Commerce in the Channels Your Customers Actually Use

Complete the entire purchase — browse, add to cart, pay — without leaving WhatsApp.

Customers in India's tier 2 and tier 3 cities already run their lives through WhatsApp. X3 meets them there with a fully transactional shopping experience: product catalog browsing via interactive list messages, add-to-cart with quantity control, address selection from saved profiles, and UPI payment link delivery in-thread. MSG91 is the primary messaging provider for India, with automatic failover to Twilio's WhatsApp Business API for international tenants. Order confirmations arrive as structured messages with line items, totals, and a tracking link. Cart abandonment is recovered in-channel — a follow-up message with a deep link back into the interrupted session, sent 30 minutes after drop-off, recoverable without re-entering any data. The entire session is stateful: pause mid-checkout, resume the next morning.

• MSG91 primary integration for India (sub-3 second delivery SLA)
• Interactive list messages for catalog navigation
• UPI payment link generated per order, expires in 15 minutes
• Order confirmation + invoice PDF in-thread
• Cart abandonment recovery at configurable intervals (default 30 min)
• Full session state persists across days
• Tenant-scoped catalog: only your SKUs visible to your customers
• WhatsApp Business API verified sender for high deliverability

The same catalog, ordering, and support capability as WhatsApp — on Telegram, without duplicating any configuration.

Tenants register a Telegram bot token in the admin panel under Settings → Messaging Channels. X3 provisions the bot, connects it to the same NL interpreter and product catalog as the WhatsApp channel, and activates it within minutes — no separate catalog sync, no separate prompt writing. The bot supports inline keyboard navigation for browsing categories, a persistent cart across Telegram sessions, multi-language responses (English, Hindi, Arabic, Tamil — language auto-detected from the user's Telegram locale or overridden by explicit command), and the same order lifecycle events as WhatsApp. Conversation context is stored per Telegram user ID: a customer who asks about an order status in the morning and returns in the evening to place a reorder gets a coherent, continuous experience without re-identifying themselves. Support tickets raised in Telegram escalate to the same agent queue as WhatsApp.

• Single bot token registration in admin — no dev work
• Shared NL interpreter, catalog, and order state with WhatsApp
• Inline keyboard navigation for category and product browsing
• 4-language support: English, Hindi, Arabic, Tamil (auto-detect)
• Persistent session state across Telegram conversations
• Same escalation queue as WhatsApp for live agent handoff
• Webhook-based — no polling, sub-second message delivery
• Tenant-isolated: each tenant's bot sees only their data

One shared model interprets customer messages across every channel — fix a prompt once, improve every channel simultaneously.

The NL Intent Interpreter is the shared brain beneath WhatsApp, Telegram, and any future conversational surface. It classifies each incoming message into one of seven top-level intents: product search, order status, return request, loyalty query, reorder, support escalation, or out-of-scope. Within each intent, slot-filling extracts the required parameters — for order status, that means the order ID or the customer's phone number; for product search, it's category, price range, and any attribute filters. Multi-turn slot-filling means the bot asks only for what's missing: if a customer sends 'where is my kurta order from last week,' the interpreter resolves the product category and recency, looks up the matching order, and responds — no 'please provide your order number' prompt. Graceful fallback routes genuinely ambiguous messages to a confirmation prompt before escalating, so edge cases never produce confident wrong answers. Because the interpreter is channel-agnostic, a prompt refinement deployed by a Kynetra regent propagates to WhatsApp and Telegram simultaneously — there is no per-channel fine-tuning.

• 7 top-level intent classes covering full commerce lifecycle
• Multi-turn slot-filling: asks only for missing parameters
• Recency and attribute resolution from natural language ('last week', 'blue one')
• Confidence threshold gate: low-confidence responses escalate rather than hallucinate
• Channel-agnostic: one model update propagates everywhere
• Tenant-specific catalog and policy context injected at inference time
• Intent telemetry dashboard: see what customers are actually asking
• Kynetra regent CI deploys prompt changes without code releases

When the bot reaches its limit, a human agent takes over — with the full conversation context and customer profile pre-loaded.

