Do Reco For Realz

[strubled]

Context

https://github.com/strubled/optimizr/issues/7#issuecomment-631170776

Game Plan

Store Form Details and Card Values

• Create a table called transaction_category_totals
  • fields: user, grocery, gas, hotel, dining, airfare, amazon, other, total
  • things to note: we don't need to create new records each time...but for now we will. maybe it'll be useful for the user to see what their transaction categories were a year ago
• Update the controller to create new records upon submit, and store the variables in the controller so that view can see them (replace storing them in session)
• Create a table called card_value
  • fields: user, grocery, gas, hotel, dining, airfare, amazon, other, total
  • things to note: same as above...it could change over time based on card worth as well
• Update reco view to use results from card_value

Create Reco

• Create queries in the controller to sort the card_value categories, and return the card name for the card w/ the highest value.

gas_winner = ...

• We need to then find which name or names are appearing. not sure how to do this right now, but i'm finding promising things.
• we need to calculate a total value for each card - which should be easy once we know which cards won which thing, but we have to not over charge the annual fee

let's consider the following possible results:

I assume we want to show the user the total amount of rewards they'll earn with each card that we recommend. in order to create the below table, we compared the results of each category - the annual fee. let's say for example the annual fee for hilton was $100. in this case, the hotel rewards were actually worth 200, airline 150 and total 500. so really we're over counting the annual fee by 200. we probably need to add even more logic that says if a card wins more than 1 category, let's say 3 categories, we have to find the annual fee and add back in 3-1 * annual fee - in this case the $200 back.

grocery_winner = capital one gas_winner = chase hotel_winner = hilton 100 airline_winner = hilton 50 dining_winner = bofa amazon_winner = amazon other_winner = hsbc total_winner = hilton 400

this is an extreme example...but for hilton, we should show they can earn $750. $100 + $50 + $400 + $200 (the annual fee we double counted).

[strubled]

i've figured out how to store the transaction values and determine the winner per category. the problem is the annual fee. i just tried to do something where i found the top winners, and counted how many times they won (e.g. say if capital one won 3 categories), then found their annual fee and divided by the number of categories it won, then re-evaluated if that new value per category was enough to win (e.g. originally their grocery value may have been $200, but after subtracting a $100 annual fee, the total value is $100, so now they likely lose to another card).

the problem is that ends up making everything change, and i'm not sure you can write enough if statements to figure it all out. example, if capital one no longer wins 3, and only wins 2, and now i move onto another category like gas and find they don't win that either, now they're down to 1 win, which if it has a high annual fee probably makes it lose that category too, which makes some other card win, but now we need to check the annual fee on the new winning card and after subtracting it's value probably ends up losing to some other card...just doesn't seem like this path works.

i'm thinking we have to factor in the annual fee BEFORE evaluating against the other cards. we could do this in maybe a couple different ways:

1. by % of spend (I think @sknechtel you suggested this on text)...we could look at what a user's categories of spend are, and if say 50% is in the grocery spend category, we'd assume that groceries would account for 50% of the annual fee. so if it's a 2% cashback on 10000 of spend, w/ a $100 annual fee, we'd say you earn $200 - $50 = $150
2. by # of categories we're using...we could just divide the annual fee by 5 since we're looking at 5 categories (grocery, gas, hotel, dining, airfare, other...we're not doing amazon yet). that would mean in the same example above the annual fee would be $20 per category, so $180
3. by # of categories the card supports...say a travel card doesn't pay out for groceries, so only 4 categories. it would be $25 annual fee, so $175
4. we could gloss over this whole thing and just roll w/o an annual fee in the calculation except for the "total" value calculation. this would end up making the reco tell someone to use a particular card for a particular category when if they did that, and only that category, we're likely going to be wrong.

another idea is to roll with 1 - 3 above and instead of recommending 1 card per category, we could recommend a few cards that end up with values roughly the same and add an asterisk or something. that seems lame.

[strubled]

one other idea is...

5. by number of categories the customer entered. if they only enter in grocery and gas spend, we'd divide the annual fee by 2.

[sknechtel]

I wracked my brain about this last night. I’m going to talk to Juan and see if he has any ideas. There has to be a way to do this.

Sent from my iPhone

On May 21, 2020, at 9:59 PM, Dan Struble notifications@github.com wrote:

 i've figured out how to store the transaction values and determine the winner per category. the problem is the annual fee. i just tried to do something where i found the top winners, and counted how many times they won (e.g. say if capital one won 3 categories), then found their annual fee and divided by the number of categories it won, then re-evaluated if that new value per category was enough to win (e.g. originally their grocery value may have been $200, but after subtracting a $100 annual fee, the total value is $100, so now they likely lose to another card).

the problem is that ends up making everything change, and i'm not sure you can write enough if statements to figure it all out. example, if capital one no longer wins 3, and only wins 2, and now i move onto another category like gas and find they don't win that either, now they're down to 1 win, which if it has a high annual fee probably makes it lose that category too, which makes some other card win, but now we need to check the annual fee on the new winning card and after subtracting it's value probably ends up losing to some other card...just doesn't seem like this path works.

i'm thinking we have to factor in the annual fee BEFORE evaluating against the other cards. we could do this in maybe a couple different ways:

by % of spend (I think @sknechtel you suggested this on text)...we could look at what a user's categories of spend are, and if say 50% is in the grocery spend category, we'd assume that groceries would account for 50% of the annual fee. so if it's a 2% cashback on 10000 of spend, w/ a $100 annual fee, we'd say you earn $200 - $50 = $150 by # of categories we're using...we could just divide the annual fee by 5 since we're looking at 5 categories (grocery, gas, hotel, dining, airfare, other...we're not doing amazon yet). that would mean in the same example above the annual fee would be $20 per category, so $180 by # of categories the card supports...say a travel card doesn't pay out for groceries, so only 4 categories. it would be $25 annual fee, so $175 we could gloss over this whole thing and just roll w/o an annual fee in the calculation except for the "total" value calculation. this would end up making the reco tell someone to use a particular card for a particular category when if they did that, and only that category, we're likely going to be wrong. another idea is to roll with 1 - 3 above and instead of recommending 1 card per category, we could recommend a few cards that end up with values roughly the same and add an asterisk or something. that seems lame.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[strubled]

I did idea # 5 above...we can always change it but i think it's fine for now. this ticket is done!