MsgHub AI Suite

One platform. Eight AI superpowers.

MsgHub doesn't bolt AI on as a feature — AI is woven through every screen. Ask your analytics a question. Let the agent run your workspace. Train intents with a few examples. Test two flows and let the winner take over. Rescue customers before they churn. All in plain English, no data scientist required.

The AI suite, at a glance

Jump to any feature — each one works on its own, but they're stronger together.

01 · AGENT

MsgHub Agent — the AI teammate that actually does the work

Most AI copilots only talk. MsgHub Agent takes action inside your workspace. You ask in plain English — "Send a reminder to everyone who abandoned cart yesterday" — and the agent looks up the segment, drafts a WhatsApp message, confirms with you once, and sends it.

It can create segments, build chatbot flows, launch campaigns, remember customer preferences across sessions, tag conversations, and escalate edge cases to you. Every destructive action needs a one-tap confirmation — so you stay in control.

What you can ask it

  • "Create a segment of customers who ordered biryani twice in the last month."
  • "Launch a 15% off WhatsApp campaign to cart abandoners from yesterday."
  • "Build me a chatbot flow that collects name, phone, and booking day, then hands off."
  • "Remember that Priya prefers morning appointments."
  • "What's my unanswered-question rate this week?"

MsgHub Agent

Online · Connected to your workspace

Thinking
Send 15% off to cart abandoners from yesterday
Found 187 customers who abandoned cart on Apr 17. Drafting WhatsApp template now…

📄 DRAFT TEMPLATE

Hey {{name}}, you left something in your cart 🛒 — here's 15% off if you finish your order today. Shop: msghub.co/r/xyz

Recipients

187

Est. cost

₹93

Channel

WhatsApp

How different businesses use the Agent

🏥 MEDICAL

"Remind patients with appointments tomorrow & list no-shows from last week."

🍽️ RESTAURANT

"Push today's lunch menu to customers who ordered in the last 30 days."

🛒 E-COMMERCE

"Find customers who bought shoes but not socks, send a bundle offer."

📈 LEAD-GEN (IndiaMART)

"Auto-reply to every new IndiaMART lead within 60 seconds on WhatsApp."

🎯 GOOGLE / META ADS

"Qualify every click-to-WhatsApp lead by asking 3 questions, then tag hot."

Ask Analytics — Plain English to Answer

"Which channel had the highest conversion last month?"

Answer

WhatsApp led with a 4.8% conversion rate across 2,340 campaigns. SMS was 2.1%, Email was 1.4%.

WhatsApp 4.8%
SMS 2.1%
Email 1.4%

Tip: WhatsApp outperforms when you include a product image + CTA button.

02 · ASK ANALYTICS

Type a question. Get an answer.

Your analytics shouldn't require you to learn SQL, open ten filters, or wait for a data analyst on Monday. MsgHub's Ask Analytics turns any plain-English question into the right chart, number, or list — in seconds.

It understands messaging-specific metrics: delivery rate, open rate, funnel drop-off, CSAT, cost per lead. It connects the dots across channels and returns a short explanation, not just a dashboard.

Ask anything like

  • "What was my cost per conversation this month?"
  • "Which WhatsApp template has the highest reply rate?"
  • "Show me IndiaMART leads that haven't replied in 7 days."
  • "Compare click-through rate for Meta Ads leads vs organic contacts."

Why each business loves it

🏥 MEDICAL

"Which clinic branch has the most no-shows this month?"

🍽️ RESTAURANT

"What's the most reordered dish by my repeat customers?"

🛒 E-COMMERCE

"Top 10 cart-abandon reasons in customer chat last week."

📈 INDIAMART

"Lead-to-WhatsApp-reply rate by product category."

🎯 GOOGLE / META ADS

"Cost per qualified lead split by ad campaign."

03 · CONVERSATION INTELLIGENCE

Read between the lines of every conversation

You have thousands of chats every month. Nobody reads them. MsgHub's AI does — and it tells you what matters: which questions your bot couldn't answer, which customers are angry, where people drop off, and what the top topics are this week.

It turns raw conversation logs into CSAT scores, sentiment heatmaps, funnel drop-off alerts, and a weekly "things to fix" list — without reading a single chat yourself.

What it surfaces

  • Sentiment — positive / neutral / negative, per conversation & customer.
  • Unanswered questions — the top 10 things your bot couldn't answer, clustered.
  • Funnel drop-off — where people abandon flows (and why).
  • Topic heatmap — what customers are asking about this week vs last.
  • Escalation signals — chats that need a human, ranked by urgency.

Conversations this week

Apr 11–17

CSAT

4.6/5

▲ 0.3

Resolved by AI

81%

▲ 4%

Avg. reply

3.2s

— stable

Trending topics

Order status
412
Refund
289
Menu / pricing
231
Technical issue
145

Bot couldn't answer: "GST on gold jewellery"

Asked 34 times this week. Add a KB article to auto-answer.

By business type

🏥 MEDICAL

Spot anxious patients early; flag chats mentioning "severe pain" for same-day callback.

🍽️ RESTAURANT

Detect cold-food / late-delivery complaints by branch, so you catch service issues fast.

🛒 E-COMMERCE

See the top 5 reasons people ask for refund — fix them at the source.

📈 INDIAMART

Tag leads by buying intent (just browsing / shortlisting / ready) automatically.

🎯 GOOGLE / META ADS

Find which ad creative brings in the angriest / least-qualified leads and pause it.

AI rewrote your system prompt

❌ BEFORE — vague & generic

"You are a helpful assistant for our shop. Answer customer questions politely."

