From a failed escrow idea to an AI co-pilot for MSME finance teams
Fintech, accounting, AI | OnTheProwl Technologies Pvt Ltd, Jan 2022 to Jul 2023
Designed an AI finance co-pilot for small business finance teams, from idea to a shipped pilot product in three months.
TL;DR
- Problem: Small business finance teams had plenty of accounting data and little clarity. Accountants exported Tally lists every day to find who owed money, then chased them by phone and WhatsApp.
- What I did: As founding designer, I interviewed 12 MSME owners, accountants and CFOs, helped the team drop an escrow product that did not solve the problem, and designed an AI co-pilot with automated payment reminders and dashboards generated from a single prompt, taking it from idea to shipped pilot product in three months.
- Result: In the pilot, payment collections rose 40% and CFO clarity rose 20% (self-reported). The prototype went to investors before development began and the company raised its first funding. The product was later shut down.
- Company
- OnTheProwl Technologies Pvt Ltd
- Role
- Founding Designer
- Team
- Four people: CEO, CTO, COO and me
- Employer period
- Jan 2022 to Jul 2023
- Industry
- Fintech, accounting, AI
- Users
- Accountants, CFOs and owners of small and medium businesses in India
- Platform
- Web application
- Scope
- User research, product vision, interaction model, UI and UX, dynamic dashboard system, investor prototype, input on tech stack
- Status
- Shipped to pilot customers, later discontinued
- Outcome
- 40% increase in payment collections. 20% increase in CFO clarity. Self-reported.
Product names and pilot customers are not named to respect confidentiality; the problem, decisions and results shown here are unchanged.
Context
The company's first product was a digital escrow for freelancers and their clients. The founders, who came from finance, wanted to solve two problems: working capital, and trust between small businesses and the people they pay or get paid by.
The pivot: the escrow did not work
The escrow solved neither problem. Freelancers did not want their money held in an account, and clients who planned to pay after completion found their money stuck before the work was done. It did not improve working capital.
I helped the team drop it. We went back to the real problem: businesses did not know where they were losing money, or who still owed them.
The new problem
Business owners, accountants and CFOs already used accounting software such as Tally and QuickBooks. These tools showed all the numbers, but the numbers were hard to read. In one interview I found out that accountants exported a Tally list every day, worked out who was due for payment, and then called people or sent WhatsApp messages by hand.
The brief from the CEO and COO, both from finance, was to build a personal assistant for the CFO of a small business.
My role
I was the founding designer on a team of four with the CEO, CTO and COO. I interviewed 12 MSME owners, accountants and CFOs. I developed the product vision with the founders, designed the whole interface and experience, built the final prototype shown to investors before development began, and took part in the tech stack decisions and the choice of the charts library. From idea to a shipped product took three months.
Constraints
Trust. ChatGPT had only just launched on GPT-3.5. The idea of a co-pilot inside accounting software was new to these businesses, and they were wary of AI handling money.
Users. Deep domain expertise and no technical background.
Time. Three months from idea to ship, with a four-person team.
Research and insight
Interviews. 12 MSME owners, accountants and CFOs.
Key finding. The problem was not missing data. It was that nobody could ingest the data they already had.
Decision 1: a co-pilot, not another dashboard
The product became a text-based co-pilot panel, like the chat panels common in apps today, where users type what they need. For example, "list the people who are due in the next five days" or "create a workflow that sends email reminders to people who are due in the next ten days". The co-pilot did it.
Behind it, automated workflows found accounts due in 5 days, 15 days or any window the user set, and sent WhatsApp messages, emails or in-app alerts to the accountant or CFO.
Decision 2: a dashboard that did not stay fixed
This was the hardest call. A single prompt could produce a full picture of a company's financial health: receivables, payables, accounts due and more. There was no fixed dashboard. If a user asked for one, the co-pilot built a custom one.
The challenge was making something that changes on request still feel reliable. We built a library of chart components with Chart.js and programmed the AI to compose dashboards only from that library.
Decision 3: built for trust
Because the idea of a co-pilot was new and the data was money, I designed for checking.
Every figure linked to the data behind it, so a user could check where a number came from.
Every action needed human approval. The co-pilot proposed, the user confirmed.
Outcome
40% increase in payment collections in the pilot businesses, which used the automated payment reminders. Self-reported.
20% increase in CFO clarity. Self-reported.
The prototype I built was shown to investors before development began.
The product shipped to pilot customers, three months from idea.
The company raised its first funding.
The product was later shut down.