I'm really curious how much effort there was in building up the data set - before, training the model before you got to "music"
Reading the steps feels like 9 months to a year before you got to credible music.
What kept you going in the belief this would work. I can think of 20 reasons why this shouldn't work - hence its "surprising" that it does. Its quite easily something you could have worked on for 5 years with no results.
reading your background - it also sounds like your time would be tightly constrained hence figuring out where to deploy it - you need to have some conviction you'll have success
- Loads of account details available - US, AUS, USD, EUR
- Super cheap all of the time (think revolut is cheaper for some amounts and routes)
- Plastic card
Hey Andy — love this. But really you are developing a framework for figuring out how much time to invest in research.
A couple of questions — which was prompted by your blog
What situations is there no value in research ?
Corrollary — are there ways you can systemically reduce your risk by increasing your gut feel for probabilities without reesearch ?
Final one — you make the great point on unknown unknowns — apart from systemically working through all possible states and thinking through various envrionmental changes — are there any systemic approaches to reducing this risk.
In working with startups — I would summarise an alternative approach that covers all 3 by the following pieces of accepted wisdom from startups.
Launch early — to validate in the market with real people
Talk to customers to build an empathy or gut feel for relative priorities of stuff — and what really matters
I think these two principles go some way to solving the problems you are highlighting through a different approach.
I’m sharing — as I’m sure you have considered this appraoch — and have a reason for discarding it.
Hi Nilan - thanks for your words. It’s true, that as someone who is ultimately employed as a researcher in some role, I do wonder (a) when research is valuable and (b) whether the people controlling the decisions of when/how much research to undertake can justify this decision quantitatively. (But rather than think of “research” I’d prefer to back that out into a more fundamental concepts like “information”.)
I think your question 1 is partially answered by the inequality — it provides a upper bound for how much utility you could gain from the research. More practically, perhaps a decision maker could use gut feeling to break it down into quartiles — outcomes which are worst case (0th quartile), below average (1st quartile), average (2nd quartile), above average (3rd quartile) and best case (4th quartile). If the estimated cost of research is much less than ROI(3rd quartile) minus ROI(1st quartile), then the research seems worthwhile even for “typical” outcomes. If the estimated cost of research is greater than ROI(4th quartile) minus ROI(0th quartile) then the research is obviously not worthwhile (that’s the inequality).
I think your questions 2 and 3 are the questions I would also like to see answers to :) I think that given decision makers come from a variety of (often non-mathematical) backgrounds, tools should be simple and easy to use/remember, which is why I’m focusing on using things like “worst case scenario” and “above average scenario” and so on. That way we get to combine quantitative analysis and gut feeling to get good outcomes from typical decision makers.
And yes — I completely agree that the agile “release often, get feedback early” and the startup “move fast and break things” philosophies are designed for revealing information quickly so that you can make good decisions as early as possible (thus also maximizing the utility of that information). As a general work pattern, I put this under a kind-of “don’t be stupid” mentality — even if this maybe wasn’t obvious 25 years ago. I think the topic of the article is to be able to address specific, large decisions (like exercise 2).
The borderless bank account seems available in a ton of countires too - https://transferwise.com/gb/borderless/