The GenAI Genius Business Model

At the risk of sounding like a broken record, the GenAI business model is out-of-this-world genius. A technology that impersonates humans, leading people to question whether the tech is sentient. A tool marketed as a superpower employee in place of a team, at the price of one individual.

But is it?

I want to share facts with you and, if you are in an influential position, guide you on how to direct conversations in AI. In the end, it boils down to a business decision. First things first, GenAI operates on tokens, and that, my friend, is everything you need to know to discuss costs and ROI.

A token is the smallest unit of data that an LLM translates into numbers. For example, a sentence-tokenizer identifies sentences from a document. Similarly, a word-tokenizer identifies every word in a document. In fact, word-frequency algorithms operate by counting word frequencies in a document, so they first identify the words present in the document.

A token is smaller than that; it can either be the whole word, like "apple", or small units of that word. For example, "computer" can be split into "comp", "uter", and the token "comp" could be related to computer, complaint, compliment etc... for the geeks like me: beautiful!

When you write to your GenAI tool "Thank you that was very helpful", you're burning 7 input-tokens.

Every prompt-response interaction involves Input and Output tokens separately. But this is where things get creative. The cost-per-token on the output is much higher than the cost-per-token on the input.

Figure 1: Generated by Claude Pro. Prompt "generate a small image to explain input vs output tokens"

But that's where the value is - the output of the GenAI!

The ratio depends on your subscription's pricing bracket, but even if you pay for a subscription, you either have a usage limit or pay for usage per seat. For example, in Claude Pro, there is a Usage Limit in the subscription - that if exceeded, more tokens could be bought at an extra charge.

**Since it resets and I don't use it for every single thing, I never ran into a usage outage - yet!

What does it mean for ROI in AI?

One thing: Intent. The more GenAI is used blindly rather than focused, the more back and forth with the tool, the more tokens your team will burn through, and the higher the costs. Some GenAI have been described as overexplainers or as outputting a bunch of jibberish and emojis. Do you wonder why? That is why it's genius! And we can't control it when we're using it in an area we know little about, or worse, to brainstorm.

What should Businesses do?

AI is a great enabler, and I recommend hiring a true Expert. As guidance, any 'expert' who uses terms like 'brainstorm with AI' is probably not a good start. Instead, hire with intent. Don't you believe me? Let me share Anthropic's guidance on Usage limit best practices, which I applaud, since they're potentially losing money by sharing it.

Firstly, drop any "brainstorm with AI" evangelism that was promoted to drive your cost-per-user ratio sky-high. Any unfocused prompting will cost you more than a human or five, apart from the salary of that person. (Talk about ROI!) Do not treat the AI as a human or a colleague; every extra word it outputs counts as a token.

I wrote "Thank you" in Claude to which it returned "You're welcome! Feel free to come back anytime — whether you want to dig deeper into AI pricing, build something with the cost calculator, or explore anything else. 😊" ~ 43 tokens. (The emoji counts as 1 token)

Instead, do what any financially mature person does with money - spend intentionally. Hire skilled individuals and empower them with AI. If you provide AI tools for individuals to do a job they're not skilled to do, the costs will not be worth the returns. Hiring data and AI experts or upskilling your employees to use AI in their areas of expertise is the secret to ROI in AI.