Eigenquestions: What should you do first?

I’ve been playing around recently with Shirshir Mehotra’s concept of eigenquestions. His idea in short is to attack difficult choices by figuring out which of the key questions unlocks the answers to the rest. Usually there is one domino that knocks the others down, if you can find it.

Fantasy example: if I’m preparing to host some friends for a dinner party I need to figure out several questions:

  • Main

  • Sides

  • Dessert

  • Wine? What kind?

  • Start time

Usually for me, the entree dictates the rest. But if we abstract this one layer further, the eigenquestion is: who’s coming?

I’ve started using this idea as a tool for figuring out the MVP for a new feature/product. 

Any new endeavor is basically a process of making a series of choices with ambiguous information from “who is the target user” all the way down to “should this be a button”. Most of these can and must be answered with logic, intuition, or guesses. Some are isolated requirements that illuminate nothing else. But there are usually a couple of big, important decisions about order and size that determines whether the project succeeds or has a long and damaging failure. 

For those decisions I am usually weighing two approaches with similar apparent viability. I have 3 choices: X, Y, X and Y. 

Doing X and Y has a certain safe allure. Surely one of our two ideas will work and our research suggests we probably need both eventually. But the cost of X and Y is a significant increase in the length of the project. It takes longer before we have anything, and before we have an answer. 

So, X or Y? I’ve found that identifying which will determine if we should do the other usually clears things up. 

Another way of attacking this is asking if more time/data will dramatically improve our decision on one of the pieces. This can help clarify the order when we realize doing X will illuminate the technical feasibility of Y. It can also reveal flimsy ideas masquerading as strong ones. If doing Y won’t tell us anything about the viability of the product then it certainly shouldn’t be part of the MVP and maybe wasn’t such a good idea after all.

Some fun examples to consider:

  • Pinterest thinks a new dynamic ad format will increase appeal to e-commerce advertisers. Do they start with the ad format or the advanced metric breakdowns that will show its value?  

  • If Substack wants to test bundles, should they start with writer or subscriber-side features? 

  • My local butcher wants to start selling meals too. Should they do sandwiches or bowls?

  • When Stripe was building Stripe Capital how deep did they go in modeling the loan viability before figuring out their repayment approach?

This can be a particularly helpful tool for reducing the risk of investments with high uncertainty, like those that rely on the assumption that data science will be able to build an accurate, precise machine learning model. If there’s a way to test if the idea will work for users before starting the modeling, you can save literal months and millions.