Jiva

Cash Deposit: Moving SJ loan repayment off the bank counter and into the app.

What stopped SJs wasn’t confusion about the new flow but fear of it.

Role
Senior Product Designer
Team
2 PMs, 6-7 Engineers, 2 Design Researchers and 1 UX Writer
Outcome
Repayment dropped to about 7 minutes and Finance stopped checking payments by hand.

Background

Jiva extends credit to Sahabat Jivas (SJs), local partners who buy corn, order inputs and lend to farmers in their area. SJs repay that credit in corn or in cash.

Cash repayment was never designed. It was a workaround that hardened into standard procedure: collect money from farmers, travel to a bank, deposit into a permanent account number, keep the paper slip, photograph it, upload it, then send a handwritten tabulation to an Activation Coordinator who re-uploaded it to a Google Form for Finance.

Jiva, not the SJ, carried the interest on every day that money sat in transit. At 2,500 SJs this was survivable. At the scale we were planning for, it was not.

My role

I was Senior Product Designer and owned the work end to end alongside two PMs and an engineering lead, with two design researchers on the team. I was responsible for the design and the UX writing, which I worked through with our UX writer. I designed both sides of the system: the SJ payment flow and the Finance dashboard that had to stay in step with it.

Repayment went from a reported 28 hours to about 7 minutes in the app, and Finance stopped checking standard payments by hand.

Mapping the whole service before touching a screen

Before designing anything, we mapped repayment end to end: what the SJ does at each stage, what they feel, which teams touch it, and what each failure costs Jiva. It ran across five groups — SJs, Activation Coordinators, Growth, Customer Support and Finance, and it told us this was a service problem rather than a screen problem.

Three stages: learning what’s owed, collecting cash and depositing it at the bank, recording it in the app. MCs were later renamed Sahabat Jivas.
Three stages: learning what’s owed, collecting cash and depositing it at the bank, recording it in the app. MCs were later renamed Sahabat Jivas.

The cash deposit step was where it broke. SJs collect cash without categorising it the way Jiva’s ledger needs, and the bank’s own deposit process never asks them to account for the breakdown, so there was no existing habit to build on. They recorded deposits fast to avoid penalties, not accurately.


One of seven problems the blueprint surfaced. This is the one the project was built to solve.
One of seven problems the blueprint surfaced. This is the one the project was built to solve.

Downstream, that cost more than time. Every claim had to be checked by hand for correct allocation of outstanding type. Finding the true outstanding meant mapping corrective adjustments first, which delayed commission payments. 47% of calls from customer care and Growth to SJs went unanswered, and field teams were chasing farmers who had already repaid.

What each team carried, and what it meant for the business.
What each team carried, and what it meant for the business.

The money behind it: with about 2,500 active SJs carrying roughly Rp45M each, every day repayments sat uncollected cost Jiva an estimated Rp22–45M, about $1,500–3,000, in interest. Separately, 12.9% of SJs had defaulted on procurement cash advances at an average of Rp25M each, around Rp8.1 billion at risk while the records were wrong.


Choosing what not to solve

Automating the payment could not fix all seven problems, and pretending otherwise would have set the wrong expectation with leadership.

We took on data inconsistency and operational delay. Removing the Google Form, the paper slip and the manual check removed the places where the data went wrong, and showing the outstanding per bucket removed the guesswork behind it. Attribution improved sharply without reaching zero: an SJ can still allocate wrongly inside the limits.

We partly addressed a third, SJs not knowing what to pay or by when, with overdue nudges, but not the penalty rules themselves.

We left the rest: that SJs can’t see how commissions and adjustments reduce their outstanding, that planning repayment across several types before their due dates is genuinely hard, and that too many Jiva teams contact the same SJ about the same debt. Those needed policy and operating changes, not a payment flow.

Designing the payment flow

Moving repayment online was an org-level decision. We were scaling, and depending on bank counters didn’t scale with us. One-time account numbers, locked to an exact amount and expiring in 24 hours, came with the payment provider.

What that left me was a harder design problem than the one it solved. SJs had memorised a single account number for years, and now it changed every time.

