Project overview

I owned the application flow end to end: the drop-off analysis and interviews, the step structure, the UI and the microcopy, through to QA with engineering. The work was part of a 10-week redesign, from research to implementation.

The problem: an application people gave up on

Female entrepreneurs were abandoning loan applications partway through. The old process was confusing and long, and applicants could not see how far along they were. It took a median of 6 minutes to complete, and many applicants never finished.

The old application lived inside the member dashboard and asked one or two questions per screen: 12 question screens, an identity check that appeared as a pop-up, a summary and a bank details screen: 15 screens before submitting, then a confirmation.

Going through the old designs screen by screen showed why people gave up:

• The loan came last. The first screen promised "Let's get some details about the loan you want", but the amount, purpose and repayment period only appeared on screen 12.
• Questions that had nothing to do with the loan. Instagram and LinkedIn handles, professional associations and "Who referred you to us?" all sat in the middle of the flow, each on its own screen.
• Progress that did not mean anything. A bar of 15 unlabelled segments, no time estimate, and "Great you're almost there!" printed under every screen, including the first.
• A surprise identity check. Halfway through, a pop-up announced "Hey, it's getting serious. We need to verify your identity" with a single Proceed button and no word on what was needed or why.

Analytics showed that only 22% of people who started an application submitted it. The biggest single drop was the identity verification pop-up, where a quarter of all applicants left. In interviews, applicants said it felt sudden and they did not know what they would be asked for.

What shaped the design

User research and support ticket analysis pointed to overwhelming form complexity as a main reason people abandoned. I interviewed 12 applicants, 8 who had abandoned and 4 who had finished, and reviewed 340 support tickets from the previous 6 months. 41% of those tickets were about the application, most often "where was I?" after an interruption and "why do you need my Instagram?"

• Applicants are multitasking, running a business while caring for children, so a long application gets interrupted.
• They need to know what they are getting into: clear communication about fees, requirements and timelines.
• Chunking has to match how they think. Information needed to be grouped the way applicants think about their business and their loan, not the way the data is stored.

These led to two principles for the flow: every step shows progress, and every decision point has contextual help.

Decision 1: Five grouped steps instead of a long chain of screens

I grouped the questions into five steps that each answer one thing, starting with the business (name, category, address, years in business, registration number, monthly revenue, employees) and then the loan (amount, purpose, duration, number of installments, repayment method). Each step shows where you are and how long is left.

I tested two alternatives with 6 applicants on paper prototypes. One long scrolling page felt overwhelming and lost people on small screens. One question per screen felt endless and doubled the number of taps. Five steps matched the five things applicants said they think about, and each step fits in about two minutes, which is short enough to finish between interruptions.

Every question in the old flow was kept, merged or removed:

Decision 2: Save and resume, and smart defaults

Because applicants are often interrupted, they can save progress and resume anytime. Smart defaults pre-fill common answers based on business type, so there is less to type.

Answers save automatically as you go, and the rail confirms it ("Saved on this device just now"). "Save and exit" sits in the top bar on every step. Coming back shows a "Welcome back" screen with the saved application: the next step, percentage complete, the amount asked for and the time left, with one button to continue where you left off and an option to start again. Uploaded documents are kept too. In the prototype the application is saved on the device; in production it would also be tied to the applicant's account so she can resume on another phone.

Smart defaults use the business category from step 1. For a fashion and tailoring business, step 2 opens with a short note saying which answers were pre-filled, "Buy equipment" is marked Suggested, and direct debit is marked Recommended. Every default can be changed.

Decision 3: Review and consent before submitting

The flow ends with a review of every answer, editable before submitting, and a short consent step with explicit checkboxes for verification, terms and repayment authorisation. Applicants see exactly what they are agreeing to.

Financial scams are common, and in interviews several applicants said they had been asked to "approve" debits they did not understand. Splitting consent into three plain-language checkboxes, instead of one "I agree" box, meant applicants knew exactly what Shecluded could do, especially the repayment authorization. Support tickets asking "what did I agree to?" dropped to almost none after launch.

The consent step spells out each permission in plain language: checking identity, BVN and business details with credit bureaus; confirming the loan terms; and authorising the exact monthly collection (for example ₦98,334 a month for 6 months). A line under the checkboxes reassures applicants that submitting does not commit them to a loan.

This replaces the old "Hey, it's getting serious" pop-up: identity and BVN checks are now something the applicant agrees to, with the reason stated, instead of a surprise halfway through.

Documents

The old flow did not collect documents at all. Staff chased them by email and WhatsApp after submission, which added days to every decision. The new flow asks for them up front, so documents needed the most care. Each document has its own card with a clear status, and uploads continue from where they stopped if the connection drops.

The result

Three months after launch:

The rate of sign-ups who started an application stayed roughly flat, while application completion went from 22% to 44%, so most of the lift came from the application flow.

What I'd do differently

• Test with users at different business maturity stages earlier in the process.
• Consider users with older or lower-quality phones throughout the design.

I would test the document upload earlier. It was a brand new step, so there was no old data to warn us, and I only saw how often it failed on weak mobile connections after we watched session recordings in week 6. Testing uploads on real devices over mobile data in week 1 would have given engineering more time for resumable uploads.