Field collectors visit customers who are behind on their loan payments. I designed the app they use for the whole visit: choosing who to see, finding the house, recording what happened and giving the customer a receipt. Before this, most of them worked from paper printouts instead of the official app.
Collectors
1,000+
Android, tier 2 and 3 cities
Cases in the field
~50,000
a month, out of over 1,50,000 cases in collections
Collected by the field
~₹100 cr
a month
Updates in the app
~2,80,000
a month, about 2,00,000 of them by voice
Role
Product Designer
Research to handoff
Team
1 PM, 4 engineers
I was the only designer
Timeframe
About 6 months
~40cases a collector handles each month, up from 30
90% → ~0collectors on paper liners (estimate)
~2%higher collection rate, from faster collectors
Mine: the research, every screen and flow, and design checks with the engineers as the app was built. I was the only designer, so there was no separate researcher either.
Shared: scope and what to build first, with the PM. I proposed the case priority order and took it to business and product, who accepted it. What would work on cheap phones with patchy network was worked out with engineering.
Set by business and compliance: the selfie and location proof at each visit. My part was fitting it into the update flow without making it a separate task.
02 · Context
Everyone used the official app, but they worked from paper.
A field collector gets 40 to 60 overdue cases a month, across every loan product the company sells, and works from a low or mid-range Android phone. The tool they were given was LCS, built on Microsoft Power Apps. Updating it was mandatory, so usage looked like 100%. In practice it was done in the evening.
LCS recorded every update, but some basics weren’t there yet. Phone numbers were often absent and locations were never there, so collectors looked the address up on a map and then searched on foot. That cost 15 to 20 extra minutes per visit. The update form had a lot of fields, and cases came in no set order.
So about nine in ten worked from paper "liners". Excel printouts, one case per row. LCS got updated in the evening, because it had to be.
The real record lived in phone calls. Every evening each collector called their BRM (the manager who oversees 4 to 5 collectors) and read out the day's updates, 15 to 20 minutes each, while the BRM typed them into another Excel sheet. Customers who paid cash got no receipt.
How a case moved, before and after
Before
LCS on Power AppsPrinted linersManual map searchEvening calls to the BRMBRM's Excel trackerOther manual tasks
↓
After
Field Collections AppOne record per case
The paper liner
A printed liner. All customer details have been changed.
The old app, screen by screenLCS · 6 screens
01
Home
Power Apps home, with the collections app under Recent apps.
02
Menu
Collection Cases, histories and Deposit Cash.
03
Case list
Active Collection Cases, one row per loan.
Three more screensLoan summary, EMI details, Collection processShow more screensHide screens
04
Loan summary
Open a case to see the loan and applicant.
05
EMI details
Amounts overdue and outstanding, one long scroll.
06
Collection process
Team, owner, managers and field agent for the case.
Screens from a collector's phone, in the order a case was opened. Names, loan IDs and amounts have been changed.
How I researched it
I spent full workdays with 5 collectors across 4 branches, talked to their BRMs about the daily calls and the Excel upkeep, and talked to product and business about what the app had to do and what compliance needed.
03 · Goals
Usage was already 100%, so I needed a better test.
Updating LCS was mandatory, so counting logins told me nothing. The real test was whether collectors still reached for the paper liner.
Observed
The liners disappear.
If collectors still print their cases every morning, the app isn't working.
Observed
Time to reach the customer.
The 15 to 20 minutes lost finding the house was the biggest single waste in a visit.
Business data
Collection rate.
Collections rise as the loan book grows, so the rate is the number that matters. The team isn't growing, so each collector has to get through more.
Observed
Effort spent reporting on cases.
The evening calls and BRM spreadsheets should shrink once everyone can read one case history.
04 · Improving what was already there
Three parts of the daily work, made lighter.
LCS got the basics in place, and updating it was mandatory. Three parts of the daily work stood out as the ones to make lighter.
4.1 · Case order
Start the day with the right customer.
What I saw
Collectors worked from paper liners with no set order, so they picked cases by instinct or by who was nearest. A customer due to pay that day could be anywhere in the list, and the collector had to check each case to find out.
