In the dynamic world of digital lending, efficiency isn't just a luxury—it's a necessity. At Apollo Finvest while we seamlessly ploughed through most of our operational mountains with tech and automation, there was one critical area where our control was limited: collections.

Collections are the cornerstone of any lending operation. They influence not only the repayment rate but also the overall borrower experience. Unfortunately, when borrowers faced unprofessional or harsh interactions during collections, complaints would inevitably find their way back to us, sometimes even escalating to the regulatory bodies. These incidents risked tarnishing our operational expectations and, more importantly, jeopardized the borrower experience we strived to perfect.

Chapter 1: The Struggle of Manual Processes

Our initial approach to handling collections involved manual sampling of call recordings—a daunting task when you're receiving between 50,000 and 60,000 calls a month. With each call averaging about two to three minutes, it was humanly impossible to analyze every single one. Instead, our operations team would sift through a mere 5% to 10% of these calls, hoping to catch any anomalies. However, this method was fraught with inefficiencies.

Typical complaints included:

• Collection agents using rude or abusive language.

• Calls being made at inconvenient times.

Despite our best efforts, many negative calls slipped through the cracks. The operations team, bogged down by this monotonous task, couldn’t catch every issue, leading to delayed responses and unresolved borrower grievances. It became clear that we needed a solution that could handle the scale and provide real-time insights.

Chapter 2: The Birth of Senti

The Eureka Moment

The answer: AI. We needed a tool that could analyze every call, identify sentiment, and flag problematic interactions instantly. This realization led to the birth of Senti, our in-house AI-based audio sentiment analysis tool.

Building the API

Creating Senti was no small feat. We began by leveraging an archive of over five to six lakh call recordings collected over several years. These recordings were not just stored but had already been sample-checked and tagged with feedback, providing a robust dataset for training our AI model.

The real challenge lies in the language. In India, collections often occur in Hindi, with many calls switching between Hindi, English and a variety of different regional languages. We needed Senti to understand this linguistic fluidity and accurately interpret the sentiment of these multilingual interactions.

Our team spent nearly a year refining the AI model. We trained it to distinguish between positive, negative, and neutral tones, accounting for cultural nuances and linguistic variations. By the time Senti was ready, it could seamlessly analyze calls in multiple languages, delivering precise sentiment scores and remarks.

Chapter 3: Transforming Collections with Senti

Real-Time Analysis

The impact of Senti was immediate and profound. As soon as a collection call is uploaded, Senti will analyse it and provide a sentiment score and a concise explanation. This enabled us and our fintech partners to address negative interactions in real time!

Operational Efficiency

Internally, the transformation was just as significant. Where the operations team previously spent 5 to 6 hours each week manually reviewing random calls, they now needed less than 30 mins to review only the flagged negative calls. This shift allowed them to focus on more critical tasks, ensuring that every problematic interaction was handled swiftly and appropriately.

Chapter 4: A New Era for Apollo Finvest

Senti didn’t just streamline our processes—it revolutionized them. With Senti, we regained control over the entire lending lifecycle, from disbursement to collections. Borrowers received faster resolutions, and our partners could maintain higher standards of professionalism in their interactions.

Conclusion: The Future with Senti

With continuous improvements and adaptations, we are confident that Senti will not only keep up with the evolving needs of digital lending but will also set new standards for what’s possible with technology in the digital lending space.

Today many of our Fintech and NBFC partners have deployed Senti to monitor all their collection calls

This has now equipped them to ensure that every borrower interaction—from the first loan disbursement to the final collection—is handled with the care and professionalism our customers deserve.

Go ahead, try out Senti yourself: https://www.apollofinvest.com/apollo-intelligence

1. Sourcing Requirements

Our journey begins with a spark of inspiration, either from our ingenious internal teams or from our beloved external lending partners. When a requirement flows in, we dive into the task manually first. This step is our secret sauce, offering a plethora of benefits:

• Grasping the Problem: It helps us really understand the heart of the issue.

