AI-led Support Automation
- Status
- Field report v1
- Timeline
- Two quarters to implement; ongoing expansion to more issue types
- Case type
- Service experience + automation
- Role
- Head of Product Design
- Team
- Design Head + Product Head, 3 PMs, data team, AI team, customer support team
- 30-50%
- Inbound queries automated
- 40-60%
- Response-time reduction
- 12%
- Support-cost reduction
- 22%
- Bot resolution rate
- 5 intents
- Freeze, charges, transaction failure, fraud, MPIN/password
Support scale was growing faster than human support could.
As transactions, onboarding, and customer volume grew, support demand grew with it.
Every spike in call volume created the same problem: longer waits, higher call-center cost, more pressure to keep hiring agents, and a worse experience for the customer.
The issue was not always that a digital solution did not exist. In many cases, the solution already existed inside the app. The experience problem was discovery, routing, and timing.
Users came through IVR, chatbot, or email describing symptoms like "transaction failed" or "unable to transact," while the real root cause could be something else, such as account freeze.
From support contact to predicted intent.
The goal was not to replace support with bots.
The goal was to understand why customers were reaching out, predict the likely issue using behavioral and profile-based signals, and route them to the fastest digital resolution path.
Instead of treating every contact as a fresh support request, the system used issue signals to decide whether the customer should go to IVR, chatbot, email response, SMS deep link, or an app self-serve journey.
The system covered:
- Chatbot workflows
- IVR routing
- Email bot responses
- SMS deep links
- Proactive app surfaces
- App self-serve journeys
- Issue-intent mapping
- Lifecycle-triggered communication
- Tier-1 bot workflows for high-frequency issues
Primary automated intents:
- Account freeze
- Charges
- Transaction failure
- Fraud
- Forgot MPIN / password
The Issue Prediction API acted as the intelligence layer behind support routing.
It used behavioral and profile-based data from a central data lake to infer likely customer issues with reasonable accuracy.
- Transaction success rates
- Load times
- Banking-point visits
- Account status
- Recent registration
- Recent charges
- Transaction failure data
- Freeze status
- Other behavioral and profile indicators
This allowed IVR, chatbot, email, and app surfaces to route customers toward the right resolution path before a human agent had to investigate the issue from scratch.
Start with Pareto, not bot coverage
The team identified the few high-frequency intents driving a large share of support volume, instead of trying to automate everything at once.
Digitize the resolution journey
Automation only worked when there was a real self-serve path behind it. High-volume issues needed digital journeys, not just bot replies.
Treat IVR as a routing layer
IVR was redesigned from a generic menu into a smarter routing surface that could acknowledge known issues and guide users to app-based resolution.
Reduce human routing for high-confidence issues
For high-confidence blocker cases, the system first guided customers to the relevant self-serve journey instead of sending them through generic IVR or agent queues.
Use SMS as a bridge to self-serve
When users came through IVR, SMS deep links helped move them from phone support into the correct app journey.
Proactively surface help in the app
If signals suggested a likely issue, chatbot or help entry points could be surfaced on relevant app screens or the homepage for critical scenarios.
- Customer faces issue
- Calls IVR / opens chatbot / sends email
- Generic menu or generic bot
- Customer explains symptom
- Issue may be tagged incorrectly
- Human agent investigates root cause
- Resolution delayed
- Customer reaches IVR / chatbot / email / app
- Issue Prediction API reads behavioral/profile signals
- Likely intent is predicted
- Customer routed to relevant self-serve path
- SMS/app deep link opens resolution journey
- Human escalation reserved for uncertain cases
Account freeze
If freeze status was detected, IVR could acknowledge the issue and send an SMS link to the relevant app journey. A freeze API then identified the freeze type and routed the customer to the right action.
Charges
Charge-related queries were routed to Transparent Banking, where users could see past charges, upcoming charges, and the exact charge logic behind deductions.
Transaction failure
Users were shown refund timelines, failure context, and whether the issue needed to be resolved through the receiver-side bank.
Forgot MPIN / password
Handled through a fully self-serve reset journey, reducing the need for human support.
IVR acknowledgement
“We noticed your account is frozen. We have sent you an SMS to resolve this in the app. If you are calling for something else, continue.”
- Primary automation
- 30-50% of inbound queries automated through tier-1 bot workflows.
- Time compression
- 40-60% response-time reduction through real-time conversational and self-serve routing.
- Cost impact
- 12% reduction in support costs through call-center deflection.
- Resolution
- 22% bot resolution rate.
The goal was to protect wait time and CSAT as business volume scaled, without linearly increasing call-center headcount and cost.
As Head of Product Design, I worked directly with the Product Head from an experience strategy point of view.
This was not a UI project with dedicated designers. It was a service-experience redesign across support channels.
- Framing support automation as a customer-experience problem
- Identifying why users were contacting support
- Helping map high-volume intents
- Influencing the move from human-first support to self-serve routing
- Shaping the high-confidence blocker logic
- Working with PMs, data, AI, and customer support teams
- Ensuring automation reduced cost without abandoning the customer
A bot is not the product. The resolution path behind it is the product.
Customers describe symptoms. Systems need to infer root causes.
IVR can be more than a menu. It can become an intent-aware routing layer.
Human support should be reserved for uncertainty, not used as the default crutch.
Automation improves experience only when it gets users to the right resolution faster.