Recap: Pricing AI Products

Hey, hey, we are officially live with this edition of AI Fridays. Excited to talk about pricing AI. So if you’re building with AI, how should you think about pricing it? We can also talk, Ian and I were chatting right before this about some pricing frustrations that we’ve had as consumers of AI. So happy to talk about those as well. Quick reminder for everybody joining, you can always submit questions in the comments. You can also upvote questions. So we have a bunch of pre- submitted questions. Thanks to everybody who submitted questions in advance.

Please go ahead and upvote any that you want to make sure that we talk about or add any as we’re discussing that you want to make sure get mentioned. As always, this is put on by OneGuide. So if you found us through your investor, fantastic. This will be available, the recording will be available on your investors portal afterwards. And if you are yourselves a PE or VC firm, portfolio ops person, welcome. Ian’s one of our experts on our network. Happy to make intros if you’re looking for a pricing guy, which is a perfect segue to Ian.

Do you mind giving your introduction? Yeah, sure.

So I’ve been in pricing for a very long time. I pretty much started my career in pricing and specifically in software. That’s what I do a lot of. Most recently, I was at Alpine Investors, which is kind of a mid- market private equity firm that does software and services companies. And I ran their monetization practice there for about five years. So I was doing all of the pricing for all of their companies. I ran pricing at a mid- sized tech company. I was a consultant with Simon Kutcher and Partners, which is the largest pricing firm in the world.

And now I run Crescendo. And so Crescendo is a boutique pricing consulting firm that helps primarily tech companies and tech- enabled services companies with monetization.

Awesome. All right, let’s get into it.

Let’s talk monetizing AI. Let’s talk monetizing AI. So guys, here’s the runner show here. I have a few topics that I think would be really great to cover, but there’s a lot of questions in the live stream of questions. And so I’m gonna take those as they come up because I’m sure that we’re gonna talk about them. But for the time being, yeah, let’s dig in. So the first thing, what’s going on with pricing and AI? So there are a bazillion thinkfluencers out there who are saying pricing is changing.

All of pricing is gonna be totally different now that we have AI. We’ve got the great pricing convergence. I don’t even know what that means. But there’s a lot of people out there, a lot of words that are being spilled in LinkedIn about how AI is changing all of pricing. And if I were to impress upon you one thing about AI pricing these days, it’s actually that you need to think about what is staying the same, because a lot of AI people are kind of overreacting.

And what we really need to do is just do good pricing practice and apply that to AI rather than say, hey, you know, her seat pricing is dying a slow death because as this person says on LinkedIn. So that’s what we’re gonna talk about today. So before we get started, there are seven components of pricing that we like to talk about. And the reason that we do this is because you really need to break pricing and all of what we call monetization into component pieces. Otherwise your head’s gonna explode.

So for example, you know, somebody said like, hey, how are traditional, here’s a question on here. How are traditional SaaS and B2B companies integrating AI products into their pricing structures? Right? Well, what do we mean by structure? Do we mean the tiers? Do we mean the usage? Do we mean who gets access to it? The promotion of it? So there’s a lot of different sub questions in there and you have to answer them separately and individually.

So we’re gonna go over these seven pieces and then talk about which ones are the ones that are really kind of big for AI these days. So I usually like to start with metric, which is that one in the second row there. So metric is what are you charging for? This is the biggest thing in AI pricing these days. And we’re gonna talk about that a good amount today because that’s the biggest, hottest topic in AI pricing. Are you charging for usage, for users, for capacity, for what, what are you charging for? Outcomes, tokens, that sort of thing.

Structure is how does the metric relate to time and volume. Do you have volume tiers? Are they flat volume tiers?

Do you have overages?

Do you have caps?

How does that work? Do you buy up front? You spend down a bank of tokens. That’s all structure. Level is just how expensive is it, $ 10 or $ 11. We probably won’t talk about that too much today, and it’s also the easiest one to walk through. Then there’s packaging and bundling. Packaging and bundling, the red bar down there, those are the intersection of pricing and product. Okay, we came out with a bot, who gets access to it? Does everybody get access to it? Do you have to be in our gold tier to get access to it? Do we sell it a la carte?

Those are all questions about packaging and bundling. And we might get to some topics about that depending on how people are talking in the questions, but that’s those ones. And then the last one is promotion. Promotion is how do we encourage non- customers to use the product because definitionally they have a lower willingness to pay. That is, hey, we’ve got this new bot. Do we offer it via free trial? Do we offer it via premium? That would be an example of promotion strategy.

And then all of that is wrapped in monetization strategy up at the top, which is, hey, should you ever throw out all of the stuff I just said and deliberately undercharge or overcharge a particular customer segment or a particular product for a strategic reason? Is there a reason you might want to deliberately sacrifice revenue in service of some other thing? So somebody on here said, how can you encourage adoption with AI? And I don’t remember, I don’t see the- I’ll find it, yeah.

Right, and I have a kind of a funny answer to that, which is easy, give it away for free. That’s the easy answer. But the better answer is actually you can encourage adoption in all of the six things below. So from a packaging perspective, you make it a buffet. You don’t fence it in any way. From a pricing metrics and structure perspective, you don’t have usage- based pricing, you have flat pricing. All of those sorts of things would encourage adoption.

And I found at least one person asked a little bit more about free tiers or credits. Do you have any perspective you can share on those types of tactics specifically?

So those are, so again, the reason we break this up is because those are two very different questions. So the question of, should you have a free tier? That’s a promotion question. So you have to examine, should we have a free tier versus a free trial in one area? And then the separate thing is, should we have credits versus should we have users or capacity or some other question? That’s a metric question. So they’re actually two totally different questions. Should we have a free tier versus credits? It’s not versus, you can have both.