• Uncovering Surprises: We spot potential edge cases that might otherwise slip through the cracks.

• Future-Proofing: It lets us see if the feature is a long-term necessity or just a temporary fix.

And when it comes to requests from our lending partners? We prioritise those based on their current scale with us and their potential to grow even bigger!

2. Drafting the Requirement Document

After we've rolled up our sleeves and gotten hands-on, a point of contact (POC) drafts an initial requirement document. This masterpiece includes:

• Operational Insights: Detailed from an operational perspective.

• Thorough Details: Packed with all the nitty-gritty details to cover every angle.

3. Finalizing the PRD

Next, the Product Manager (PM) teams up with the POC to refine this document into a final Product Requirement Document (PRD). The PRD is our playbook, divided into three sections:

• Goal: A concise description of the feature's main objective.

• Impact: Quantitative KPIs to measure the feature's success.

• Detailed Description: Requirements laid out in simple, jargon-free language so everyone can get on board.

Once our PRD is polished to perfection, it’s shared with the relevant department for signoff.

4. Prioritizing Tasks

With a signed-off PRD in hand, we prioritize tasks based on:

• Effort and Man-Hours: Calculating the resources needed.

• Business Impact: Evaluating how the feature will boost key business KPIs.

5. Breaking Down Tasks

We then break down each task into smaller, bite-sized pieces akin to minimal viable products (MVPs). This strategy ensures:

• Frequent Releases: Regular updates keep the momentum going.

• Early Feedback: Continuous user feedback helps us fine-tune as we go.

6. Monitoring Progress

We keep a close eye on our progress. If all systems are go, we discuss the next phase with the team’s POC. This ensures:

• Awareness of What’s Next: The POC is always in the loop.

• Alignment: Both teams stay synchronized like a well-oiled machine.

7. Testing and Deployment

Our thorough PRD is a lifesaver when it comes to testing. The same document guides both the tech team and the department POC, ensuring top-notch technical and subject matter expertise. After testing, we deploy the feature and use analytics to measure its impact, guided by the KPIs set in the PRD.

8. Monthly Tech Updates

Every month, the tech and product teams put on a show with a tech update presentation, covering:

• Last Month’s Hits: Quantitative impact analysis of recent features.

• New Releases: The latest and greatest features fresh out of the oven.

This process keeps everyone accountable and fosters a data-driven approach to development.

9. Quarterly Housekeeping

Quarterly, we do some housekeeping on Sonic to identify:

• High-Utilization Features: Doubling down and making these beloved features even better.

• Underutilized Features: Analyzing and understanding the lack of use, reviving if possible, or learning from these to avoid future redundancies. If truly redundant, we eliminate them from our codebase to keep it lean and mean, reduce bugs, and lower AWS costs.

Conclusion

Our meticulous approach to product development at Apollo ensures that every feature we build is not only necessary and impactful but also a joy to create and use. By maintaining a structured yet flexible process, we continuously improve our products, driven by data and enriched by user feedback.

Using combinations of advanced technology with years of lending know-how and a mountain of data, we have kicked off a new era of credit intelligence enabling smarter and safer lending decisions across our portfolio.

Today we’ll talk about 2 such initiatives: 

The Crystal ball 📈 

One of our first developments using our legacy data is The Crystal ball. So what does it do? It's simple - whenever our credit team needs to finalize a credit policy, they use this magical crystal ball to check how such a policy has performed in the past on similar kinds of loans they intend to disburse in the future.

They just input the policy parameters like loan amount range, tenure, borrower demographics, age, etc. The crystal ball then scans through our entire past loan data matching those criteria and shows the total loans, amounts disbursed, and most importantly, the 30+ and 90+ day loan default rates we faced in the past.

This gives our credit team a powerful predictor to forecast future loan performance and confidently implement credit policies. Further, they can now experiment and see how even a small tweak of any of the factors like bureau policy, loan amounts, tenures, and borrower segments could potentially impact delinquency percentages.