What I did
Cases now come in a set order: On PTP, then PTP Not Given, then Not Contactable, then New Case. A promise to pay (PTP) due today lands at the top on its own, because it is the most likely money of the day. I took the order to business and product before building it, and they accepted it. It then ran in the pilot with 11 collectors.
What I gave up
Ordering by distance, for now. The best answer is a planned route for the day, which is designed and still to be shipped.
See iterations →
4.2 · Update flow
Say what happened first, then fill in only what that needs.
What I saw
The update form asked for every field for every outcome, so it often waited until the evening.
What I did
A bottom sheet asks one question with four answers: not contactable, promise not given, promise received, payment received. A short form then asks only what that answer needs, and the selfie and location proof is captured in the same pass. We tested it with prototypes in product meetings, then it ran in the pilot.
See iterations →
REPLACE: Update flowThe bottom sheet with its four answers
4.3 · Voice updates
Found after launchRolled out
Speak the update instead of typing it.
What I saw
While checking real updates after launch, we found the remarks field, where collectors add extra details and comments, was often filled in wrongly, because it is hard to type on a phone. Collectors find it much easier to speak about a visit than to write about it.
What I did
The collector speaks, in their own language: English, Hindi, Tamil, Kannada and a few more. The app transcribes and translates it, fills in the update form, and adds the context to the remarks. The collector is then shown the filled form and can check it before submitting. If the form comes back half filled, they complete the rest by hand.
What I gave up
Full automation. The collector always sees the form and checks it before it is submitted.
About 2,00,000of about 2,80,000 dispositions a month are now made by voice
4+ languagesEnglish, Hindi, Tamil, Kannada and a few more
REPLACE: SpeakingThe collector speaking the update, with the language shown
Step 1Speak
In the collector’s own language.
REPLACE: Filled formThe form filled from the voice note, ready to check
Step 2Check
The filled form, before submitting.
REPLACE: Half-filled formA form that came back half filled, with the empty fields marked
If neededFill the rest
A half-filled form is finished by hand.
REPLACE: A real remark, before and afterOne example with all details changed: what was typed, next to what voice wrote
REPLACE: A collector quote about voiceOptional, one or two lines
See iterations →
05 · Adding what the field needed
Four things collectors had been managing on their own.
Collectors had built their own ways of working around what the tools didn’t cover: a maps search for the address, a guess at who they were meeting, a call to the BRM every evening, and cash with no receipt. I made each of them part of the app.
5.1 · Finding the customer
Put the customer’s addresses one tap from navigation.
What I saw
Phone numbers were often missing and locations were never there, so collectors looked the address up on a map and then searched on foot. That cost 15 to 20 extra minutes per visit.
What I did
The case details now show the addresses collected when the loan was taken: the current address, the employer’s address and, where there is one, the collateral address. Each has a Navigate button that opens Google Maps with the address already set, so the collector goes straight to the door. These were already in the loan system, so there is no extra step.
See iterations →
5.2 · The customer’s photo
See who you are meeting.
What I saw
On a visit, collectors weren’t always sure who they were talking to: the customer, a family member or someone else. Anyone can pass as the customer.
What I did
The customer’s photo now shows in the case details. A photo is always taken when the loan is given, so we pulled that one in, and the collector knows who the real customer is. Tapping the photo opens it full size.
See iterations →
5.3 · Shared case history
One history of every case, for collectors and their managers.
What I saw
Different collectors handle the same customer in different months, and each one started from nothing. Every evening each collector also called their BRM and read out the day’s updates, 15 to 20 minutes each, while the BRM typed them into Excel.
What I did
Everyone sees the case history in the same way. Before calling or visiting, the collector sees the past conversations and the tags earlier collectors left, and BRMs and ARMs read the same history, so the evening call isn’t needed. Every update, including the ones spoken by voice, adds to it.
See iterations →
5.4 · Payment and receipt
Pay by QR code or link, and leave a receipt behind.
What I saw
Customers who paid cash to a collector got no receipt.
What I did
Cash collection has stopped. Collectors share a QR code or a payment link, and the customer gets a receipt. Payments reach the account after reconciliation. Making that instant is a next step.
REPLACE: PaymentQR code or the payment link screen
Step 1QR code or link
Shared with the customer.
