First time it happened to me when I was building one pet product for fun. The idea was simple, get data from device, crunch some numbers and Voila! product is ready. But, apparently it is not real world case, data appears to be bad quality and need to be cleaned before processing.

So I decided to use algorithms and started looking into space of target language, iOS swift, — no implementations found, then looked at Ruby situation is much better but still decided to think. And then I realised that R has full set of algorithms to play with. So I went to R and failed. Failed because I didn’t know my data well enough to understand what should I do with it and what algorithms to apply.

So I ended with excel spreadsheet and did all the job manually: got rid of garbage, unified and pre-processed data. In excel. I ended with description of algorithm how to clean data. And this is all I need.

Next step, was implement this algorithm. I did it in Ruby, because it is faster to develop and debug. Tweaked it till I get expected results on sample set. And only after prototype succeed I sat down and coded algorithm in swift and finally launched application on iOS device. This story is open as I haven’t launched app yet. But lesson learned, know your data, and do prototype in the fastest way.

The second case was when I decided analyze my personal finances. Shame on me, I don’t accurately track them in one of those apps. So I manually exported transactions, categorised and started playing with structured data. Comparing categories, sorting, making charts. And all this in order to find an answer got to optimise my family spendings.

Only here I realised how much faster I can iterate in excel rather than implementing features only to find out I need more iteration to understand feature is not useful. Obviously iterate in excel is faster than iterate by code. So this is the right way, build prototype manually and then automate if it succeeded.

Some time ago I was a little bit involved in similar web product which put on hold current;y. So there was a struggle with what to do with collected data and how to help customer with processing. What features user needs and so on. How to visualize it properly. How to build habit of user to optimise spednings etc.

Now I can say that most efficient mvp should be done manually first, got proven and only then automated. And excel is a good tool.