The Pincode Detective🕵️

Just like The Crystal ball, we've built another data-driven tool - the Pincode Detective! This tool helps us understand location-specific lending trends across India.

The Pincode Detective generates loan performance reports based on our past lending in different geographical pin codes. For any pincode, it shows the total number of loans disbursed, the overall disbursed loan amount, and most importantly, the 30+ and 90+ day loan default rates we experienced.

With this localized loan performance data, we can analyze borrower behavior and repayment patterns, area by area, and spot pockets of risk and opportunities with ease!

The Pincode Detective helps us lend smarter by accounting for geographical nuances. We can make informed decisions on slowing down or expanding in specific regions, discontinuing certain pin codes, and even optimizing collections by doubling down on high default areas - all powered by intelligent location insights from our proprietary data reserves.

Both of these features translate Apollo's extensive lending experience and data reserves into statistical significant insights. It’s time to flex those data scientist muscles!

What is Reconciliation?

A Reconciliation system is a process that assists organizations in ensuring that their internal financial records are consistent with external records, such as bank statements or records from counterparties. The primary purpose of such a system is to identify discrepancies, errors, or anomalies that might arise due to various reasons like transactional errors or fraudulent activities

At Apollo Finvest, we made a particular system that does all the work of reconciling things, auto-magically 

This system has made an enormous difference – it reduced the work of our finance team by a whopping 90%!

In this blog post, we'll explain all the details about this cool project. We'll talk about the old process, its issues, how the new system works, how we put it into action, and how it made a big difference in our reconciliation system.

Old Process:

Our old operational procedure involved the following steps:

• Every month 

  • We acquired the repayment file from the Fintech Partner

  • The second step was to obtain payment received reports from the payment gateways

• Subsequently, we generated reconciliation reports by cross-referencing the payments received with the payments provided to us by fintech partners

Problems with the old process:

Our finance team had a tough time dealing with a lot of work with demanding timelines. Countless transactions needed to be recorded, reconciled, and verified each day, leading to a high risk of errors and inefficiencies. 

Especially due to the month-end nature of the process, the following issues existed as well:

• The Fintech Partner would be informed of the loan closure status at the end of the month for borrowers who had fully repaid their loan amounts

• For loan products that had revolving credit limits (similar to how Credit Cards work) when payments were uploaded at month-end, the remaining balance sometimes did not reflect real-time changes, potentially impacting the borrower limits

• There was a chance of errors in the manual reconciliation process

Based on all these potential issues, we recognized the need for a comprehensive solution that would not only automate these tasks but also integrate seamlessly with a range of Payment Gateways to receive transactions and process them automatically.

Our Solution: An Automated Reconciliation System

Let’s dive right into it!

Architecture and Design:

The automated Reconciliation system is designed as a robust and scalable microservices architecture. We used AWS Serverless Architecture to deploy the complete workflow.

A simple high-level architecture diagram to explain how it is built and used is shown below:

Payment Gateway Integration:

During our project, a key task involved integrating multiple Payment Gateways to receive transactions and update the amortization table accordingly. To achieve this, we developed specialized adapters for each gateway, enabling efficient communication between our system and various payment gateways, including Razorpay, Cashfree, and others.

Implementation:

• We have integrated through the payment gateway webhook to listen to different events related to payment happening across the platform.

• When we have a successful payment made on the platform after processing the event, we store it and update the amortization table.

• In the case of a  refund event through the payment gateway, we update the payment details and deallocate it from the loan ID.

This strategic approach provided a high degree of flexibility for our fintech partners, granting them the freedom to select their preferred payment gateway and also allowing them to get a real-time loan status

Automation Algorithms:

Automation served as the core foundation of our system. We developed complex logic designed to automate a range of tasks:  

• Repayment calculations

• Real-time allocation to the amortization table

• Handling of different scenarios depending on the repayment type 

• These scenarios include part-prepayment, early loan closure, and seamless interest benefit passing to customers.

