Intelligent Customer Support for the Masses: How Nubank does it

Transcript

1. INTELLIGENT CUSTOMER SUPPORT FOR THE MASSES HOW NUBANK DOES IT FUAD SAUD & PAULO CASTRO
2. THE PROMINENT BITS • Fintech • Credit Card • Rewards • NuConta (Payments Account) ◦ More than R$9 billion in transactions • More than 4 million customers THE PROMINENT BITS
3. WE NEED TO HAVE A STRONG CUSTOMER SUPPORT IN ORDER TO MARKET THESE PRODUCTS.
4. WE NEED CUSTOMERS TO TRUST US.
5. WE TRY HARD TO HIDE THE COMPLEXITY OF FINANCIAL PRODUCTS BEHIND A GREAT UX.
6. BUT IT’S NOT ALWAYS ENOUGH.
7. BUT IT’S NOT ALWAYS ENOUGH. PEOPLE NEED HELP WITH THE HAIRY PROBLEMS THAT WILL SHOW UP.
8. COMPLEX DOMAIN • Failure in third-party financial systems • Customers need to negotiate their debt. • People get scammed/frauded
9. ENTER THE CUSTOMER SUPPORT DESIGN SQUAD
10. WE OFFER HELP RIGHT IN THE APP
11. WE STRIVE TO BE EFFECTIVE: QUICK AND CLEAR
12. SOMETIMES SELF-HELP IS NOT ENOUGH
13. AND WE’LL BE THERE WHEN YOU REALLY NEED TO TALK TO TO US.
15. CUSTOMER SUPPORT FIGURES • Xpeers represent 60% of Nubank’s headcount (24/7 team) • 13k emails/day • 20k chats/day • 7k phone calls/day
16. HOW DO WE ORGANIZE ALL OF THIS?
17. HOW DO WE ORGANIZE ALL OF THIS? DIVIDE TO CONQUER!
18. SQUADS MODEL OPS TEAMS • Bills • Customer Acquisition • Chargebacks • Collections • Fraud • Limits • Logistics • Payments • NuConta • Rewards
19. • Card delivery delay • Address change • Delivery problems (address not found, etc) LOGISTICS
20. • Questions about investment returns • Transfer in/out issues • Boleto payment delays NUCONTA
22. “GIVE ME THE NEXT TASK I CAN WORK ON!”
24. BUT NOW WE NEED TO TELL ONE PROBLEM FROM THE OTHER.
25. • Human made • Let the users correct it when necessary • Good to prototype the routing system • Capture the performance • Use as benchmark RULE BASED
26. A CLIENT THAT HAS A CARD IN ROUTE SHOULD GO DIRECTLY TO THE LOGISTICS QUEUE
27. A CLIENT THAT ONLY HAS A PAYMENTS ACCOUNT SHOULD GO DIRECTLY TO THE NUCONTA QUEUE
28. • Learns to route • Improves with time • Still uses rules when necessary MACHINE LEARNING • Human made • Let the users correct it when necessary • Good to prototype the routing system • Capture the performance • Use as benchmark RULE BASED
33. ROUTED CORRECTLY 75% CHATS 85% EMAILS
34. BUT WHAT’S THE WORST THING ABOUT BEING IN A LINE?
35. BUT WHAT’S THE WORST THING ABOUT BEING IN A LINE? THE WAIT!
36. SIMPLE QUESTIONS DESERVE SIMPLE ANSWERS.
37. XPEERS ALREADY USE PRE-BAKED RESPONSES TO ANSWER SIMPLE QUESTIONS. WE CALL THEM MACROS.
38. “You can ask for a credit limit increase using the app, it’s easy: on the home screen, look for the 'Limit Settings’ shortcut. Then tap 'Request new maximum credit limit amount’ and you can enter the amount needed. Then tap 'Request' and select a reason. You’ll get feedback about the request in realtime and further details will arrive at you email inbox, ok?”
39. Hi, I’d like to get a nubank card... I lost my password!!!!... ... order-card recover-password ...
42. ESTABLISHING A MODELLING CYCLE EARLY IS MORE IMPORTANT THAN APPLYING THE MOST INNOVATIVE MODELLING TECHNIQUE.
43. RULES ARE A WAY TO JUMPSTART SUCH PROJECTS AND CREATE A MODELLING CYCLE. DON’T BE AFRAID OF THEM.
44. FUTURE WORK • ML everywhere: reply chats, FAQ, dimensioning • Rethink the self-help experience to make it more interesting and effective
45. THANKS. FUAD.SAUD@NU.BANK LINKEDIN.COM/IN/FUADSAUD PAULO.CASTRO@NU.BANK GITHUB.COM/PRCASTRO