You can have either. I think they may be asking free credits. Like, hey, we’re gonna give you 20 free credits to test this thing out.

Oh, yeah. So if you want, what you’re saying is, should it be 20? Should it be 30? Should it be 50, right? That actually is in the kind of like structural way of looking at it. How wide is that free tier?

Cool.

And there’s no right answer for when, should it be 20?

This is not gonna be any depends question, yeah.

Yeah, yeah, yeah. I mean, that’s one that you, there’s a way to answer it. And you look at your data and you see how, what is the inflection point in terms of how many credits people need before they start to actually use the product regularly. So that’s how you would figure that out. But there’s no such thing as like, 20 credits is the right answer.

First off- Yeah, that makes sense. So if you’re gonna give free credits, this is traditional like freemium or PLG type thinking where, okay, how long does it take to get them hooked? We should give them that much.

Totally, totally.

Yeah.

So the four things that I wanna talk about today, and we might not cover all four of these, but the four things that I wanna talk about are these four, which is really, there’s a couple of new things that AI is introducing. And the question is, does it actually break the model and we actually have to think about it? Or is it something that’s just more of the same and we actually need to use pricing best practices to kind of like understand what’s going on. So let’s talk about costs first.

So the whole thing with costs is that one of the problems with AI is that everybody is saying, hey, now I’ve got marginal costs. I gotta deal with the fact that my model costs me money. And you see these things, here’s a good example of this, right? Everybody’s freaking out about this. They’re saying, oh my God, my gross margins are compressing. Here’s a company here, it went from 80% gross margin down to 65% gross margin. Crazy, crazy compression. We gotta change the whole pricing model, right?

Now for this company, I deliberately hid the X- axis and I’m not gonna tell you who this company is, that’s for dramatic effect. And I will tell you who it is at the end and you’ll see why. So the thing with costs is that they matter a lot less than people think they do. So people think that costs matter, they really don’t. What matters is willingness to pay. And I’m gonna show you why that is. So with costs, imagine that I come out with like Ian’s newsletter, for what it’s worth, Ian does have a newsletter, but it’s free.

So I come out with Ian’s newsletter. And let’s assume that I have these 10 customers and I have this magic eight ball and I know exactly what they’re willing to pay for Ian’s newsletter, okay? And I’ve listed them right here and it’s 987- 654- 321, that’s what they’re willing to pay.

So, you can look at this and we can say, what’s the optimal price? The answer is $ 5. You can see it there that if I charge five bucks, five people buy it, we get 25 bucks. If I charge either six or four, we lose a dollar because of either too much or too little price. You can see that. If you really wanted to be fancy, you could go back to Calc 1 and solve that derivative over there on the right. Kate’s laughing. Kate did a lot of derivatives, I’m sure.

I did, but if you ask me to do it now, it would take a minute.

Yeah. The amount of times where I have had to look up how to take the derivative of this thing in recent years is embarrassing for two reasons. One is that I’m still taking derivatives, and two is that I can’t remember it. In any event, here’s the answer. It’s like, hey, it’s $ 5. If you remember what this looks like from a math perspective, it looks like this. You’ve got this, hey, that’s my price on the x- axis, that’s how many people buy it on the y- axis, and you get this inverted parabola. You say, great, the optimal price, the maximum of that is $ 5.

Sweet. Now, here’s the question. Does this change if I introduce fixed costs? Does the optimal price of $ 5 change if I introduce fixed costs? Here’s us introducing fixed costs. Here I’ve introduced $ 15 of fixed costs. The answer is no, the optimal price does not change. If I’ve got a bunch of rent, I’ve got a bunch of engineers working on this thing, I’ve got a bunch of all that stuff, the optimal price still doesn’t change. You should still charge $ 5. You’re looking at the space between the orange line and the green line, that is optimized at $ 5.

Now, the question is, does the optimal price change if we introduce variable costs, essentially gross margin? The answer is yes, mathematically, yes, but not as much as you think. Let’s take a look here. Here is the exact same parabola, but I’ve introduced 50 cents of cost for each one of these things. You see how the apex of that parabola is moving over to the right? It does change, but look how much it’s moving over to the right. It moves over to $ 6 there as the optimal price. When you hit $ 1. 50 of costs, that’s a lot of costs on a $ 5 product.

I’ve quantified that over here on the right with this Christmas colored chart over here. What this is saying is, what percentage of the maximum possible revenue could you achieve at different price points? If you hit the maximum and you see what’s important here, is that it doesn’t really change until you’re getting at 50 percent gross margins. My point here is that technically speaking, yes, by introducing costs, your optimal price does change, but not as much as you think. Most people don’t have 50 percent gross margins in their software product.

If you do, maybe you’re selling infrastructure, maybe you’re selling AI infrastructure, and then we got to chat. But most of these companies, again, they’re going from 90 percent gross margins to 80 percent gross margins. Everybody freaks out about the price, and the reality is you still need to price the willingness to pay, you still need to not take cost into account.

Yeah. Ravi brings up a really interesting question. We don’t know how much it’s going to cost yet.

Don’t even worry about it, even better.

Yeah.

They’re thinking, hey, we’re going to give this away for free to drive adoption, and then we’re going to get a sense of cost later. What you’re saying is, hey, if the costs don’t drive your gross margin down to 50 percent, you should keep probably giving it away for free versus increasing.

Not necessarily. What I would say is, you should, first off, giving it away for free is a strategic decision. But let’s say that I can tell you what is the optimal price for this thing that maximizes revenue.

Got it. If you’re going to pay more, you should charge more, but it shouldn’t be about this. Exactly.

It’s like, where is the point where we are maximally capturing value? That’s where I would recommend you price it.

Right, which is almost certainly not free.

Yeah.

