Yellowcard Onboarding

Four steps, on yellow. The first asked which country you live in, over a map of the continent. The second asked for your name. The phone number, which is what you actually log in with, came after that, so someone who already had an account and tapped the wrong button was two screens in before the flow told them so.

We ran usability sessions, a survey and a pass through support tickets before I changed anything. The findings repeated: the steps felt disconnected, the yellow made the forms hard to read, and customers in different markets said the product did not feel made for them. That last one is the one that set the direction. This was not only a shorter-flow problem.

The first thing I moved was the phone number, to the front. It is the credential, so asking for it first catches the people who meant to log in before they have typed anything else. Then the country step became redundant: a Kenyan number starts with +254. The dialling code is the same fact the map screen was asking for, already typed.

So the country stopped being a question and became something the system reads. That took the flow from four steps to three, and it did something more important than saving a screen. It meant the market could configure the account, the currency, the welcome, without the customer having to know that any of it was configurable.

If the customer has already told us something, we do not ask again.

Fewer steps did not mean fewer fields. The shipped flow still asks for a nickname, a referral code, a marketing opt-in and a recovery email. What changed is which of them can stop you. Only the information needed to create a safe, valid account blocks Next: the name and date of birth that have to match an ID, a password that meets five rules, and an email, because a phone-number login needs a way back in.

Everything else says what it is. The nickname says (optional) in the field. The referral code lives behind a link, so the field only exists for people who have one. The promo opt-in is a box, off by default, in a yellow strip that makes it clearly a choice. And the fields are grouped by why they exist: name with date of birth, password with email, so each screen asks one kind of question.

Account creation is blocked only by what is required to create a safe, valid account.

Every step has a Help? pill in the corner, next to the step count. The research said the old steps felt disconnected, and the fastest way to make a new flow feel the same is to send someone to a help centre in the middle of it. So help opens over the step, says why this screen is asking what it asks, and closes back to the step with everything typed still there.

The step count and the thin progress bar do the same job from the other side: not how much is left in the abstract, but where you are in a flow short enough to count on one hand.

The yellow was the brand, and the old flow used it as the ground for every form. In sessions people read the yellow screens more slowly and missed field labels. I ran a colour workshop with our product illustrator, and it surfaced what the sessions had shown: the load at sign-up was partly visual. The forms moved to white; the yellow moved to the places that needed emphasis, the help pill, the opt-in strip, the splash.

The same step, before and after, is the clearest way to see it. Same question, same fields, same brand. One of them you can read at a glance.

The customers who said the product felt made somewhere else were right. Nothing in the old flow knew where they were, except the country screen, which used the answer for compliance and nothing else. Once the phone step carried the market, the flow could know. The question was where to let it show.

Three ways to do it, and the difference between them is where the branch goes.

Two numbers I tracked and one I watched. Time to complete and drop-off were the ones the redesign was aimed at; consistency across markets was the guardrail. A fix that worked in Kenya and broke Ghana would not have counted, so the numbers below are across all markets together, by design.

Window and cohort: time to complete and conversion for the new flow against the previous one, measured across all markets together rather than market by market; the period and cohort sizes were not kept, so they are not claimed here.

The assumption that changed. I went in believing simplification meant fewer fields. The old flow did ask for too much, too early, and the count did come down. But the change that made the flow feel simple was not a removed field. It was deciding which complexity belonged to the customer and which the system should absorb. The country is the clearest case: the customer stopped answering a question the phone number had already answered.

What it taught me about localisation. Localisation is a decision you make at the start, and mostly it is a decision about where the branch goes. Put it early and you fork the whole flow; every fix has to be made twenty times. Put it at one point near the end and the flow stays one flow, and the market still gets its own welcome. The warmth people noticed was cheap because the structure paid for it.

Not the same outcome. A flow that is 23% faster is not a flow that is 23% better understood. Speed can come from people skipping optional fields they should have filled, or from the shorter path hiding a step they will hit later at verification. The time number is real; what it says about comprehension is limited.

What I would measure earlier. Drop-off per step, per market, from the first day. One number across twenty markets says the average moved. It cannot say whether Ghana moved for the same reason Kenya did, and the whole premise of the design was that they might not.

What is still unclear. Whether the dialling code is the right proxy for the market for everyone. Diaspora customers, people with a foreign SIM, and anyone who moves between markets get the wrong welcome and possibly the wrong configuration. The flow assumes phone and residence agree; the earlier flow, for all its faults, asked.