REPLACE: ReceiptThe receipt the customer gets
Step 2Receipt
The customer gets one every time.
See iterations →
06 · Designed, not shipped yet
What I have designed for what comes next.
These are designed but not live. They show where the app goes next, and none of the results further down come from them.
6.1 · Finding the customer on a map
Designed, not built
Show every location we have, not one pin.
What I saw
The addresses from loan time are not always where the customer is today, and collectors only see the current, employer and collateral addresses. Our customer app can capture a customer’s location, which could give the collector more places to look.
What I designed
A map with numbered candidates: the addresses from loan time, the location the customer applied from, and any location the customer app captures. The collector picks one and navigates in one tap, and makes the final call.
Options I weighed
Leave collectors to the manual map search. Show one stored pin. Or show every location we hold for the customer.
What I gave up
A cleaner map with one confident pin. A single stored pin can be wrong, and a wrong pin costs a collector’s trust fast.
Status
Designed, not built. The extra locations would come from the customer app.
REPLACE: Manual searchThe address on a liner, or typed into a maps app
BeforeManual map search
Look the address up, then search on foot.
REPLACE: One-pin versionA map with a single stored pin
OptionOne stored pin
Cleaner, but a wrong pin costs trust fast.
DesignedEvery known location
Numbered candidates. The collector makes the final call.
See iterations →
6.2 · A planned route for the day
Designed, not built
Plan the day’s visits as one route.
What I designed
An ordered visit plan for the day, so the collector doesn’t have to work out the route from the case list. The case order puts the customers most likely to pay first. A route would take distance into account too.
See iterations →
REPLACE: Route planThe day’s visits as one ordered route
Also next
Payments showing in the account instantly
Daily incentives, so effort spreads across the month
07 · The full design, in one visit
One visit, start to finish.
A prototype of the whole design, from opening the app to the receipt. It includes the designs that haven’t shipped yet: the map and the route plan.
REPLACE: Prototype or screen recordingOne visit in one pass: home, case detail, update, proof, payment and receipt. The Home screen is standing in until then.
08 · Delivery
We piloted in one branch before it went to a thousand people.
I worked with one PM and four engineers for about six months.
Scoping came before the spec. The field findings shaped what we built first, and the big decisions were argued out with prototypes.
The pilot ran at Vasanth Nagar, Bengaluru, with 11 collectors. We wanted an app that needed no training, but we still gave a short demo in the branch. What we learned went back into the build before the wider rollout.
Every screen went through Visual QA. Frontend engineers built each screen, and as the designer I had to review it against the design and approve it before it could go forward.
Language and devices. The interface is in English, which nearly all collectors are comfortable with alongside Hindi and regional languages. Voice updates let them speak in their own language instead. Everything was built for low to mid-range Android phones.
What shipped
Prioritised cases and monthly numbers on home
Customer addresses with one-tap navigation
Past conversations and customer photo
Two-step update with selfie and location
Voice updates in the collector’s own language, added after launch
QR and payment links, with receipts
One case history for collectors and their managers
What I have designed but not shipped yet is in section 06.
09 · Results
The printouts are almost gone, and the same collectors handle a bigger book.
~40
cases a month, per collector
Before30
Now~40
Business data
Collectors now handle about 40 cases a month, up from 30.
The team stayed about the same size, so the app is what let the same collectors take on more.
~2%higher collection rate, from collectors getting through their cases faster
~90% → ~0
collectors on paper liners (estimate)
Before
Now
Observed in the field
Almost everyone has stopped using liners.
Everything a collector needs is now in the app, so the printout has nothing left to do.
One record
read by collectors, BRMs and ARMs
Collectors4 to 5 per BRM→BRM→ARM
Observed in the field
BRMs and ARMs see the team’s day without waiting for the call.
Every update sits in the same case record, so the evening calls of 15 to 20 minutes per collector are no longer needed.
What I learned
The last week of the month still does most of the work.
Collectors do most of their work in the final days of the month, because that's when the numbers get counted. The app made each visit faster but hasn't changed when visits happen. We're trying to change that with incentives that reward working through the month, starting with daily incentives.
Get in touch
If this was interesting, I'd like to hear from you.
I'm happy to walk through the project or answer any questions.