Somebody is willing to pay something for it.

No, I’m embarrassed that I… It honestly took me a minute to get there, but this makes perfect sense. A, we might increase the price, but the increase should be about willingness to pay, probably not about…

There we go.

There we go. By introducing cost data, nothing changes.

Yeah.

All right.

And more importantly, the model doesn’t change. So do we need to start doing cost plus pricing? We need to charge for all of our usage. No. No, you don’t need to do that.

Right?

Charge for the primary way that you charge. Match willingness to pay. Don’t change the model. Don’t do whatever the great pricing convergence is. Just price to willingness to pay. So my dramatic reveal here, this is Microsoft in 2010. And that was when the cloud was introduced.

Yeah.

And the question is, did pricing… Did they suddenly change all the way that they priced based on the cloud? And it’s like, you know, in a certain sense, they did in the sense that it was more subscription- based. Sure. But it wasn’t… They didn’t go to cost plus pricing when the cloud was introduced. And no one would suggest that you should do that today.

Awesome.

All right.

All right.

No new questions are coming in on cost- based pricing. So let’s keep cruising. All right.

Don’t worry about your costs. That’s what I’m telling you. Okay. The second one is metric. We’re going to spend a bunch of time here, I’m sure.

Yes.

And we do have some questions on metric that I’ll pull up.

I’m sure we do. Everybody wants to talk about this.

Yep.

So metric. Outcomes, usage, are seats dying? These are all of the things that the blogosphere is worried about. So when we think about metric, there are four things that matter when you are picking a price metric. Those four things are that the metric has to be feasible. So you got to be able to track it and enforce it. It’s got to be communicable, which is that the metric has to… People have to be able to understand it and not get pissed off that you’re charging for it.

Yep.

The third thing is segmentable. So that is that it actually has to differ widely, kind of high diversity amongst customers. And the last one is valuable, which is that customers… It has to actually predict willingness to pay.

Okay.

So I’m going to take a couple of examples here. But what I want to impress upon you guys is that something like seats- based pricing versus usage- based pricing, that actually is going to differ based on the company and the use case of the AI, not AI in general. So there is no sense that AI should or should not be user- based pricing. It is only for your company should it be user or user- based pricing. So I’m going to run through four different case studies that have kind of a yes or no version of this. So a couple of these.

Here’s a company, top left, called Voltron. So Voltron sells RFP generation to government agencies. So what they do is that they’ve got a bot. It reads through all these things, and it generates an RFP for government. So their target customer is both the White House lawnmower and Lockheed Martin. Both of those have to make RFPs. Now, the question was, should they be user- based pricing or usage- based pricing?

Or, I guess, to throw a monkey wrench in there, performance- based pricing. So that’s…

Sure.

We’re going to talk about that one.

Yeah.

We’re going to come back to performance- based pricing. That’s my fourth piece that we can talk about.

All right.

Sounds good.

But let’s pretend that it’s users or usage. How many RFPs are you going to generate versus how many people are generating RFPs? So one question you might ask is, does the number of people generating RFPs, is that segmentable? As in, what’s the range in the RFP team? So we asked these guys, and they said, well, the White House lawnmower, that’s one person. Their RFP team is one person. We said, well, how many are at Lockheed Martin? You guys want to take a guess? It’s 12.

Yeah, that scales pretty linearly, I guess.

Right. Exactly.

Right?

Not at all. So you can’t do user- based pricing. It’s not segmentable. Right? It doesn’t work. Now, let’s take the opposite version of this. So the other one is Seismic. So Seismic, what they make, they make an AI tool, among other things, but they make an AI tool that helps salespeople with sales enablement. So imagine that I’m going to generate pitch decks, and I’m going to talk to the AI tool and practice my pitch, and I’m going to do a bunch of other stuff with this AI tool. So you could say, well, okay, it’s making every salesperson better.

So maybe that’s a reason to price per salesperson. But also, if you’re using it a lot more, if I’m generating a lot more pitches, for example, maybe that’s what I should price for.

So the question is, which of those two is the right metric? So what we looked at here was the valuable piece. So remember I said, going back a slide here, it needs to, that last one, valuable, it needs to predict willingness to pay. So you could look at this and you could say, well, what’s going to predict willingness to pay of this product? Then you could measure that if you want to. Is it users or usage? Turns out that users, as in the number of salespeople you have, is a really good predictor of how big your company is.

Really good predictor of how big your company is, as in how big your wallet size is. And that is actually a better predictor of wallet size than how much you use it. When we talk about pricing to willingness to pay, it is both wallet size and utility, both of them. And the intersection of those is what we’re trying to hit. The intersection of those two is what we’re trying to hit. And so pricing to wallet size in this example was a better predictor of willingness to pay than pricing to usage. If you had high usage but low wallet, no willingness to pay.

High wallet, low usage, yes, willingness to pay. So these guys, user- based, user- based pricing. So that whole idea of user- based pricing is going away. No dice. This one should have been user- based. Third one, we’ve got usage or capacity here. So this is a company called Black Tiger. Should I price for all the usage that you do? These guys, essentially, they clean up your data. So should I do all the usage that you do every time you hit clean? Great. You’re going to get a ding for that. Or is it just how much data do you have?

How much data am I cleaning all the time? The question here was, again, which one of these is best correlated with a willingness to pay, number one? And number two, which one of these things is really driving the value? And so the value of this product was keeping all of your data clean. Not the act of cleaning it, but keeping it clean. And that scaled better with the amount of data. So that’s a capacity question, rather than the usage. So it’s not every time you hit clean, it does it again. But it’s how much data do we have to keep clean?

