The Large Language Models that power generative AI have been called by some “prediction machines.” That’s because they’re trained on vast amounts of data, and then produce the words that are most likely to come next based on what they’ve seen before.
To summarize the process:
- Take in existing information.
- Draw connections and identify patterns.
- Encounter something new.
- Predict the correct response based on existing patterns.
Sound familiar?
When you or I make a decision, we follow a similar process: We consider the available information, whether that’s our past experiences, our education and training, or data we’ve just been handed. Then we select the option that we anticipate will have the intended outcome we want.
But unlike us, foundation models like ChatGPT have ingested billions (if not trillions) of words — an internet’s worth of stories, case studies, business strategy, and insights — and have demonstrated impressive reasoning capabilities.
All that to say, AI can be a really powerful decision-making tool that you can use to:
- Select optimal strategies: Evaluate multiple business strategies and recommend the most effective one based on data-driven insights.
- Judge past projects or campaigns: Recommend which campaigns to replicate or scale based on past performance.
- Prioritize tasks: Use AI to prioritize daily tasks and assignments based on deadlines, importance, and available resources.
- Select key points for discussion: Identify and select the most important points for discussion in a meeting or report.
- Advise on meeting agendas: Recommend agenda items for meetings based on outstanding tasks, project status, and team feedback.
- Recommend process improvements: Suggest ways to improve workflows or processes based on analysis of current practices.
- And more.
✏️ Decision-Making Starter Prompts
What are the potential outcomes of [decision/action]? Weigh the pros and cons and recommend the best course of action.
Given the following options: [list options], which one would be most effective for achieving [general goal]? Explain your reasoning.
Based on the provided data, what is the most logical next step? Provide a recommendation with supporting arguments.
How would you prioritize these [number] objectives? Explain your decision-making process and rationale.
What are the potential risks and benefits associated with [general situation/decision]? Should we proceed, and why or why not?
Considering both short-term and long-term implications, what decision would you recommend regarding [general issue]?
Using a cost-benefit analysis, evaluate the following alternatives: [list alternatives]. Which option provides the best value?
What factors should be considered when making a decision about [general topic]? How would you weigh these factors?
In light of [changing circumstances/new information], should we adjust our current approach? If so, how?
Apply [decision-making framework of choice, e.g., SWOT, PESTEL, Six Thinking Hats] to this scenario. What course of action does this analysis suggest?
Given the following criteria: [list of criteria], rank these [number] options in order of preference in order to achieve [specific goal].
Evaluate the potential outcomes of implementing [strategy A] compared to [strategy B]. Which strategy aligns better with our long-term goals?
How would [thought leader, expert, etc.] solve this problem? How could I use this method to approach the problem I've outlined?
💡More Tips
When using AI for decision-making tasks, try these:
Provide context.
First and foremost, you’ll notice that many of the prompts above are missing one big thing: context. What is the problem you’re facing? What is the challenge you’re trying to solve? What are the factors or options you’re weighing? What limitations exist? The more information and specificity you can provide, the more likely you are to get a helpful response.
As Ethan Mollick wrote in “Co-Intelligence” (emphasis added)
...the default output of many of these models can sound very generic, since they tend to follow similar patterns common in the written documents the AI was trained on. By breaking the pattern, you can get much more useful and interesting outputs. The easiest way to do that is to provide context and constraints.
Ask it to force rank choices.
Part of an LLM’s ethos is to be agreeable (the “helpful” part of “helpful, harmless, honest”). This can pose challenges when you want a critical perspective, i.e. “These options all look good!”
To avoid this, request a ranking of the proposed options. This will force the LLM to evaluate them across various criteria and recommend the most effective one. Plus, requesting justification for its rankings will challenge the LLM to make a case for or against each, and give you more insight into the pros and cons of each approach.
Here's a recent example where I asked ChatGPT to do this. After explaining the challenge I was facing, I wrote:
Given all of this, creating a Gantt chart with dependencies is challenging.
1. Suggest a few ways I should think about and consider approaching this project to achieve my goal of creating a Gantt chart with dependencies.
2. Evaluate the various approaches you have suggested for their feasibility and simplicity. Consider the pros and cons of each approach.
3. Force rank the options you've presented based on your findings.
Apply a specific framework or person.
Have a methodology you follow or a person you admire? Ask AI to assess something using a particular approach. Or, provide an example of exactly HOW you want it to reason before presenting your request. This is called “chain-of-thought prompting” and data suggest that this can lead to better outcomes than prompts that skip this.
Encourage AI to show its work.
If you’ve experimented with LLMs, you know that responses can sometimes be verbose, and it can be tempting to ask it to cut to the chase. But this language generation is effectively how LLMs “think.” For reasoning tasks, allow it to provide a full explanation of its reasoning.
If you prefer, you can ask it to reason its way through the problem, and then provide a summary. (By the way, I asked ChatGPT to fact-check that last paragraph. Here’s what it said.)
Embrace multi-model.
Use multiple models collaboratively to build on and critique each other’s work. This might be as simple as telling Google Gemini “Here's a prompt I gave ChatGPT … Here's the response … What else would you add?” (Tip: Do this seamlessly in a single conversation thread using a tool like BoodleBox.)
Most important of all, LLMs should never replace human judgment and critical thinking. To quote our company’s AI policy, “AI assists and augments, but does not replace, human ability.”
AI can be a valuable tool to correct your own blindspots — so long as you keep its shortcomings in mind. I recently heard these foundation models described as “persuasive not passive.” In other words, in their pursuit to satisfy us, they are inclined to convince and encourage and agree.
It’s why AI is a valuable decision-making assistant, but should always approached with a healthy dose of our own critical thinking.
Want more? Subscribe for more growth, fewer growing pains.
Ever wonder how other growing companies make the tough decisions — like how much to spend on marketing, when it’s time to expand your org chart, or when and where to use AI?
Us, too. So we decided to create growthcurve: A newsletter dedicated to practical advice for growth without the growing pains.
Get frameworks, templates, and advice to smooth out the bumps along the way.