Every bucket shows what’s owed against it. A deposit has to be split across outstanding types, and in the old flow SJs did that split from their own handwritten records with no way to check it against Jiva’s. The allocation screen now lists each type with the amount owed, totals the deposit as the SJ builds it, and won’t let them proceed on an invalid one. Research had found SJs recording deposits fast to avoid penalties rather than accurately. Showing the number removed the conflict between speed and accuracy.

Buckets with nothing owed stay on screen, greyed out, under the line “You have 0 outstanding for these 2 loans.” Hiding them would have been tidier and would have left SJs wondering where a category went.

The screen says what changed. A banner on the allocation screen reads “You no longer need to upload proof of transfer in the payment process.” The explainer video did that work before the SJ opened the app; this did it at the moment they would have reached for their phone camera.

Deadlines sit in an alert bar, not in body text. A pending payment carries “Pay before 22 Nov, 14:59” where it can’t be skimmed past. An SJ sees four statuses in their history: waiting for payment, approved, expired, cancelled.

Failure got the same attention as success. The outstanding screen has seven states, including an API failure, a general error, and a screen of its own for no internet, because SJs work in fields and storage rooms on low-end Android.

Two levels of nudge on the home screen. A neutral banner when something is owed, a red one when something is overdue: clear these as soon as you can. The outstanding screen then flags each overdue item with its own due date.

Farmer selection stayed as it was, so each payment still ties to the right credit record.

Designing the Finance side in parallel

The old process had a shadow system: a Google Form, a separate Finance record, and a handwritten tabulation feeding both. Ship the SJ flow alone and that shadow system reappears somewhere else.

So the Retool dashboard was designed alongside the flow, not after it. Finance needed to see payments as they landed, resolve exceptions and handle disputes, and every state an SJ could see needed a matching path internally.

Designing for fear, not confusion

Usability testing found the flow wasn’t confusing. It was scary. Three anxieties came up: a number that changed every time felt untrustworthy, “Virtual Account” meant nothing, and many SJs believed a bank visit was still required.

☝🏽
Change management spread through the same social infrastructure the old broken process had relied on.

None of those could be answered on the payment screen, because by then the SJ had already decided whether to trust it. So we answered them before it: a sub-minute explainer video, made with Brand and Comms, that ACs shared on WhatsApp ahead of launch.

69% of views came from returning viewers, many rewatching while making their first payment. 70% of traffic came from WhatsApp, which told us ACs were sharing it without being asked.


Getting the organisation behind it

This changed how money reached Jiva, so it needed more than design approval. I presented the case to leadership in a pod meeting with the CEO present. The goal wasn't to walk through screens, it was to make leadership certain enough to green-light the change. The CTO noted afterward that the presentation gave the organisation the confidence it needed to move forward.

Outcome

  • Repayment moved from a reported 28 hours to about 7 minutes in the app. The 28 hours came from survey responses and AC interviews, covered the full cycle including visits to farmers, and was never instrumented. The 7 minutes covers the payment in the app, start to finish.
  • Gone from the process: the bank visit, the banking-hours dependency, the paper slip, the photo upload, the Google Form, and Finance’s manual verification of standard payments.
  • Cash rotation fell from 14 to 10 days the month after launch. A penalty policy and the Total Outstanding redesign landed in the same window, so this isn’t mine alone. At the interest figures above, four days of rotation was worth roughly Rp190–390M a month, about $13,000–26,000, across all three.

What I learned

  • Designing for behaviour change means designing the transition, not just the destination. The video, the first-time screen, the explicit VA copy, none of these were features. They were the transition layer between what users knew and what we were asking them to do. Easy to skip when focused on shipping; on this project, it was load-bearing.
  • Precise language is a design tool, not a copywriting flourish. The exact wording on the payment deadline prevented real financial errors. Generic fintech copy in that screen would have caused real harm.
  • The internal tool is part of the service. The Retool dashboard wasn't a side task, the SJ-facing flow and the Finance tool were one system, as the diagnosis above laid out.
  • Organisational alignment is design work. Getting leadership confident enough to ship a change this significant required the same clarity of thought as the interface itself.

Built in India & UK · Every heading was once a different heading. © 2026 Sai Shinde