3. Our domain
is very
complex.
FINTECH BUSINESS
People that own Nubank
some amount of money.
DEBT
We need our customers to
understand some very
complex rules.
REGULATION
People get scammed and
are frauded.
FRAUD
Failures in systems in other
banks or services.
THIRD PARTIES
We develop and deliver new
products which users,
sometimes, don't
understand.
NEW PRODUCTS
Some of our users are not
tech savvy.
DIGITAL
10. Rules
Uses human knowledge about the business.
Human made
If the routing is wrong, the XPeer will tell us.
Let the user correct it
Very fast and good for an MVP.
Good to prototype
Better at corner cases and learns things that
are hard to encode as rules
Learns to route
The model is as good as the data we have,
so we need lots of it.
Improves with time
Needed for new products and to give the
teams more flexibility on how they route
their tickets.
Rules when necessary
Machine Learning
11. Language Models
Text Waitlist?
Pq não fui aceito? Yes
Como faço um cartão? No
Quero mais limite! No
...
Não queria mesmo! Yes
Xpq
Xnã
o
Xaceit
o
Xcartã
o
... Xmesmo
Waitlist?
1 1 1 0 0 Yes
0 0 0 1 0 No
0 0 0 0 0 No
...
0 1 0 0 1 Yes
F(Xpq
, Xnão
,Xaceito
,Xcartão
,…, Xmesmo
) → {Yes,
No}
F(Xtext
) → {Yes, No}
Is this email a waitlist ticket?