AI-Generated Business Plans

When OpenAI released GPT-2 back in February 2019, the tech world was enamored. The natural language generation model, composed of 1.5 billion parameters, quickly became the most talked-about tool of the year. Partly because OpenAI fed the press a salacious story that they couldn’t release the whole model because it was too dangerous for the public. Nonetheless, its ability to generate text based on provided prompts with human-like quality was magical.

I experimented with GPT-2 quite a bit back then through an integration called Talk to Transformer.

I published a story titled “How I used AI to write a business plan,” where I leaned on GPT-2 to help me build an executive summary for a fictional lead-generation business. (You can read the AI-generated Executive Summary here.)

At the time, it felt revolutionary. Not because the business plan was that great. It definitely wasn’t good enough to raise money. It felt revolutionary because it was coherent. I was mesmerized that the AI could stay focused on one subject for 400+ words without trailing off.

Looking back on it (and comparing it to GPT-3’s abilities today), the business plan sucked. AI-generated text was passable. But certainly not good enough to automate any writing tasks.

Three years later, though, it’s not absurd to ask the question...

Can AI Be Your Business Partner?

Ethan Mollick wrote an article on how to use AI to generate ideas, where he showed how ChatGPT helped him come up with revolutionary toothbrush ideas. Some of the ideas aren’t that bad and (at the minimum) helpful to spark your own ideas.

Ethan outlines a few methods for getting AI to help you think creatively:

1. Play with constraints. Prompt the AI to generate extreme use cases of a product or service.
2. Interview the AI. Ask the AI to answer a series of questions from a specific persona’s perspective.
3. Yes, and… Let the AI go off on tangents by responding to its output with a follow-up “yes, and…” statement. (Based on the well-known improv technique.)
4. Go for volume. Tell the AI to generate 10, 20, 50, or 100 solutions to your prompt. One of them is bound to spark your thinking.
5. Make it weird. Provoke the AI with strange parameters like presenting you a solution in the form of a Johnny Cash song, a script from Seinfeld, or described by an alien.

What this tells me is that I might have had the collaborative relationship wrong back in 2019. I should have given AI the role of helping me formulate ideas, not formulate a formal document.

So, I took another stab at writing an AI-Generated Business Plan. Although this time, I focused on using AI to help me generate the premise for the business and used some of the methods that Ethan outlined.

You can view my AI-Generated Business Idea here.

Overall, it was fun and much more coherent than my experiment from 2019. But I still wouldn’t trust AI to create a finished product.

More Than A Hobby

At the moment, interacting and working with generative AI is a fun hobby. To me, that’s the most emphasis I can place on it in my workflow. I haven’t tried all of the generative AI platforms out there for every niché use case imaginable. But from my tests with LEX and Jasper (for writing) and Midjourney (for images), I haven’t been able to consistently use it in lieu of my own efforts.

Although I’m calling generative AI a fun hobby today, I’m not saying it’s a distraction. It’s certainly not a waste of time because learning how to operate/collaborate with these tools will be a major competitive advantage in the very near future.

I’m still doing daily sessions with generative AI because I know how quickly it will translate from a hobby to a core part of my daily work routine.

Playing with the AI’s parameters until I get something creatively worthy takes too long and requires too much reworking. Currently, it’s quicker for me just to do the work.

But I know the trap of saying, “It’ll be quicker if I do it.” It’s a managerial pitfall. Anyone with experience managing teams knows that at some point, you have to let go and trust your team to support you; to maximize what you can do.

Figuring out what you can and cannot delegate to AI is the million-dollar question. And everyone should be running their own tests on this front.