You let us worry about how long, how many times we keep it clean. So that’s going to be more capacity. And then the last one was this tokens one. So there are two funny things with AI. One is tokens, and one is outcomes. And we’re going to talk about both of them. But here’s one with tokens. So this is one of my favorite companies out there. This is an AI psychic. And I love these guys because every, first off, they’re one of the most sophisticated companies out there in terms of data science and whatnot.

But I also just love talking to their customers because it’ll be things like people didn’t like the AI psychic because it wasn’t giving real predictions. And stuff like that. It’s just I love this company, in any event. It’s a psychic marketplace. And you’ve got this AI bot that comes in and gives you psychic predictions. And the question is, do you have tokens for it? Or do you price for time? So the problem with tokens, in this case, is what is it a token of? A query?

With a number of predictions, of course.

Number of predictions. And that question just ends up being too difficult for users to understand. And so it’s a lot easier to say, hey, you know what? We’re just going to fence this. You’re on the free plan. You get five minutes a week or whatever it happens to be. You’re on the paid plan. You get unlimited minutes per week. And pricing for tokens just complicates things. So let’s go back. And I’m going to skip a section here. We’re going to talk about packaging.

But I want to talk about those two questions that came up, which was the tokens, number one. And number two was outcomes. I’m going to show you two versions of outcomes that I think are kind of an interesting way to think about this. So the first one is outcomes- based pricing can get really good press. But it’s not clear yet that outcomes- based pricing is going to actually win.

And the reason I say that is the problem with outcomes- based pricing is that even if the metric is objective, as in we reduced this metric or we improved this metric some amount, even if the metric is objective, what’s almost always in doubt is how much you improved it, how much you drived. And so I am just generally skeptical, even when it gets good pricing like this. This is Sierra, by the way. Even when it gets good pricing like this, I’m just always very skeptical that outcomes- based pricing is actually going to work for most customers.

The other question I have is about predictability of cost. How often do you run into…

clients, your clients, clients that care a lot about predictability of how much this is going to cost. Oh man.

Kate, I’m so glad you mentioned that. We’re going to go back to this. Predictability is a structural question. So we can solve the structural question, for instance, by having super wide volume tiers. We have three outcomes. We’ve got, we did a lot of damage. We did minimal damage. We did low damage.

Yeah.

That’s super predictable, but it still is outcomes based pricing.

Got it. Okay.

We got to think about them independently.

Okay. Yeah. All right.

Um, do, do, do, do, do, do, do, do. Okay.

I think we have a follow- up question. I’ll throw this one in. What if you can measure outcomes exactly? For example, if you’re setting up meetings for the client and you charge according to every meeting you set up.

Yeah, yeah, yeah, yeah. So, so, um, that, first off, my question might be, how is that different from usage?

I guess it’s how many meetings you tried to set up versus how many you succeeded in setting up.

Exactly. But, but still that still is usage, right? Like how many, I think about cost per click or something like that, right? Cost per click is still usage. It’s not really outcomes. It’s still usage. Every time somebody clicks, you get a ding and, or maybe in volume tiers or something like that. Right. So with, with outcomes, what I think about with outcomes is like.

Revenue. Yeah.

Or like we reduced your customer acquisition cost or like how many successful meetings you’ve set up. And the way that I like to think about this is AI is probably a proxy for human labor. And it’s just a scaled version of human labor. So I think the question you should ask yourself is, could I, as a human price to that metric or would somebody say that is actually not what I want. So if I, as a human could say, Hey, Kate, every meeting I set up for you, I’m going to take a cut of that. I’m going to ask for, you know, 10 bucks or something like that.

Every time I set up a meeting for you, that’s, that’s like reasonable. It’s not crazy. Kate might not like it, but it’s not that crazy. Right. And you know, but again, that maybe that’s usage. Let’s think about Sierra though. So Sierra prices for every solved customer interaction. So imagine I am Kate’s personal customer service agent and I handle all of Kate’s customer service. They got to go through me before they get to Kate.

And the way that the way that we price this is I say, Kate, every time I talk to a customer and they don’t get to you, I get money. Might you think pretty soon that I’m just being the world’s worst customer service agent and not letting people get to you. Right. It gets a little dicey. Yeah. The other one, Mike, a good example of this was we worked for a company that did AI pricing for grocery stores. Not, sorry, not pricing for, we were doing the pricing. They did AI reordering for the grocery store.

So imagine like, Hey, let’s optimize when you need to reorder raspberries because we’ve got all the purchase data and we got how long they go bad. You know, we’re going to optimize that. And there’s this measure, there’s this metric that they have in their financials, a grocery store does called shrink.

Okay.

And that’s essentially just how much food did they lose due to spoilage? That’s that measure. That is an objective measure. It is very clear what that measure is. And you could say, here was your shrink beforehand. Then we implemented the AI and here was your shrink after. Look how much money we saved you. The question is, can you price to that?

The problem with it is they could say, well, I see that our shrink changed, but what percentage of that change can I attribute to your software versus my new management practices that came along with the software versus the great weather we had this year. So what you end up having to do is find a different metric that is not outcomes- based and it’s still there. You know, you still have to go back to what we talked about before. Feasible, communicable, segmentable, valuable.

Awesome. Okay.

The last one here is tokens. And we’ll talk about Cursor here. We wrote a big post about this one.

Yeah.

This was the wrong kind of press. So Cursor changed a lot of different things about their pricing. So the first thing is that they changed from a speed metric to a usage metric.

Okay.

So there’s an open question about which one of those things would be better anyway. The other thing, though, is that they introduced tokens.

Okay.

And tokens, the problem with tokens is that tokens are oftentimes just a mask for cost plus pricing.

And you’re just going to say, here are all of my costs.

Let’s add them all up. Here’s a token.

