Data Mining: Tez aka Google Pay Rewards

Tez now renamed Google Pay has been quite a successful application having clocked over 30 million transactions within 37 days of its initiation. I was though, late to the party having it installed it only recently in July 2018.

Besides the cool UI and a very sleek experience, the most interesting aspect of the application is the rewards program where you get a shot at a certain amount of cashback on making a transaction.

This rewards program has been of a great intrigue given that it’s largely random and not many understand the nuances behind how the cashback is computed for each transaction.

I have attached the sheet of all my transactions :

  • Transactions with non-zero rewards are coloured green.
  • The sole transaction in red is the one where the offer was deterministic, available on spending a specific amount on FreshMenu application.

Screen Shot 2018-09-12 at 10.03.40 PM

In an attempt to understand the exact mechanics, I made some custom columns for the data :

  • Date Since Last: Days since the last transaction on Google Pay.
  • Transaction Value: The amount transacted through Google Pay.
  • CumTV: The Cumulative transactional value exchanged until then.
  • Reward: The corresponding cashback received.
  • CumR: Cumulative rewards earned till then.
  • Item: Whether it is a transfer to another Google Pay user or payment to some kind of service.
  • Special Offer: This column is just to identify the specific row when we got a deterministic reward.

Hypothesis #1

The reward is a function of days since the last transaction and not the value of the transaction. This makes sense if the number of active users in terms of daily, monthly is a more important metric than the notional value of transactions made.

So, if you want to know if you are up for a reward next, check if the last time since, you made a transaction is been a while. Transacting on consecutive days is likely to result in zero rewards.

Screen Shot 2018-09-12 at 10.12.48 PM

The red row was removed to maintain consistency in the type of data. Indeed, the trend is fairly apparent.

Hypothesis #2

Your cumulative reward(CumR) is a strong function of cumulative transaction value (CumTV).

Screen Shot 2018-09-12 at 10.43.25 PM

This trend is a lot stronger, it will be interesting to know if the trend slope is the same for every user or is adaptive as usage matures.

How has other’s experience been, one strong caveat has been the fact that I’m still a new user and thereby my current rewards pattern might not be the same as that of a more mature user?

Take Away

  • Time as a component solves lots of issues for them.
  • If you transact frequently, no need to incentivize you as they have already converted you.
  • If you send it to a new person who has not transacted a lot, your likelihood of reward increases given that, you a new connection for Google Pay and also engaging someone new into the system.
  • I further strengthened the anecdotal evidence by sending some fresh transactions recently which have resulted in no rewards. Thereby, it seems like hypothesis one rather than two.

Another real-life data case study: Photofeeler

Another real-life data case study: Chegg Tutors

Machine Learning Notes

Advertisements