• The above process resulted in reduced manual errors and ensured swift, accurate repayment processing.

The Implementation Journey

Building the Automatic Reconciliation System was a multi-faceted project that required meticulous planning and execution. Here are some highlights of our implementation journey:

Agile Development: We adopted an agile development methodology, allowing us to respond swiftly to changing requirements and prioritize features based on their impact on reducing the finance team's workload 

Development Process: The complexity of the repayment process lies in its dual nature, as it can either be executed manually or through automation, depending on the specific operational model of the Fintech Partner. To cater to both scenarios effectively, our development team embarked on creating an adaptable system that could seamlessly handle repayments from either approach.

We chose the AWS Serverless architecture to host our complete application, benefiting from its excellent scalability that aligned seamlessly with our requirements. Our approach involved the utilization of AWS Batch, Lambda functions, and SQS, carefully optimizing their usage to align with our distinct needs.

We used webhooks to ensure the integration of payment gateways was smooth and responsive. This real-time event-driven architecture not only allowed us to receive and process payments promptly but also granted us the agility to adapt to ever-evolving business requirements seamlessly.

The process to make the magic happen:

1. Payment Collection: We receive payments from various payment gateway portals. These payments are logged and stored in our database for record-keeping.

2. Partner Interaction: Partners interact with our system using our Repayment API. When they do so, we have a few important checks in place:

   1. Data Validation: We carefully validate the data sent by the partner to ensure it's accurate and complete.

   2. Payment Confirmation: We cross-verify whether the payment associated with this interaction has indeed been received in our payment gateway system. To do this, we integrate with multiple payment gateway APIs.

3. Communication with Fintech Partners: Once all the necessary checks are completed, we promptly update the Fintech Partner with the outcome of their repayment request. We achieve this by using a webhook integration, which allows us to send real-time responses to our partners.

4. Loan Table Update: Assuming all validation checks pass and the repayment is accepted, we update our loan table with the data. This ensures that the payment information is accurately reflected in our records.

5. Refunds Handling: If a refund is initiated directly from the payment gateway portal, it directly impacts our loan table. This means that refunds are seamlessly integrated into our system, keeping it up to date.

   The primary trigger for initiating a refund is when a dispute arises from the borrower's side, and the objective is to resolve the matter in favor of the borrower. Alternatively, a refund may also be initiated when the borrower makes an extra payment during the loan closure process and subsequently requests the return of the surplus amount.

6. End-of-Day Settlement: At the close of each day, we perform a crucial task - settling all the transactions. We achieve this by invoking the Settlement API provided by the respective payment gateway. This process is essential for ensuring that all financial transactions are properly balanced and aids in the later reconciliation process.

In essence, our system was meticulously engineered to harmonize the diverse needs of the fintech partner, encompassing both manual and automated repayment processes.

Testing and Validation: Rigorous testing was conducted at every stage of development. We performed unit testing, integration testing, and user acceptance testing to identify and rectify any issues promptly.

The Impact: A 90% Reduction in Workload

The implementation of our Automated Repayment System brought about a transformational change in the way repayments were handled at Apollo. The results were nothing short of remarkable:

Efficiency Gains: Repayments that once consumed hours of manual effort were processed in mere minutes, leading to a 90% reduction in the finance team's workload

Accuracy: Automation virtually eliminated the possibility of human errors in repayment calculations and transaction processing

Fintech Partner Satisfaction: Partners valued the freedom to select their preferred Payment Gateway and the ability to update their lending books in real-time, which significantly enhanced their overall experience.

If this sounds interesting and you would like to participate in similar adventures, we’re hiring, please check out our open engineering roles.

Credit Bureau APIs

Apollo has integration with different credit bureaus to get the borrower's credit score and credit history.