And the question you have to ask yourself is, are we actually giving anything to the customers by doing token- based pricing? Or are we just kind of trying to simplify our own costs? And our argument for Cursor is that it was the latter, that they were just trying to simplify their costs, price to cost. And we already talked about why pricing to cost is not what you want to be doing. It generally just kind of makes the price a lot more confusing. You say, well, how many tokens do I need? It doesn’t actually solve any problems.

So we’re generally pretty bearish on tokens. And we’re also pretty bearish on outcomes.

Got it. And just to fully answer this person’s questions, key challenges of token- based pricing are?

I love it. Let’s talk about it.

Number one, communicating what a token gets you in terms of value.

If I say, let’s say I have a database product.

I say one token equals 100 reads, 10 writes, this much bandwidth, this much storage, all of these things. That’s one token. Well, now I got to figure out, OK, how many tokens do I need as a customer given that? That’s problem number one. Number two is, why is it 110 and this and this and that? Give me some reason why it’s that ratio that is other than cost. Right?

Certainly, you can’t.

And so you end up doing cost- plus pricing rather than value- based pricing, which is what we want to do.

Cool. Confusing and cost- plus. Awesome.

Sweet.

All right. So I think we have some PLG and freemium and adoption questions we can get into, or I guess we’re listening. What?

Let’s do some of them, yeah.

Cool. So one, when should enterprise and B2B companies consider freemium pricing for new AI features?

All right. For this, I’m going to switch slides. So hold on one second. How do I stop sharing?

Stop sharing.

And we’re going to go over to this guy.

OK.

Come on. Sorry, folks. Trying to share my PowerPoints here. There we go. That was the problem. I was presenting two PowerPoints at the same time.

I can’t do that.

No, streamer must not like that.

Way, way too complicated. One PowerPoint at a time.

All right.

We’re getting there. I still don’t have.

While he’s looking for that, feel free to share in the comments what kinds of products you’re pricing, or if you have specific questions about your products, I’m happy to weave those in. And we can get to specific questions as we go through this on stuff that’s related to the products that you’re pricing.

All right, guys. I’m going to do something a little bit crazy. I don’t know why it’s not doing this, but we’re going to make a.

We can do that. We’re going to have the.

Whoa, whoa. Never mind. We’re not going to do that.

OK. No, it’ll work if you share a different screen.

I’m just going to share this, and we’re going to go with that. So hold on one second. I don’t know why it went away. It was so good, and now it’s not. OK. So here we go.

So here’s the framework for figuring out how to do freemium versus free trial. So there are generally four types of promotion strategies that are free, and there are two types that are paid. I’m not going to talk about the paid ones here for a minute, but in case you’re wondering, it’s paid pilot and pays you go. So it all has to do with how value accrues to the customer over time. So when you have gradual value accruing to the customer over time, then you would do a freemium.

So that is, hey, I have unlimited time.

I have unlimited functionality.

I have limited usage.

So that would look like the first 10 questions are free. I don’t care when you ask them. That would be a freemium for AI. The alternative to that is a free trial. So that is where value pops on day zero, and then it’s flat over time. It doesn’t really grow over time. That’s free trial. You guys all know what that looks like. Deep discount is where I have to expend some sort of energy to get this thing to work.

I either have to do some integration cost or I gotta start talking to it and give it all this context or something like that. That would be a deep discount where you say, hey, we’re not gonna charge you for that cost, get it all set up, then we’re gonna charge you.

That’s a deep discount. And lastly, there’s a reverse trial.

That is where you can accelerate the time to value by adding more functionality.

A reverse trial is that you get gold for the price of bronze. So you have to have different tiers in that situation. You have to have a gold version and a bronze version. But it’s when you convert, you get gold for the price of bronze.

For most AI products, you’re debating between a freemium and a free trial.

That is most AI products.

You could argue that if you have tiers of features in your existing product and the question is, where do you put the AI? That could be a reverse trial question. But the difference between freemium and free trial, it has to do with how does value accrue to the customer over time? So do they gradually increase their usage of the AI? Then maybe you could look at a freemium. Does their usage kind of stay flat over time? Once they’ve figured out how to use it, they’re gonna use it exactly the same way in perpetuity.

That looks more like a free trial.

You should check that with your product and your company because it may or may not actually translate from company to company. But you can see it in your data.

Another awesome question. What are the best practices for implementing and learning about willingness to pay during a pilot or GA phase for enterprise AI products?

Okay, so willingness to pay, there’s five ways that you can get at willingness to pay.

And some of them are relevant for what you said and some of them are not.

So the five ways you can get at willingness to pay.

Number one is data.

You can look at your data analysis. If nobody is using the product, you don’t have any data.

So you can’t do that, right?

But data. Data is really good at getting relative willingness to pay over time or across different segments. It’s not good at getting absolute willingness to pay. So if you think you’re gonna kind of look at your data and get the absolute price that you should charge, probably not.

The second one is customer interviews, also prospect interviews.

So customer interviews would be for people who use the product and the AI. Prospect is for new ones. Essentially, they’re the same thing.

It’s only just when you’re entering a new market, you need to do prospect, not customers. But that’s the best way.

So when you’re trying to figure out, hey, how much could we charge for this AI? What are the use cases? Is value gonna grow over time? Is it gonna be flat over time?

All of these questions that I’ve been talking about, customer interviews. We can talk about how to conduct a customer interview. But customer interviews is like the number one way to look at that.

The third way is surveys.

That’s number four out of five. So surveys.

Surveys are really wonderful because they get a high enough N where you can actually get like a demand curve out of a survey. That’s really great, but you have to have some sort of PLG motion. You have to have enough customers where that N is gonna be meaningful.

And here we’re talking like thousands of customers, like 5, 000 plus customers is what you would need to have to do to run that kind of survey.

