7 steps to price AI products

A frequent question these days is about pricing AI products. I’ve worked with a few companies on this, and I have a feeling other people have this question too. So here’s my take.

It helps that I’m in the thick of this with skyp.ai, so it’s a practice-what-you-preach moment. Or, “build in public”, in the parlance of our times.

Founders always underprice

Founders are both overconfident and insecure. They're sure they can build something crazy that people want—but terrified people won’t. So they often price too low, hoping that’ll drive usage. Signups. Closed-won. Many run out of money simply because they never asked anyone to pay—or pay enough.

Big companies don’t have this problem. IBM and Salesforce salespeople will take you to the cleaners. I talked to a nonprofit that was paying Salesforce $50k a year for a “mobile app”. Never mind that Salesforce doesn’t actually make a mobile app, and any (free) SQL database could have done what Salesforce was charging them $50k to do. It was despicable. But someone got a bonus, and the customer was actually happy about it.

Another founder I spoke with charged one large enterprise $250k for a proof of concept; six months later, desperate to raise more cash, he charged a different one ~$2.5 million. For the same pilot. Turns out to Fortune 100, a zero makes zero difference. That remains the same, AI or otherwise.

The difference with AI

There are two core differences with AI today. First, everyone wants it. Whether it’s an “AI strategy” or simple FOMO, there is demand for AI solutions. Period. This is one of those “no accounting for taste” things–it just is what it is.

The second core difference–which is, truly, different than regular software–is that AI costs money to deliver. Real money.

A founder friend was messing around building an AI to write unit tests. (Unit tests make sure your app or code does not break when you add or change something.) His company had <50% coverage. His tinkering increased that to something like 60%. But “it was expensive”.

“Expensive compared to hiring a human?” I asked.

“Hell no. It was like $40.”

There’s not a lot of software that costs $40 in hard cash to run off and on for a week. But AI is that software. It did replace a $8k/mo human, however, so there’s that to consider.

The fact that it costs money AND people are willing to pay for it makes AI very different than prior software waves. Instagram got pretty far on $1mm in seed funding; YouTube was burning through cash with hosting videos–but only ever raised $11.5 million. A lot at the time, but most of it was an $8 million Series B just before it was acquired for $1.65 billion. YouTube was able to crack the Top 10 internet sites globally–on its first $3.5 million. A free AI service could never hope to do that.

Which is why all the AIs worth using cost money.

How to price AI software

Here’s a seven step process for pricing AI software. I’ll get into some details about tactics at the end. Note, this assumes you are very early in your company’s journey. If you have a lot of data already, including usage and payment data, you can do a lot to price tactically you cannot do at the beginning. While these steps can be applied at any stage, if you have data–use it.

1. Discover the value

Understand the value your product creates in the eyes of your customers. What you think it creates is 100% irrelevant. Don't even say it out loud–ask questions only. This was a mistake I made and many founders make: just because, logically, your product is valuable to your prospect does not mean they think it is.

Value is not willingness to pay–another rookie mistake. For example, unit tests might not be worth paying $8k/mo for a human to write–which is why they haven’t been written. Value is the value of having them, of your software not breaking when you launch a new feature or update. Maybe it’s closer to $40. Who knows? Your customer knows, actually, so have this discussion with them.

2. Discover willingness to pay