CollectWise AI-POWERED COLLECTIONS ORCHESTRATION PLATFORM

Inspiration

Collections is a critical function for every lending business, yet borrower engagement often relies on repetitive, one-size-fits-all calls. A borrower who has missed an EMI by two days may need a gentle reminder, while someone overdue for 45 days may need a completely different conversation focused on understanding their financial situation.

We wanted to explore a simple but powerful question: What if every borrower received the right conversation, from the right AI agent, at the right time?

That idea led to CollectWise — an AI-powered collections orchestration platform designed to make debt collections smarter, more personalised, and more empathetic.

What it does

CollectWise intelligently orchestrates borrower conversations based on their Days Past Due (DPD) and collection requirements.

Instead of deploying a single generic voice bot, CollectWise uses five specialised AI voice agents, each designed for a different delinquency stage:

Early stage (1–3 DPD): Gentle payment reminders.

Follow-up (4–15 DPD): Understand reasons for payment delays.

Mid-stage (16–30 DPD): Encourage realistic payment commitments.

Late stage (31–50 DPD): Explore feasible repayment arrangements.

Resolution (51+ DPD): Understand financial hardship and identify appropriate next steps.

The platform automatically identifies a borrower's DPD bucket, selects the appropriate AI agent, initiates an outbound call, and captures the conversation outcome.

Following the call, the system updates the borrower's record with relevant information such as call status, payment commitments, expected payment dates, conversation summaries, and recommended next actions.

The goal is not merely to automate calls, but to make every borrower interaction more contextual and outcome-oriented.

How we built it

We built the CollectWise MVP using three core components:

Google Sheets: Acts as the central collections dashboard and borrower database. It maintains customer information, overdue amounts, DPD, call eligibility, and collection strategies.

Google Apps Script: Powers the orchestration layer. It reads borrower data, determines the appropriate DPD bucket, selects the corresponding Vaani agent, triggers outbound calls, and processes webhook responses.

Vaani.dev: Powers five specialised AI voice agents that conduct borrower conversations using different strategies based on delinquency stages.

The workflow is straightforward:

Borrower data is maintained in Google Sheets.

The system evaluates the DPD and assigns a collection strategy.

Apps Script identifies the appropriate AI agent from the configuration.

Vaani initiates an outbound call to the borrower.

After the conversation, Vaani sends the call information through a webhook.

CollectWise updates the corresponding borrower record with the call outcome and next action.

This modular architecture allows lenders to configure different agents and collection strategies without having to build an entirely new calling system.

Challenges we ran into

One of our biggest challenges was designing conversations that were genuinely different across delinquency stages.

A generic voice agent can make calls, but effective collections require context, empathy, and a clear understanding of the borrower's situation. We had to think carefully about how each agent should approach a conversation, what information it should capture, and when it should recommend human intervention.

Another challenge was connecting the different components reliably. We worked on integrating Google Sheets with Apps Script and Vaani's outbound calling APIs, while also designing webhook-based updates to ensure that call outcomes are mapped back to the correct borrower records.

We also had to consider the sensitivity of financial conversations. Our agents are designed to avoid threatening or pressuring borrowers, making unsupported promises, or sharing sensitive loan information with unauthorised parties.

Accomplishments that we're proud of

Built an end-to-end MVP connecting borrower data, AI agent selection, outbound voice calls, and post-call data processing.

Created five specialised AI voice agents instead of relying on a single generic collections bot.

Implemented DPD-based agent routing, allowing the system to select a conversation strategy automatically.

Designed a feedback loop where call outcomes can be captured and written back to the original borrower record.

Established a modular foundation that can potentially support multiple lenders, loan products, and collections strategies.

Most importantly, we transformed an idea into a working prototype that demonstrates how AI agents can participate in a larger, intelligent collections workflow.

What we learned

We learned that building an effective AI product is not just about making an AI agent talk. The real value comes from the intelligence surrounding that conversation.

Contextual data, agent selection, conversation design, structured information extraction, and workflow automation are equally important.

We also learned how powerful tools like Vaani.dev, Google Apps Script, and Google Sheets can be when combined to rapidly prototype an AI-powered business workflow.

Most importantly, we realised that collections is not simply a communication problem. It is a decision-making problem that requires the right engagement strategy for every borrower.

What's next for CollectWise AI-POWERED COLLECTIONS ORCHESTRATION PLATFORM

Our vision is to evolve CollectWise into the AI Operating System for Collections.

Some of the capabilities we want to explore next include:

Omnichannel collections: Extend borrower engagement beyond voice calls to WhatsApp, SMS, and other communication channels.

Predictive collections intelligence: Identify borrowers at risk of delinquency before they miss payments.

Dynamic agent orchestration: Select conversation strategies using repayment history, previous commitments, customer responses, and behavioural signals.

Collections analytics: Give lenders visibility into contact rates, promise-to-pay rates, kept promises, recovery rates, and agent performance.

Human-in-the-loop workflows: Intelligently escalate complex cases, disputes, and financial hardship situations to human collection teams.

Continuous optimisation: Learn from conversation outcomes and improve engagement strategies over time.

We believe the future of collections will be more intelligent, personalised, and empathetic — where technology helps lenders improve recovery outcomes while treating every borrower with dignity.

CollectWise — Smarter collections. More meaningful conversations. Better repayment outcomes.

Built With

• appscript
• sheets
• vaani
• webhook