So you really gotta be in the kind of like SMB, PLG motion. And the last one’s A- B testing and experimentation. And the problem with A- B testing is that it’s really expensive. And it’s expensive in terms of time and expensive in terms of your engineer’s distractions. Think about if you wanted to run an A- B test to see, hey, should we be a freemium or free trial? First off, you gotta code both of those up. Second, you actually have to run the A- B trial for at least the amount of time of the freemium and free trial.

So that could be seven days, but it could also be a month. Then you gotta get enough data to actually make a decision. And then you gotta make a decision.

Or you could talk to 10 customers next week.

Got it.

I’m very curious to get the answer to this question myself. How do we think about AI pricing so that the AI doesn’t become a bolt- on, but instead drives adoption and expansion and measurable business value for customers?

So the first, there’s two pieces to that. So let’s talk about the second one first.

So not all features drive measurable business value for customers. So the first thing you have to figure out is does your AI actually drive business value for customers? Essentially, do they have reasonable willingness to pay for this AI? Are you solving a pain point for them?

If you’re not, go back to square one, build a better AI, right?

Well, or I guess, are there ever cases where AI improves usability of your existing product? And it’s not necessarily so much that they would pay more for it, but it does improve your existing product in the same way that a UI UX refresh might improve your existing product.

Sure, yeah, absolutely. So what that would look like is high popularity.

So lots of people use it. Lots of people like it. Low willingness to pay. And when you have that situation, what happens is you usually give it away for free.

If you do anything else, that’s like my favorite example of this is dark mode. Dark mode is a good example. Pretty much nobody wants to pay for dark mode, but lots of people like it. So go ahead.

But your AI might be actually driving a lot of extra value. So I talked about the RFP one, I talked about the sales enablement one, those guys actually drive a lot of value. And there’s real willingness to pay for that AI. Well, if there’s real willingness to pay, you gotta think about essentially, what’s the P and the Q for that willingness to pay? How many people want it, and are they willing to pay a lot or a little? If lots of people want it, and they’re willing to pay a lot for it, it’s a new product.

If lots of people want it, and they’re not willing to pay very much for it, you include it anyway, or maybe it’s a minor upsell. Just kind of like, it’s what we call a filler in pricing problems. If very few people want it, but they’re willing to pay a lot for it, that’s what we call a killer. That’s usually an add- on or an a la carte item. Sometimes you sell it in gold, and we can talk about when you would do that or not, but it kills bundles, so you wanna sell that a la carte.

And then lastly, is not that many people need it, they’re not willing to pay that much, we call that a dud. Hey, feel free to give it away for free, unless it costs you a lot of money, and then maybe turn it off.

Awesome, we’ve got a bunch of new questions coming in. So one of them, freemium and free trial, what if there’s a heavy service component? So setting this thing up for your client is expensive for you.

Yeah, okay, all right. So wait, wait, expensive for you, or expensive for the customer? So that’s one thing to think about. So let’s pretend that it takes a long time to set this up. Palantir, Snowflake are good examples of this, right? It takes a long time. That would be in the deep discount version of promotion. So neither freemium nor free trial, it’s, hey, we’re gonna set this up for you, and then we’re gonna charge you. That’s a good example of deep discount.

For what it’s worth, most software companies get this wrong, they charge implementation fees. And there are very few times when you should charge an implementation fee, but that’s not one of them.

What are the ideas when you should?

Okay, yeah, so my example of this that I showed earlier in the promotion slide was Verizon. So Verizon knows that it’s a pain in the ass for you to go and change your phone carrier. You gotta go in, you gotta give them your social security number, you might have to tell all your friends you got a new phone, you gotta port all your data, it’s a huge pain in the ass. And so because of that, they give you half off on your phone plan for next year, and they give you a free iPhone to do it.

Okay, what if instead Verizon said, actually, we have an implementation fee?

Yeah.

It’s the opposite.

Yeah.

It runs completely the opposite, right?

Yeah.

They go, ah, but our iPhones cost money.

Yeah.

Of course they cost money. We’ve got costs.

Yeah.

You don’t think the iPhone has costs?

Yeah.

Right, so stop thinking about your costs. Think about the willingness to pay. So when are the good reasons to do it? Let’s start with the bad reasons because the inverse of that is the good reasons. So the bad reasons are, they have to do it, they can’t use it without implementation, and you’re the only one who can do it. There’s no third party and they can’t do it themselves.

Got it.

If you have those two situations, you really can’t charge for implementation or you shouldn’t.

Yep.

If however, there’s a marketplace, anybody could implement this, but we’re better at it. We’re faster, it’s our software, we know it. Feel free to buy one of the other implementers out there. Okay, that’s a good reason to charge for it. Or you could do it yourself, or you could pay us to do it. That’s another really good reason to charge for it.

Got it. So in the AI case, in the case of this question, what if, so spinning up new AI for a particular customer, let’s assume that there’s a significant professional services configuration for me to be able to set this up for you to use in the way that will be compelling. And it’s gonna take 50 hours of my team’s time to set this thing up. I don’t wanna do that unless you’re really gonna use it and you’re really bought in. Like how do I deal with that situation?

Yeah, so I’m gonna say there’s two ways. And really what you have to figure out is how much red tape is there in the purchasing process. So the like straight down the fairway, correct way to think about this is the customer doesn’t care that you have costs. So suck it up and do it. Right? So the customer doesn’t care. You should charge for what they care about, which is only charged when they start getting value out of it. And then separately deal with your costs on your own. Okay, that’s the like straight down the fairway.

There are a couple of situations where that doesn’t work. And so remember I said that there’s four free trials and there’s two paid trials. So this is a paid trial. So essentially that’s a paid pilot and you have to get your customers to have a skin in the game and you want them to go. Essentially you do a paid pilot when one of two things is true. Number one, the red tape for getting the contract over the fence is so significant that you want to make sure that contract is going to get over the fence before you even start. That’s one. Okay.

