Mateus • Design
Amsterdam —

Nubank Personal Loans

Transparency drives better decisions.

Role
User research, product design
Year
2019

Ten months, and a textbook double diamond. There was no researcher on the team, so I ran the research as well as the design.

It’s still the project I’m proudest of.

Three screens from the Nubank loan product: choosing an amount and a term, a list of active loans, and the instalments of one of them.

Discover & define

Market analysis

Day one, and we knew almost nothing about personal loans. We went after market reports and found exactly one. Not much — enough for a landscape.

13%

want it approved fastest

17%

have no time to compare

21%

need the instalment to fit

Surveys

The report couldn’t tell us how our own customers differed from everyone else. Surveying our base was the fast, cheap way to find out.

Most people borrow from the bank they already have, after comparing 1.5 institutions.

User interviews

So we brought 14 of the respondents into the office and asked why.

“I can’t understand why they treated me so badly. I went there to pay.”

“I asked for R$1,000 and they charged me R$1,200. It’s absurd.”

Personas

Cooper’s method: personas built on behaviour, with demographics only where they earn their place. Behaviour becomes a persona in three moves.

  • Define the rulers — agnostic and clear. Tech savviness, dependence on an account manager, how money gets discussed at home.
  • Place every interviewee on every ruler, from what they told us.
  • Read the clusters. When 2, 5 and 9 keep landing together, that’s a persona.
A whiteboard covered in behaviour rulers, each with fourteen sticky notes placed along it and clusters circled in red.
Fourteen people, fifteen rulers, and the clusters circled in red.
Two persona cards side by side: Chris, the primary persona, and Robbie, the secondary one, each with a statement about what a loan should be and a list of goals.
Chris wants a way out. Robbie wants a way there.

One primary, one secondary — so whatever we designed for Chris would also serve Robbie, with some tailoring.

Design & deliver

Information architecture

There are only three things anyone does with an interface: take information in, change it, and move through it. I mapped the loan that way before drawing a screen.

  • Green is what we need from the customer.
  • Yellow is what we owe them, so the decision is theirs to make.
  • White is a group of related information, not a screen. Relationship, not sequence.
The information architecture in four steps — dashboard, simulation, contract, terms — each holding green input blocks and yellow information blocks.
Dashboard, simulation, contract, terms.

The hiring flow

We built three of them: a list of ready-made proposals, a simulator, and a chatbot.

Three first-round concepts on phone screens: a scrolling list of loan proposals, a simulator with amount and term fields, and a chatbot asking how much the customer needs.
Round one — proposals, simulator, chatbot.

The simulator won. The chatbot was warmer, but too linear to leave anyone feeling in control, and the proposals buried people in numbers.

Four simulator variants: a stepper, a dropdown list, two sliders, and a row of number chips.
Round two — four ways to set an amount and a term.

A few rounds later, four simulators went to user testing. The stepper won, and being the plainest of them, it was also the one that held up at every screen height.

The simulator in three states: typing an amount on the keypad, stepping the number of instalments, and picking the first payment date from a calendar.
The simulator in detail.
The same simulator screen on four handsets of increasing height, the spacing stretching to fill each one.
Vertically responsive.
The complete hiring flow across ten screens, from the offer through purpose, simulation and the full cost, to the terms, the card PIN, naming the loan, and the money landing in the account.
The whole flow, offer to money landing.

The rest of it

I’ve stayed on the hiring flow to keep this short. Briefly, the rest:

The loans dashboard in four states, each with a different title: nothing borrowed, all up to date, one instalment due in seven days, and two overdue.
The title carries the state — nothing borrowed, on track, due soon, overdue.
One loan’s instalments, with the paid ones collapsing into a single expandable row and struck through when shown.
Paid instalments collapse into one line, so progress is something you can see.
Choosing which instalments to pay early: as more are ticked, the total rises and the amount saved rises with it.
Paying ahead, and what it saves. This one deserves a case study of its own.

Results

5×

the benchmark, 10× at roll-out

4+

NPS from people who simulated

“I want to thank you all for this dream come true. Yeah, we bought the land. Thank you, thank you, thank you.”

Next project