Love this framework, Anna. One thing I’ve found: spikes without decision architecture become throwaway research. When I come back 6 months later, I would have to start over because I’d documented findings, not the decision scaffold. I therefore anchor every spike to “what needs to be true to act”, so exploratory spikes build conviction instead of stopping at intelligence gathering. Drafting a piece on this approach now…
Ooo that’s such a good point. I’ll have to tweak my prompt to include a trace of the decision points too. Please share your article once it’s live. I’d love to read it.
It’s very organic. Like when DBT was acquired by Fivetran and I saw that in the news I researched what other options might be in case we need to pursue them. Or yesterday one of my PMs raised the problem of how we extract information from free text fields and turn it into structured parsed data (a common challenge with health record data). So it’s usually a reaction to a headline, a conversation, a problem I’m thinking through, etc.
The harder part is once you’ve got a backlog of ideas figuring out which is worth days or weeks of discovery time and engineering capacity.
Love this framework, Anna. One thing I’ve found: spikes without decision architecture become throwaway research. When I come back 6 months later, I would have to start over because I’d documented findings, not the decision scaffold. I therefore anchor every spike to “what needs to be true to act”, so exploratory spikes build conviction instead of stopping at intelligence gathering. Drafting a piece on this approach now…
Ooo that’s such a good point. I’ll have to tweak my prompt to include a trace of the decision points too. Please share your article once it’s live. I’d love to read it.
Hi Anna! It’s live now: https://open.substack.com/pub/shuba/p/stop-researching-start-deciding?r=awq0n&utm_medium=ios
Thank you! Off to read it after school drop off!
Spot on. The 'plane while flying' is so real. How do you pick your exploratory AI spikes?
It’s very organic. Like when DBT was acquired by Fivetran and I saw that in the news I researched what other options might be in case we need to pursue them. Or yesterday one of my PMs raised the problem of how we extract information from free text fields and turn it into structured parsed data (a common challenge with health record data). So it’s usually a reaction to a headline, a conversation, a problem I’m thinking through, etc.
The harder part is once you’ve got a backlog of ideas figuring out which is worth days or weeks of discovery time and engineering capacity.
Yes please keep cross-posting to LinkedIn!