Backlot · Fintech · 0 → 1 · Mobile + web

One payment chain across three apps, from the crew who spend to the producer who pays. Every claim is paid once.

Measured in testing · v1 → final

4 → 0testers paid a planted duplicate
41 s → 9 sto approve a claim
64% → 94%task success
58 → 86SUS score
After: the Backlot producer console overview
Before: a WhatsApp group, a paper bill, a petty-cash sheet, a UPI payment, a notebook and a reminder, none of them connected
BeforeAfter

Before: money lived in six places that share nothing. After: one console that sees all of it. Drag to compare.

Role

Founding product designer

Scope

Research to tested product in 14 weeks

Team

Founder, PM, two engineers

Research

50 interviews on 4 live sets

Product

Snap the bill, say who paid. Sent in 20 seconds.

Problem

About 40% of on-set spends come out of a crew member’s own pocket. Here is one ₹1,440 bill, today and with Backlot.

Today3–5 days
  1. Day 1 · 9 pmNaveen pays ₹1,440 for lanterns, from his own cash
  2. Day 2The paper bill goes in a pocket, or gets lost
  3. Day 3The coordinator can’t recall approving it
  4. Day 4The producer checks it isn’t a duplicate
  5. Day 5, maybeNaveen gets his money back
With BacklotSame night
  1. 9:02 pmSnaps the bill. Spot reads it
  2. 9:03 pmSays who paid. Sent to Riya
  3. 9:10 pmRiya approves between shots
  4. Tonight’s batchD. Rao pays tonight’s batch, once
  5. Same nightIn Naveen’s account, with a UPI ref

Executive producer. All 8 producers I spoke to fear paying twice more than paying late.

So I reframed the brief

Asked forAn expense app for crew
BuiltA money loop for three roles
Asked forCapture the receipt
BuiltPay the right person, once
Asked forAI that reads bills
BuiltAI that catches risk before money leaves

Research

50 interviews · 28 crew, 14 coordinators, 8 producers · 4 live sets

11 of 14coordinators couldn’t remember by evening what they’d approved on setApprove on the phone, on set
~40%of spends come from a crew member’s pocket, but every tool had one payee field“Who paid?” comes first
8 of 8producers fear paying twice more than paying lateA pay-once guardrail

Built for a film set, not an office

No signalClaims save offline and send later.
Shared phonesPhone number and OTP. Light screens.
Mixed literacyNumbers and icons first. Any language.
UPI isn’t instantQueued, sent, landed. Shown honestly.

The apps

Crew app asking who paid for this
Crew app asking who paid for this

Producer console

Overview

Opens on what needs him, with Spot’s brief and tonight’s payout.

Expenses

Every claim with its bill, who raised it and where it’s stuck.

Purchase orders

Ask before you buy. A PO changed after approval gets flagged.

Daily wages

Paid from the call sheet. Not on one, not paid.

Talent

Contracts paid by milestone, with what’s due next.

Departments

Budget against shoot progress, so overspend shows early.

Wallet

One pre-funded wallet, its funding sources and every transaction.

Compliance log

Every rupee and every decision, logged by name.

AI

Why AI here

Rules catch an exact duplicate. They miss the same bill photographed by two people, a worker on no call sheet, or a department that runs dry by day 39. Spotting those means checking every bill against every other, which no one does at 9 pm.

Where it works: four moments, three apps

Reads · crew app

Every bill is read on capture. The crew checks fields instead of typing them.

Suggests · coordinator app

Approve or decline, with a one-line reason.

Flags · producer console

Duplicates, ghost workers and edits after approval, before money leaves.

Looks ahead · daily brief

Which department runs dry, and on which day.

Kept honest by design

Producers see how Spot is doing, per check, against a target.

Never decides

It can’t approve, reject or pay. Every decision is logged under a person’s name.

Every dismissal has a reason

The reason is logged and used to tune Spot.

Rolls itself back

Below 0.70 precision for two weeks, a check returns to fixed rules.

Checked for bias

Flag rates are reviewed monthly by department and craft.

Owns its mistakes

When a hold is overturned, Spot says so on the overview for a week.

Decisions

01

Ask “who paid?” first

Generic payee fields paid shops for bills the crew had already paid.

4 of 12 → 0 of 12 paid the duplicate

4 of 12 → 0 of 12 paid the duplicate

Before: the form assumes the shop gets paid. After: one question decides.
Before: the form assumes the shop gets paid. After: one question decides.

02

Approve on the card, not a page

Coordinators work from their phone between setups.

41 s → 9 s to approve

41 s → 9 s to approve

Before: open, read, decide, go back. After: decide on the card.
Before: open, read, decide, go back. After: decide on the card.

03

An attention queue, not an inbox

The producer opens on what needs him, sorted by urgency.

64% → 94% task success

64% → 94% task success

After
Before · v1
Before · v1After
Every claim, newest first.What needs him, then tonight’s payout.

04

A flag, not a chatbot

Investors loved an “ask anything” chatbot. I ran it and a simple flag through the four tests I set for every AI feature.

The testAsk-anything chatbotSpot, as a flag
Does it remove work?Fails. Adds a conversation to keep upPasses. Checks every bill so no one has to
Can a human cheaply undo it?Passes. Close the chatPasses. Dismiss it with a reason
Is there ground truth to check it?Passes. Bills and call sheetsPasses. Bills and call sheets
Is it worth the trust it spends?Fails. One wrong money answer costs morePasses. One-line reason, never decides
Passed2 of 44 of 4
Every flag says why. Dismissing one takes a reason, and it’s logged.
Every flag says why. Dismissing one takes a reason, and it’s logged.

Guardrails

Pay once

A second press returns the same batch.

New payee check

No first payment until bank details are checked.

Ghost worker

Not on a call sheet, not paid.

Same bill, two films

Spot shows which production paid first.

Dispute lane

Out of tonight’s batch, labelled “quick check”.

Proof of payment

Queued, sent, landed, with a UPI ref.

Impact

12 testers · 5 crew, 4 coordinators, 3 producers · six tasks on v1 and on the final design

12 testers, one planted duplicate
₹₹₹₹
Nobody paid the planted duplicate.v1: 4 of 12 testers paid it→Final design: 0 of 12
Naveen · crew
Same night3–5 days
Paid back before he leaves the set.
Riya · coordinator
9 s41 s
Approves on the card, between shots.
D. Rao · producer
94%64%
Of tasks done right, with every rupee logged.

Easier to use, too: SUS rose from 58 to 86.

The startup raised a $500K pre-seed round. These results are from usability testing, not live usage.

Learnings

01

Kill the chatbot in week one

My own four tests would have rejected it. I let the debate run instead.

02

Test top-ups with the accountant

The accountant moves the money, not the producer. I tested with the wrong person.

Both prototypes run in your browser. Try paying the same bill twice.

Screens are an NDA-safe recreation; I rebuilt them. The research, testing, and decisions are mine. Names and data are fictional.