Need to Write a Generative AI Business Case for Your Boss or Client? Here's the Data You Need.

How much productivity boost could you expect to get using generative AI over the next 12 months, and who benefits the most, the highest or lowest performers?

Writing a GenAI business case. Imagined using DALL-E 3

Do you use generative AI (like ChatGPT) at work?

Or maybe you use it in your personal life?

Or perhaps both.

I know I do.

The fact is, over 100 million people signed up to use ChatGPT by spring 2023, and the number of users, after the rate of growth allegedly flatlined over the summer (school’s out and all), is still increasing according to OpenAI CEO, Sam Altman.

So what are people doing with these Large Language Models (LLMs) since the novelty of writing a “… rap in the style of Chaucer about the dangers of climate change” (try it) has since worn off?

And more importantly, what can we expect in terms of ‘gainz’?

Particularly, “productivity gainz”.

(Yeah ‘gainz’, not a spelling mistake ;).

Let’s take a look at what the literature and research say.

If you need to make a business case or convince your boss, or clients to invest in generative AI, then this is the post for you.

Let’s dive in.

---

Let’s have some fun—a guessing game.

And yes, I do have the answers so won’t leave you hanging:

Question 1.

How many unique occupations (jobs) do you think there are in a well-developed economy like the USA?

Question 2.

How many unique tasks do you think make up each of those unique occupations?

Take a guess.

Go on. And don’t peek!

The good news is you don’t need to guess because in the U.S. there’s the O*NET® database.

In their own words,

“The O*NET database contains a rich set of variables that describe work and worker characteristics, including skill requirements.”

—Source: O*NET® database

The O*NET® database lists some 1016 unique occupations in the US economy that have been documented to date.

And of those 1016 unique occupations, there are 18,008 unique tasks (according to the ChatGPT Advanced Data Analysis plugin I used to analyse the data) that make up those occupations, although there are 19,281 individual rows in the task spreadsheet due to task duplication across multiple occupations.

Here’s a small sample of some of the occupations documented,

The first 10 occupations from the O*NET Database. I analysed the data using the OpenAI Advanced Data Analysis plugin. Source: O*NET® database

How did you do?

Ballpark correct, or way out?

Why is this data important?

Well, it turns out you can use it to analyse the impact of LLMs on the job market.

In last week’s newsletter on AGI, I decoded the obscurely titled “GPTs are GPTs” paper, meaning “Generative Pretrained Transformers are General Purpose Technologies”.

If you haven’t already read it, it’s worth a look as it explains how AI is likely going to become a general purpose technology of which only 25 have been identified by academics and economists throughout the whole of human history to date—examples being, the wheel, printing press, electricity, and more recently, the Internet.

The upshot is that, by analysing the data in the O*NET® database and integrating both human expertise and using GPT-4 to help classify occupations and tasks, the researchers predicted that,

“… around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted.”

—Source: GPTs are GPTs

Taking the figures of, 80% of 10%, and 19% of 50%, and assuming the remaining 1%, has 0% automation, it means that, on average,

17.5% of the tasks for a worker in the U.S. could be affected by the introduction of LLMs.

How many organisations are using LLMs today?

Next, let’s try to understand what percentage of organisations are using LLMs (like ChatGPT, and others) today to get work done.