How Tumble Dry handles 100% of pickup-booking inbound calls with a voice AI agent and a deep-API MCP layer
Tumble Dry runs India's largest organized laundry chain — over 1,500 stores in 350+ cities. Their inbound call volume is dominated by customers asking to schedule pickups, check order status, or raise complaints. Caller Digital deployed a voice AI agent that handles all inbound booking and ticket-handling traffic, integrated to Tumble Dry's order/ticketing APIs through a Model Context Protocol (MCP) layer.
An inbound call queue that scales with store count — and a customer expectation that every call is answered fast, in-language, end-to-end
Tumble Dry's customer-experience surface is dominated by inbound voice. A customer wants to book a pickup, ask where their order is, raise a complaint about a stained garment, or reschedule a delivery — and they call a single national number. With 1,500+ stores feeding the same support layer, the call queue is large, bursty, and skews heavily toward booking and status questions rather than complex resolution.
Two things made this hard. First, the back-office logic for any given call — checking pickup-slot availability for the customer's pincode, looking up an order's status across the franchise network, raising a ticket against the right store, scheduling a re-pickup — required pulling data from multiple internal APIs in real time. Most contact centres can't do this; they read from a script, capture details, and hand off to back-office for follow-up. The customer is told to wait for a callback that may not come.
Second, language. Tumble Dry's footprint is genuinely pan-India — Tier-1 metros, Tier-2 capitals, and Tier-3 towns. A Patna customer wants Hindi; a Coimbatore customer wants Tamil; a Pune customer might switch between Marathi and Hindi mid-call. Building and staffing a multilingual contact centre at the scale Tumble Dry needs would require hundreds of agents and still wouldn't run 24×7 cleanly.
Queue depth peaking unpredictably
Pickup-booking demand spikes around weekends, festivals, and weather events. Hold times during peaks were eroding NPS and pushing customers to WhatsApp where resolution was slower.
Back-office integration on every call
Almost every call required real-time API access — slot availability, order status, ticket creation. Human agents either had to be trained on multiple internal tools or hand off to back-office, which broke first-call resolution.
Multilingual coverage with consistent quality
Even when regional-language agents were available, conversational quality varied. A customer who wanted Tamil sometimes got broken Hinglish.
Tickets being mis-routed or dropped
Complaints raised over voice often didn't make it cleanly into the ticketing system with full context (order ID, store ID, photos requested, customer expectation), forcing follow-up cycles.
A voice AI agent fronting the inbound queue, wired to Tumble Dry's APIs through an MCP layer that lets the agent take actual back-office actions in-conversation
The deployment has two layers. The voice agent itself runs the conversation — language detection, intent recognition, dialogue flow for the four dominant intents (book pickup, check order status, raise ticket, reschedule). Below the agent sits an MCP (Model Context Protocol) layer that exposes Tumble Dry's internal APIs to the agent as a controlled set of tools.
MCP turns the agent from a chatbot into an operator. Instead of asking the customer to wait for a callback, the agent calls the slot-availability API for the customer's pincode in-conversation, proposes available windows, and books the pickup before hanging up. For order status, the agent queries the order tracking API, reads back the current state in plain language, and offers next-step options. For complaints, the agent gathers structured details, raises a ticket against the right store with full context (order ID, customer ID, complaint category, severity, photos requested), and reads back the ticket number for the customer to track.
The MCP wrapper is the unlock. It enforces auth, rate-limits, audit logging and field validation between the LLM-driven agent and the production order/ticket systems — so the agent is genuinely transactional but Tumble Dry's data integrity stays intact.
End-to-end pickup booking
Agent checks slot availability for the customer's pincode in real time, proposes 2–3 windows, confirms address and payment mode, and books the pickup against the right store before ending the call.
Live order-status lookup
Customer asks 'where's my order' — agent queries the tracking API live, reads back the state, and offers next steps (reschedule, escalate, no action).
Structured ticket creation
Complaints captured as structured tickets with order ID, store ID, category, severity, customer expectation and any photo requirements — landing in the right CRM queue with the right SLA.
MCP-controlled tool access
API access is scoped, rate-limited, audited. The agent can only invoke tools it's been granted, with field-level validation between the conversation and Tumble Dry's production systems.
8-language conversational coverage
Hindi, Hinglish, Tamil, Telugu, Marathi, Bengali, Kannada, Gujarati. Code-switching native. 24×7 coverage with no shift gaps.
Smart escalation
Complex or sensitive calls (refund disputes, repeat complaints, high-value orders) escalate to a human agent with the full transcript, ticket, and customer context already in hand.
Inbound queue absorbed at scale, first-call resolution lifted on booking and status calls, complaint tickets land complete
Within the first deployment phase, the voice AI began handling the majority of pickup-booking and order-status calls end-to-end without escalation, freeing human agents for the longer-tail of complaint resolution and exception handling.
| Dimension | Before | After |
|---|---|---|
| Pickup booking flow | Capture details, callback to confirm slot | Slot booked in-call against live availability |
| Order-status calls | Agent reads from CRM, often stale | Live API lookup, current state read back |
| Ticket completeness | Free-text, often missing fields | Structured (order ID, store, category, severity) |
| Hold time at peak | Long during weekends/festivals | Parallel handling, no queue depth |
| After-hours coverage | Limited or none | 24×7 fully automated for booking & status |
| Language coverage | Hindi/English-led | Hindi, Hinglish, +5 regional |
Frequently Asked Questions
MCP (Model Context Protocol) is a standardized layer that exposes internal APIs to a language-model agent as controlled tools. For Tumble Dry, it lets the voice agent actually invoke production endpoints — slot availability, order tracking, ticket creation — with auth, rate-limiting, and field validation in between. The result is an agent that can resolve calls end-to-end rather than just collecting information for a callback.
Pickup booking, order-status lookup, simple reschedules, and structured complaint capture are handled end-to-end by the AI. Complex disputes, refund decisions, and high-sensitivity escalations are routed to human agents with the full transcript, captured ticket, and customer context — so the human starts at minute one of resolution rather than rebuilding context.
The MCP layer enforces three controls: scoped tool access (agent can only call APIs it's been explicitly granted), rate-limits per session, and field-level validation on inputs and outputs. Every tool invocation is audit-logged, and write operations (booking creation, ticket creation) require structured outputs that are validated before the API call is made.
Hindi, Hinglish, Tamil, Telugu, Marathi, Bengali, Kannada and Gujarati in production. Language is detected on the customer's first utterance and the agent code-switches mid-call when the customer mixes languages.
The agent escalates to a human queue with the full transcript, current ticket state, customer history, and any partially-completed action (e.g. a pickup tentatively held but not confirmed). The human agent picks up where the AI left off rather than restarting the conversation.
The voice agent itself was live in a few weeks. The MCP layer took the most engineering time, because it required wrapping Tumble Dry's order and ticketing APIs with consistent auth, validation, and audit logging — an investment that now pays off across every new agent intent we add.
Run a similar voice AI rollout for your team
We work with Indian enterprises across BFSI, D2C, retail SaaS, and consumer services. Book a 30-minute consult to see what voice AI can do for your call queue.
© 2025 Caller Digital | All Rights Reserved