• Apollo Credit API 1

• Apollo Credit API 2

Loan APIs & Webhooks

The below-mentioned APIs are used for creating, updating and knowing the status of the loan

Loan Creation API:

• This API is used to create a loan in Apollo’s Loan Management System. We support different types of loans with different repayment frequencies:

  • Personal Loan (monthly/weekly/daily repayment)

  • Micro Loan

  • Consumer Loan

Loan Status API:

• Integrating with this API allows Fintechs to retrieve the current status of a loan within Apollo's Loan Management System. It provides real-time information for efficient tracking of the loan application.

Loan Status Webhook:

• To receive automatic updates on loan status changes via webhook, eliminating the need for continuous API calls.

Loan Update API:

• Loan Update API is used to update limited information about the loan that was previously created on Apollo's LMS

Disbursement Communication APIs & Webhooks

These APIs/webhooks are required to communicate the disbursement status between Apollo and Fintechs/Partner NBFCs

When disbursement is done by Apollo:

• Disbursement Status Read API

  • Integrating with this API allows Fintechs to retrieve the current disbursement status of loan/drawdown.

• Disbursement Status Webhook

  • To receive automatic updates on disbursement status changes via webhook, eliminating the need for continuous API calls.

When disbursement is done by Partner NBFC:

• Disbursement Status Update API:

  • To inform Apollo about the disbursement status of the loan

Repayment API & Webhook

Repayment API

• This API is needed to inform Apollo about the repayment collected from the borrower

Repayment Status Webhook

• This webhook notifies the fintech/partner NBFCs about the change in the status of the repayment post reconciliation that was previously created using the Repayment API

Other APIS

Borrower’s Status API

• This API is used to check whether the borrower belongs to the negative list (defaulter or fraud) of Apollo

NOC API

• This API is to generate NOC (No Objection Certificate) from Apollo Finvest for loans that are closed on Apollo’s LMS

Document Upload API

• This API is used to upload documents to Apollo’s S3 bucket

  ---

CKYC APIs

With our cKYC APIs, Fintechs and partner NBFCs can search, download, upload and update borrower's cKYC information. APIs available:

• cKYC Search API

• cKYC Download API

• cKYC Upload API

• cKYC Update API.

Please email us at business@apollofinvest.com to request a demo and complete documentation.

Now let’s talk about Apollo APIs. We are firing up our APIs and making them better. Much better. It’s like SpaceX improving their Falcon 9 engine🚀

We have many APIs which are powering the journey for our borrowers whenever they interact with our partner’s digital lending apps🏦. Offering a swift experience during the whole lending journey is the key to creating a great user experience🌞

Lets talk numbers

Our Loan Creation and Sanction Letter API have been optimized to be fast. Blazing fast!

• Previously the Loan Creation API took 700 milli-seconds to hit and process a request. Now, this takes less than 250 milliseconds. This is almost a 3X improvement! To be precise, it is equal to the blink of an eye😉

• Now let’s talk about the sanction letter generation API. This used to take 2 seconds. Now it takes only 500 milliseconds. This is a 4X improvement!  

• With these 2 improvements in place, an average loan gets sanctioned within 1 second at Apollo⚡️ 

The need for speed is real and a key metric for higher conversations when a borrower is applying for a digital loan!

How did we do it?

We got no superpower!🎈 There ain’t no Batman or Robin here! 

But what we got is a team who is always trying to improve existing services to offer better experiences😎 

They reviewed the code, removed duplicates, wrote new code, and optimized, optimized, and optimized! 

Sometimes improvements are better than creating a new feature🤘

Like the OG Steve Jobs said, sometimes we have to look back to move forward and build great experiences 

As always, there are more improvements coming. 

Until then why don’t you wonder how fast your eyes are blinking and our APIs are working?🤞

Here’s a collection of APIs that are fundamental to digital lending on the Apollo Platform. This is the rock-solid foundation on which 40+ companies have embarked on their lending journey 🌈

Start your engines. You are now 48 hours away from doing your first digital loan! 🚀

Overview 🤓

All our APIs are organized around REST*. Our APIs have predictable resource-oriented URLs, accept form-encoded** request bodies, return JSON encoded** responses, and use standard HTTP response codes.