So you better be selling to like, go.

Governments for something like that. Yeah, crazy enterprise. Palantir would be a good example of this. The second example is if there is so much loss aversion to taking this away that you think that by doing a paid trial you’re gonna have so much skin in the game and you’re essentially playing with psychology where it’s like put your first foot down and then you are gonna continue the whole thing. I i don’t love that reason for doing it.

Um, because there’s just very few examples of it’s more in consumer, where that’s true, like putting a credit card down to try it. In business, it tends that tends to not be as big of a reason, but the red tape one is a real reason.

Awesome. Um, i’m gonna keep rolling with people’s questions, keep doing it in the chat. Any best practices for experimentation with multiple pricing and packaging models in parallel.

It’s really hard to experiment with multiple ones in parallel because you are needing to code them up and you need to ab test them and all of that. You can do it with a couple of things. So here’s there’s there’s easier and there’s harder ways to do this. So, first off, level is super easy. You can maybe test level to your heart’s content. Three dollars, five dollars.

Fine.

Um metric is sort of uh sorry, structure, which is how your volume tiers work. That’s a little bit on the easier side. So what you can do is you can say: you can go to customers in a sales motion, for instance, and you can say: hey, sales motion, we got two models, we got pay as you go or we got tiers. Which one do you want? And you’re indifferent to what they choose because the prices are going to be the same on average. But the customer might have an opinion and that’s going to tell you which one they have higher willingness to pay for. Um.

My example for this is like fixed versus variable rate mortgages. The bank is indifferent to whether you choose a fixed or variable rate mortgage, but the price difference in between them is exactly what they’re what they’re pricing. They’re pricing your willingness to pay for um. For the structural change between those that structure packaging is a lot harder. So packaging is: how, like how many tiers do you have? Or is this thing sold a la carte? Where do the features go in the packs?

That is super hard to have in parallel because you’ve got to keep that skew up for a whole year. So i don’t love doing that. The one exception to that is you can deprecate a package. So if your standard is three packages and you think the middle one is cannibalizing, you can get rid of that one in an ab test. And that does. That does work. Um, in general- you’re gonna hear this as a refrain- i don’t like experimenting with pricing and packaging models.

I think the best way to get pricing and packaging information is to do the primary research or to look at your data.

Yeah, right, uh, all right. This is a vague question, but an interesting one. I think we can take it a couple of different ways. So how should ai products be priced to maximize adoption? And i like there’s- at least there’s- several sub parts, but two that i’m interested in are: um, one, what if you want to maximize adoption of your existing product using the ai? And two, uh, what if you want- like you want- to eventually make a lot of money off this ai product, but you’re willing to subsidize that in the beginning?

Okay, so before we even get down the road of how, you should ask: why, why do you care that you are maximizing adoption, let’s pretend, of the ai product? And we generally find that we only have three good reasons to care about this and every other reason is 3d chess. That doesn’t pan out because the the default should be: how do you price ai products? To maximize revenue? Not adoption revenue, right, that’s what you should be thinking about. But let’s pretend you actually do want to maximize adoption and you do.

You want to sacrifice revenue in turn in service of that. Good reasons, number one: network effects. If, for whatever reason, the use of your ai product makes your product more valuable to other people because they’re feeding it, answers back and forth or something like that, that’s a reason to do it. The other reason is economies of scale. So if, whatever reason, you get great rates from your ai provider, if you get tons of people using this, something like that, that would be another reason to do it. And the third reason is stickiness.

So that is that if we have the ai product and people use it, then you can raise the price on your other product or the core suite or retain more people. Yeah, exactly, exactly. Um, i would encourage you to do the math of how much revenue you sacrifice and how much price you would have to increase to make those even out. It doesn’t usually even out, but okay, now you’ve decided we want to underprice our ai product. We want to deliberately under monetize ai, which is what you’re asking. How do you do it? To maximize adoption, okay?

Well, to maximize adoption, you give it away for free. You pay people, right, that’s maximizing adoption. But really, in each of those seven components, or the six below strategy- because strategy is that you’re choosing to maximize adoption- there’s an answer.

So

For metric, what you want to do is choose something that is already standard with the company, and the AI has nothing to do with it. You would choose, for instance, users over usage, for example. For structure, the really good way to do this is make the structure as flat as possible. Flat tiers rather than variable tiers. That would be an example. For level, as low as possible. For packaging, as monolithic as possible. You want to make it a buffet. You want to give it away to all people. Bundling is the same thing. Promotion, promote it highly.

Cool. Here’s another interesting one. Most B2B SaaS companies are valued on an ARR multiple. How should that factor into the way that you price AI? Yeah, it’s a math question. Yeah.

So in different pricing models, different pricing models, let’s pretend that you have a flat plus a variable rate. You’ve got platform fee plus, let’s pretend that you have that pricing model. Well, the flat is ARR. The variable is not ARR. Those are valued at different revenue multiples. That’s just multiply one times the other. That’s going to give you your total valuation. That is a trade- off. Now, the tricky thing is, customers might have a different willingness to pay for subscription revenue versus variable revenue.

They might, for instance, dislike subscription revenue. They might say that’s not fair as they want to pay for the variable revenue. So they might have a higher total willingness to pay for variable than subscription. That’s the trade- off that sometimes you have. Where this runs into, the biggest one that I see is actually with CapEx and OpEx. So one- time fees versus recurring fees. The question is like, hey, what should the ratio of my subscription versus my buy outright price be?

And that oftentimes comes to your needs for the right valuation against your customer’s needs for their own budget cycles. And it’s just a math question.

