intercom (to my knowledge) approached enterprise accounts from day 1.

​

OpenAI does not have this option as path to value is unclear. It takes a lot longer for OpenAI to build enough confidence and consistency in usage patterns to shift customers to upfront payments of committed volumes.

​

(If you're familiar with my taxonomy and previous work: question 2 tells us which pricing modality to use - flat, license, usage or credit - and if we want to reinforce risk and expansion elements with terms)

​

Question 3: tells us whether there is a price to be paid for uncertainty and whether we can offer the solution with fully predictable pricing before we have enough data to clearly understand usage:

​

If I have clear costs I can not, for example:

• Offer the solution for free
• Offer via a flat fee model
• Offer the solution via a license model that doesn't scale well with costs (e.g. a per user model).
• Offer the solution bundled into another product where the pricing of that product doesn't scale with costs.

​

All of the above run the risk of massive usage = massive costs. Google Gemini and Microsoft can do this, because they have enough usage data to predict average costs (and fair usage limits in place to curb outliers). They have no uncertainty.

​

But if you have a net-new AI solution you're introducing to your 1000 enterprise customers?

​

Then you have both massive cost risk and massive uncertainty - because you have no idea how they are going to use your product.

​

And this uncertainty have to be paid by someone.

​

And that someone is either you (offer the solution with fixed, predictable pricing) or the customer (solution is usage based with a potentially unlimited price tag).

​

(If you're familiar with my taxonomy and previous work: question 3 tells us which pricing modality NOT to use and if we want to reinforce risk and expansion elements with terms)

​How to do it: Get together your senior team.​

Ask them the 3 questions as it pertains to TODAY:

1) Do we have singularity of outcome? Or plurality?

2) Is path to value clear?

3) Do we have cost uncertainty that needs to be paid?

​

Now ask if you expect the answer to be different 2 years from now?

​

Now discuss options for pricing models that take these answers seriously.

​

If you need a 'bridge model' that takes you to +2 years that is fine. But make sure your contracts allow the future change.

​​
​As promised: a point about pricing, every Monday.

​PS: if you have a great topic for this newsletter - just reply to this email.