If the tech jargon is lost on you:

• *Here is a quick explainer video on REST APIs

• **Think of form-encoded as an acceptable format to post a request in.

• **JSON ( JavaScript Object Notation) is a popular programming language

It’s simple to integrate, brilliantly documented, and most importantly suited to match your tech team’s ever-limited bandwidth

For ease of testing, we have distinct test and live environments. Switching over from test to live is easy: Simply change the URL and the unique key allocated by Apollo Finvest to production credentials and boom, you're live!

Documentation 📜

To make this Day 1 guide crisp, we have hand-picked a few core APIs out of our larger bouquet of 20+ APIs and split them into 5 broad categories as mentioned below.

• KYC APIs: 

  • We have integrations with leading players in the KYC automation space (Karza & HyperVerge). Their entire gamut of services are available with us 

• Credit APIs:

  • Credit 1 API documentation, Credit 2 API documentation

  • The credit API can be chosen depending on the underlying raw credit bureau data on which the credit score is built

• Loan Creation API:

  • Loan Creation API documentation

  • This API is used to pass approved borrower personal data, loan data,  KYC documents, and Loan Agreement through a single API to create loans on Sonic 

• Disbursement API:

  • Disbursement API documentation

  • We have made disbursement updates proactive and efficient. To know the status of the disbursements you can use our Webhook API

• Collection APIs: 

  • We have integrations with multiple payment gateways like RazorPay, Cashfree, PayU, etc. to enable different modes of collections 

All our API Responses are communicated via HTTP using following the status codes :

• 200 - OK 

  • Everything worked as expected

• 400 - Bad Request 

  • The request was unacceptable, often due to missing a required parameter

• 401 - Unauthorized 

  • No valid API key provided

• 402 - Request Failed

  • The parameters were valid but the request failed

• 403 - Forbidden 

  • The API key doesn't have permissions to perform the request

• 404 - Not Found 

  • The requested resource doesn't exist

• 409 - Conflict 

  • The request conflicts with another request (perhaps due to using the same idempotent key

• 429 - Too Many Requests 

  • Too many requests hit the API too quickly. We recommend an exponential backoff of your requests

• 500, 502, 503, 504 - Server Errors 

  • Something went wrong at our end- Very Rare!

 How can Fintechs get started with testing? 🚀

• Write to us at product@apollofinvest.com

• We will share the postman collection containing these APIs for a quick trial 

• Let the magic begin! ✨

We are upgrading our Credit API to a newer, faster version!

What is the benefit?🌈

• This API now enables Fintechs to place concurrent credit requests. Simply put, Fintechs no longer need to queue requests in a sequence when processing more than 1 case at a time

• Imagine this upgrade: 4X faster speed per API call + Ability to handle concurrent API calls. This is built to scale ⚡️

• There is no other change in the request or response body- it only got more efficient and quicker!

How can Fintechs start using this? 🚀

• Write to us at product@apollofinvest.com

• We will share your unique credentials along with the updated API URL and voila you will be upgraded!

We have a couple of credit APIs available basis their underlying raw data. You can find their documentation here for more details: Credit 1, Credit 2

• A webhook API for communicating the disbursement status of loans to Fintechs

What is the benefit? 🌈

• This API proactively updates the status of loan disbursements to Fintechs without needing them to place an API request

• This enables Loan disbursement Status to be shared with Fintechs on a Real-time basis thus making the process even more seamless and efficient 🤯

How can Fintechs start using this? ⚡️

• Share the webhook URL on which they would like to receive the updates

• Apollo will configure the same at our backend

• Once the configuration is complete, this is what a sample response pushed to the webhook would look like:

  

Feel free to reach out to us at product@apollofinvest.com for any queries.