Thank you for this inspiring post! Curious about how you see generative models to help with insights creation (and how you define insights?)
Maybe you have some system like this in mind?
* Input = transactions data of the user
* Output = some text giving the user some insights into their personal finance, maybe some advice, e.g. "Bill X is due in 3 days, remember to transfer $Y from Savings!"
Or are you thinking about it more in terms of developer / data scientist tooling?
Both! I think in the first case that's a more generic form of meta-learning but it'll probably start with the latter.
I actually decided to test this with GPT very quickly by generating some fake data (45 observations) and prompting GPT3 to tell me some insights about it, the results were quite interesting!
It did primitive stuff very well! It's kind of expected but I think there's a lot of opportunity there.
One use case that I’m excited to see is using chatGPT for chargebacks/disputes. There is a known process for filing chargebacks and winning them is all about saying the right things, having the necessary proof, and following the right process. Chargeback win rates are notoriously low in the industry and friendly fraud is extremely high and prevalent. Further, people that commit friendly fraud are way more likely to do it again. Early stage fintechs can start beefing up their disputes management process with fewer team members by training GPT on their processes. Won’t be long before a startup emerges to automate this for fintechs.
100% and I imagine DoNotPay would be a good one to tackle that. I'd say though that that kind of feels like it's not technically Fintech (but awesome) for the consumer side. Fintechs actually handling the disputes could definitely be enabled by GPT! I imagine that'd actually be really useful!
Thank you for this inspiring post! Curious about how you see generative models to help with insights creation (and how you define insights?)
Maybe you have some system like this in mind?
* Input = transactions data of the user
* Output = some text giving the user some insights into their personal finance, maybe some advice, e.g. "Bill X is due in 3 days, remember to transfer $Y from Savings!"
Or are you thinking about it more in terms of developer / data scientist tooling?
Both! I think in the first case that's a more generic form of meta-learning but it'll probably start with the latter.
I actually decided to test this with GPT very quickly by generating some fake data (45 observations) and prompting GPT3 to tell me some insights about it, the results were quite interesting!
It did primitive stuff very well! It's kind of expected but I think there's a lot of opportunity there.
One use case that I’m excited to see is using chatGPT for chargebacks/disputes. There is a known process for filing chargebacks and winning them is all about saying the right things, having the necessary proof, and following the right process. Chargeback win rates are notoriously low in the industry and friendly fraud is extremely high and prevalent. Further, people that commit friendly fraud are way more likely to do it again. Early stage fintechs can start beefing up their disputes management process with fewer team members by training GPT on their processes. Won’t be long before a startup emerges to automate this for fintechs.
100% and I imagine DoNotPay would be a good one to tackle that. I'd say though that that kind of feels like it's not technically Fintech (but awesome) for the consumer side. Fintechs actually handling the disputes could definitely be enabled by GPT! I imagine that'd actually be really useful!