How AI Can Help Us Make Decisions Faster

You should make decisions faster.

When was the last time you were involved in a decision that dragged on for far too long?

Maybe your team was debating what color to make the “Sign Up” button on a landing page. Or which feature to build this sprint and which one to defer for two weeks. Or you and your partner were just trying to decide where to eat for dinner.
‍
Recently, my team spent close to a whole day going back and forth between two rental houses for our upcoming team offsite. The houses had the same number of beds + baths, they were <1 mile apart and the total cost difference was ~$300. The big difference, the tough decision to be made, was whether to optimize for location or comfort, which would have a big impact on the overall experience of the offsite. I spent over an hour of my most precious resource (time) considering and communicating about the trade-offs of these two options. Was that time well spent? Especially these days when AI is available to help?

Getting started with AI for decision making

I decided to experiment with using OpenAI’s GPT-4 to help classify decisions that can be made quickly. I came up with the following list of 7 decisions that I believe are spread across the speed vs. accuracy spectrum.

What shirt should I wear today?
Should I subscribe to a $20/month paid newsletter about the latest research and development in the field of Artificial Intelligence?
What color should we make the “sign up” button on our landing page?
Which location should we choose for next month’s quarterly offsite?
Should we make an offer to Jane Smith for our open data science role?
Should we raise an A round of investment capital in a sluggish macroeconomic environment?
Should we sell our company?

The first question you should ask yourself when considering whether you can make a faster decision is what degree of lasting impact the decision will have.

Suzy Welch coined the “10-10-10 rule” which is a powerful mental shortcut for assessing the impact of a decision over time. The 10-10-10 rule asks you to consider how you will feel about the potential outcomes of a decision 10 minutes, 10 months, and 10 years in the future.

For example, what you decide to wear today will have almost no bearing on how happy you are 10 months or 10 years from now. However, deciding whether you should sell your startup or raise another round of funding is likely to have an impact on your satisfaction 10 years from now.

Let’s explore if a large language model can help us understand the impact of a decision over time. I asked ChatGPT, on a scale of 1 to 10, “What level of impact does this decision have over time?” for my seven decisions. Here’s what I got back:

Shirt to wear: 1
Subscribe to newsletter: 4
Color of sign up button: 3
Location for offsite: 5
Make the job offer: 7
Raise capital: 9
Sell our company: 10

I found these results to be mostly in line with how I personally would assess these decisions. This is a promising start for creating an AI-powered rubric for identifying fast decisions.

Given these results, I’m going to classify anything that rates as a 4 or lower as a decision that should be made “quickly.” I’ve been in situations where a decision about what color to make a signup button, or whether to make a small recurring purchase have soaked up hours and hours of valuable time so this is useful.

AI can help analyze impact

Another way to identify decisions that can be made faster is to consider if a decision is easily reversible. Amazon popularized the terminology of “one way door” vs. “two way door” as a way to think about how easily and affordably a decision can be reversed. If a decision can be easily undone, then it’s a decision that should be made fast. It’s likely more efficient to simply unwind a “two way door” decision if the outcome isn’t desired rather than to try and get to 90% accuracy before making it.

The “signup button color” decision is a good example of this. If you have a very high-traffic site, then making the button light gray might actually have a material impact on your signups and revenue. However, if you notice a drop in signups, it’s a decision that’s extremely cheap and fast to reverse.

Let’s see if GPT-4 can help us identify easily reversible decisions using the following prompt, “On a scale of 1 to 10, how costly and difficult is it to reverse or undo the effects of each decision?” Here are the results:

Shirt to wear: 1
Subscribe to newsletter: 2
Color of sign up button: 3
Location for offsite: 4
Make the job offer: 6
Raise capital: 8
Sell our company: 10

These ratings are closely aligned with my own assessment, and I think I can classify decisions with a “Reversibility Rating” of 4 or less as “fast decisions”. While I still might want to speed up our decision-making on those with a rating of 5 or higher, those with a rating of 4 or lower are prime to be flagged as decisions that should move fast.

I asked ChatGPT to give reasoning behind these responses and GPT-4’s answer about “Which location should we choose for next month’s quarterly offsite?” decision gives us a clue about another way to identify fast decisions.

“While the choice of location for an offsite meeting cannot be undone once the event has taken place, the consequences of a less-than-ideal choice can be mitigated in future events. The impact is limited to the outcomes of that specific meeting.”

AI can help with reasoning

You can identify decisions that should be made faster by looking for decisions that are repeated frequently. For example, if you are planning a meal for your grandma’s 75th birthday, you probably want to be thoughtful since this happens once a lifetime. On the other hand, the decision of where to eat out this weekend is a decision you’ll get to make at least 51 more times this year.

Decision repetition allows you to make a decision faster for two reasons:

1) It reduces the psychological burden of potential regret. If you only get one shot at something then a looming sense of potential regret makes it hard to make a fast decision. However, if it’s a decision you get to make 100 more times, then you should waste no time making that decision because you can make a different choice next time.

2) When a decision is repeated many times, you can apply your learnings from that decision to the next time you make it.

Let’s see if GPT-4 can help us identify repeated decisions with the prompt, “on a scale of 1 to 10, how infrequently is this decision likely to be repeated?”

Shirt to wear: 1
Subscribe to newsletter: 4
Color of sign up button: 5
Location for offsite: 6
Make the job offer: 7
Raise capital: 9
Sell our company: 10

While there’s nothing surprising here and the results are closely correlated with the previous results, I believe it’s valuable to look at decisions through this lens because of the potential to help bolster our confidence in moving fast on a decision.

Now that I’ve got a three dimensional “fast decision” detector, let’s see what it thinks about our team’s offsite housing decision that we spent a day deliberating last week.

I asked ChatGPT to use the three previously established ratings systems and evaluate the decision speed of which house to rent at the team offsite in Minneapolis.

*robot noises*

Conclusion -> Fast Decision

Try out AI for your upcoming decisions

The beauty of LLM models is that they do an astounding job of taking an extremely wide lens on problems you are trying to solve. They allow us to systematically inject best practices in areas that were previously very hard and expensive to apply software solutions to. Imagine a world where the tools you use to collaborate every day can help you take note of decisions that are moving too slow.

What’s a decision that AI could help you make faster?

Here’s a prompt to get you started:

You are a "fast decision detector". You will help identify decisions that should be made quickly. Fast decisions will be identified using 3 criteria.

1) You will rate the decision on a scale of 1 to 10 for how much long-term "impact" the decision will have.

2) You will rate the decision on a scale of 1 to 10 for how difficult or expensive it is to reverse the decision.

3) You will rate the decision on a scale of 1 to 10 for how infrequently repeated the decision is.

If the decision is rated as a 4 or lower on any of these criteria it should be classified as a "fast decision". Otherwise, it will be classified as a decision that "needs consideration".