It’s not a problem. All right, we’re going to keep cruising through questions. But while we do, go ahead and give us feedback on this session and other AI sessions that you want to talk about next. This will go into our AI Fridays programming. And we’ll keep cruising through questions as we do this.

Keep on coming, get at it.

Yeah, so pick and choose models where you can place the customer based on where it will promote more growth, consumption versus users versus agents. I’m not sure exactly what this means. Julia, can you provide a little bit more background or Ian, do you know what this means?

So I think this is like the customer chooses what they want, whether they want to be priced on users versus agents versus consumption.

Yeah, let’s assume that.

Okay. So the problem with this is that customers are going to self- select into the one that’s most advantageous for them. And there’s no sense that they are pricing their risk. So unlike fixed versus variable rate mortgages, where you say, hey, self- select, and I don’t care. The customer is self- selecting, but oftentimes they’re essentially overpaying for the thing that they want. As in, if I want a fixed rate mortgage, I’m paying more for that privilege to have a fixed rate mortgage.

When you’ve got consumptions versus users versus agents, there’s no sense that one of those things I’m going to be like overpaying for, and they’re just choosing. My take on this is, I’m going to read into your question, Julia, and you’ll have to forgive me. It sounds like you don’t know which of those three things correlates best with willingness to pay. And you’re trying to decide or say, why don’t you choose customer?

And my take is, probably go and figure out which of those three things is most predictive of willingness to pay and charge for that one. I’m reading into your question there, but sorry.

One more, or possibly a couple more.

Possibly a couple more.

Do you have any data offhand showing how much a free- to- start performance- only pricing increases willingness to pay for a product? Or does it?

How much a free- to- start performance- only- based pricing increases willingness to pay for the product?

Sure.

So we’ve got that mixed up. So the willingness to pay for the product is what it is. The structure by which you capture that might be free to start, but it might not be free to start. So by having a freemium model, I don’t increase the willingness to pay for the most part. There’s some exceptions to that, which is like, oh, by giving away the value, now they see it, now their willingness to pay rises because now they’re a current customer. But really what we’re talking about here is what’s your promotion strategy.

Well, or I think the question here might be not, I guess, willingness to pay, but how much you get paid. So I imagine like if I were hiring a recruiter and they said, you know, I’m hiring a recruiter to hire a product manager. And they said, cool, I will charge you either $ 5, 000 upfront or 20% of the success fee. And either it’s $ 5, 000 whether I succeed or not, or 20% of their salary if I succeed.

I might be more willing to take the 20% and pay only if they succeed and end up paying them much more versus if I were to- Yeah, essentially your question, like let’s pretend that Kate has a willingness to pay for this thing, right?

And the question is which of those two models is more likely to capture Kate’s willingness to pay better? And more importantly, which of them is more likely to capture all of the Kate’s varying willingness to pay better?

Yeah.

Right, so like, again, we’re coming down to, well, is the success fee model going to be better correlated to all of the Kate’s willingness to pay? We have a market of Kate’s versus the flat rate. Is that gonna be more predictive of willingness to pay?

I guess it also depends a fair amount on how successful, what percent of the time you’re successful.

Well, no, so interesting. That is going to change the level, but not the metric.

Okay, yeah, well, yes.

So let’s pretend like a crappy real estate agent might be only to take, be able to take 3% or something like that, but a great real estate agent can take 5%, but both of them are taking a percent.

Got it. All right, I’m looking through for kind of the last one to two questions that we haven’t hit yet that I want to. Okay, we’ve already covered a fair amount of this. How are companies approaching pricing based on outcomes versus inputs? The part that’s interesting here is variation by industry. And so within the world of B2B software, any trends you can speak to of how different industries are.

Yeah, there’s some, so let me, I’m gonna give you two answers to this. So one of them is what people are doing, and one of them is what people should be doing. So what people are doing is that tokens are very popular in DevOps because there’s a lot of different inputs and they’re moving those around sort of things. Whereas outcomes are more popular in kind of the new things like customer service and like CRM and that kind of world where there’s a human involved here. It’s not clear though that that’s what they should be doing.

So, because especially with the tokens based one, again, we would argue that you need to figure out which of the components of the token is the thing that’s driving willingness to pay. So is it the queries? Is it the size of the query? Is it going back to our RFP? Is it the number of proposals? Is it the size of the proposal? Is it the value of the proposal? All of those things might be different inputs to the token.

You really shouldn’t find out which of those things is the thing that drives willingness to pay rather than taking an amalgamation and charging as a token.

Cool, any other industries that you can call out things you’re seeing or that they should be doing types of end markets?

Yeah, FinTech is one of the few ones that can sort of get away with outcomes- based pricing. So FinTech, the reason that it can do it is because, remember when I said that the core thing about outcomes- based pricing is that you have to believe that A, the metric is objective and B, that I moved the metric. And it generally is the case just that like in FinTech, that tends to be the place where that happens most frequently, right?

So like if I more efficiently route you to the right, let’s say credit card provider or something like that, or I manage your fraud a little bit better, I don’t know, something, it just tends to be a place where it’s like a clearer outcome to metric.

It’s tied more tightly to the financial outcome itself.

Yeah, and if you think about even FinTech, like a lot of the metrics in FinTech are like percentage of transactions. That’s a beautiful metric. Tracks amazingly with wallet size. Cool.

All right, I think we are at time. For those who asked in the chat, yes, the recording will be available. If you got invited through your investor, it’ll be through their portal. And we will take a look at the feedback and potentially have more pricing discussions as AI continues to progress.

You know where to find me.

We’ll also share Ian’s information. So if you’re looking for a pricing guy, he could be it. Awesome. Thank you so much, Ian.

See you guys.