Escalation is triggered automatically when confidence drops below threshold, when the customer explicitly asks for a human ('talk to someone', 'agent please'), or when the same issue appears in more than two consecutive bot turns. The agent sees a unified panel: the complete conversation transcript, the customer's order history, loyalty tier, previous support tickets, and any open returns — pulled in real time from the X3 data layer. No copy-pasting, no 'can you share your order number' when the bot already captured it. Agents can resolve the issue and optionally re-engage the bot for follow-up messaging (shipping notifications, return confirmation) so the human handles the high-touch moment and the bot handles the routine follow-through. Resolution is attributed: the ticket is marked as bot-assisted or agent-resolved, feeding into SLA reporting and bot quality metrics. Agents are managed in admin-web under Settings → Support Team; routing rules can assign escalations by channel, tenant, or issue type.

• Automatic escalation on low-confidence or repeated failure
• Keyword triggers: 'agent', 'human', 'complaint', customizable per tenant
• Agent panel: full transcript + order history + loyalty + open tickets
• Agents re-engage bot for post-resolution follow-up
• Bot-assisted vs agent-resolved attribution in SLA reports
• Routing rules: by channel, issue type, or agent group
• Queue visibility: agents see wait time and position
• All conversations stored under DPDP-compliant data retention policy

Every 'where is my order?' is handled by the bot — and proactive shipping events arrive in-channel before customers think to ask.

X3 normalizes carrier events from Delhivery, Bluedart, DTDC, Ecom Express, and Shiprocket into a unified event schema: picked up, in transit, out for delivery, delivered, return initiated. These events trigger proactive WhatsApp or Telegram messages without requiring the customer to ask — out for delivery at 8 AM means a message at 8 AM. The bot resolves order status queries from natural language: 'where is my order,' 'did my parcel arrive,' 'when will the saree reach' — no order number required if the customer's phone is registered. For multi-item orders, each shipment's status is reported separately. Delivery confirmation messages ask for a simple thumbs-up acknowledgment, which feeds into the delivery confirmation rate metric (used to flag suspected misdeliveries). Customers can also reschedule delivery or raise a not-received complaint entirely within the chat thread — no app download, no web form.

• Carrier normalization: Delhivery, Bluedart, DTDC, Ecom Express, Shiprocket
• Proactive notifications: out-for-delivery, delivered, return-in-transit
• Natural language order lookup — no order ID required if phone is registered
• Per-shipment status for multi-item orders
• Delivery confirmation thumbs-up feeds misdelivery detection
• In-channel rescheduling and not-received complaint
• Notification preferences: customers opt out per event type
• All events logged to order timeline in admin-web

Points balance, reward redemption, and tier upgrade notifications — all via WhatsApp, no app required.

The loyalty commands are woven into the same conversational flow as shopping and support: 'how many points do I have,' 'redeem my points on this order,' 'what's my tier.' The bot retrieves live data from the X3 Loyalty engine — no stale cached balance. Redemption within a WhatsApp checkout session applies the discount before the UPI payment link is generated, so the customer pays the net amount and sees the breakdown in the confirmation message. Tier upgrade notifications are proactive: when a customer crosses a tier threshold (Silver to Gold, for example), a rich WhatsApp message explains the new benefits — free shipping threshold, higher reward rate, birthday bonus — without requiring any action. Points expiry reminders are sent 7 days before expiry with a direct link into an active catalog session so the customer can use them immediately. Tenants configure the message templates and thresholds in admin-web under Loyalty → Notification Rules.

• Live balance lookup — not cached, always accurate
• Redemption in checkout: discount applied before payment link
• Tier upgrade rich message with benefit summary
• Points expiry reminder 7 days before with catalog deep link
• Earn notification after each qualifying order
• Customizable templates per tenant in admin-web
• All loyalty commands available in Telegram too
• Loyalty history accessible via conversational query