✅ AFTER — focused & measurable

"You are Kavya, the WhatsApp assistant for Shree Jewellery. Answer in simple Hindi or English. For any pricing question, use the price_lookup tool. For repair / order status, use order_lookup. Never invent prices. If you're unsure, say: ‘Let me check with the team.’ and tag @escalate."

Expected lift: +18% resolution rate, −22% hallucinations

Based on 340 simulated conversations from your KB.

04 · PROMPT SUGGESTION

The AI that improves your AI

Writing a good system prompt is a skill. Most people guess. MsgHub analyzes how your bot actually performs — which questions went unanswered, where it made things up, where it sounded off-brand — and rewrites the prompt to fix those exact problems.

You get a before/after, an explanation of every change, and an expected performance lift. Apply with one click, or tweak and ship.

What it changes for you

  • Adds brand voice, language, persona — so every answer sounds like you.
  • Wires in the right tools (price lookup, booking, CRM) based on your flow.
  • Adds guardrails against hallucination (“never invent prices”).
  • Ships clear escalation phrases so handoff feels human.

How each business benefits

🏥 MEDICAL

Adds medical-safety guardrails: never diagnose, always route emergencies to a clinician.

🍽️ RESTAURANT

Teaches the bot your tone — warm, punny, regional — and upsells combos naturally.

🛒 E-COMMERCE

Forces the bot to only quote live prices from your catalog — never guess stock.

📈 INDIAMART

Rewrites openers so every IndiaMART lead gets a contextual reply, not a template.

🎯 GOOGLE / META ADS

Tunes the qualifying questions based on your ad copy, so click-to-chat converts more.

05 · INTENT TRAINING

Teach the bot your customer's language

Every business has its own slang. "Order status", "tracking ID", "mera saman kahan hai", "how far is my package" — they all mean the same thing. Intent Training lets you group these into one intent ("track order") with just a few examples. The AI handles every variation after that.

No ML engineers. No regex. You type 5 example messages, pick an action, and the classifier learns. It auto- improves from real chats — and tells you when customers are saying something new you haven't covered yet.

What you can do

  • Define custom intents: track order, book slot, complain, cancel, refer friend…
  • Multilingual out of the box — Hindi, English, Gujarati, Marathi, mix.
  • Route each intent to its own flow, agent, or tool.
  • Continuous learning: AI suggests new examples from real conversations weekly.

Intent: track_order

Active

Examples you added

"where is my order?"
"tracking id please"
"mera saman kab aayega"
"order kahan hai"

✨ AI suggests adding

"delivery status?"
"how far is my package"
ROUTES TO 📦 order_lookup flow confidence ≥ 0.82

Real intents for real businesses

🏥 MEDICAL

book_appointment, report_symptom, get_prescription

🍽️ RESTAURANT

place_order, reserve_table, complaint_food

🛒 E-COMMERCE

track_order, request_refund, size_exchange

📈 INDIAMART

quote_request, bulk_enquiry, sample_request

🎯 GOOGLE / META ADS

price_qualifier, location_check, book_demo

Running experiment

2,340 chats
A

Current flow

"Hi, how can I help today?"

Conversion 12.4%
CSAT 4.1
B 🏆 WINNER

AI variant

"Hey 👋 looking for {{recent_category}} or something new?"

Conversion 18.7% ▲
CSAT 4.6 ▲

Auto-promote Variant B — 95% confidence

Expected +₹1.2L additional monthly revenue.

06 · FLOW A/B TEST

Two flows go in. The winner takes over.

Every copy change, every button order, every opening line matters. MsgHub splits traffic between two flow variants automatically, tracks conversion + CSAT + drop-off, and promotes the winner once it's statistically significant.

You don't set sample sizes or p-values. You pick A, pick B, set a goal (conversion / reply rate / CSAT / revenue). MsgHub runs the experiment and tells you when you have a real winner.

What to test

  • Greeting copy: generic vs personalized (“looking for sneakers again?”).
  • Button menu vs free-form AI chat.
  • Qualification order: ask budget first, or timeline first?
  • Handoff point: after 2 failed bot answers, or offer human from turn 1.

What to test per industry

🏥 MEDICAL

Booking-first flow vs symptom-first flow — which books more consultations?

🍽️ RESTAURANT

Menu buttons vs "what are you craving?" — which drives larger orders?

🛒 E-COMMERCE

Product recommendation carousel vs one-at-a-time story. Which converts?

📈 INDIAMART

Template reply vs AI reply — which converts more enquiries to hot leads?

🎯 GOOGLE / META ADS

Qualifying-questions flow vs book-a-demo flow. Which fills more slots?

07 · CHURN RESCUE

Rescue customers before they're gone

Losing a customer quietly is the most expensive thing a business does. MsgHub's churn model watches each contact's activity (last order, last open, reply rate, sentiment) and flags the ones slipping away — before they stop responding.

Flagged contacts land in a rescue queue. AI picks the right message, channel, and timing, sends it, and only brings the ones who reply back to your team. Rescue rates typically beat static reactivation campaigns by 2-3×.

How it works

  1. Score — every contact gets a 0-100 churn risk updated daily.
  2. Rank — high-risk + high-value contacts bubble to the top.
  3. Rescue — AI drafts a personalized win-back, you approve, it sends.
  4. Learn — every reply (or silence) trains the model for your business.

🚨 High-risk customers this week

47 flagged
PR

Priya Rathi

Last order 42 days ago · Was monthly buyer

92
RM

Ravi Mehta

Stopped opening WA campaigns 3 weeks ago

78
AK

Arjun Kumar

CSAT dropped to 2.3 on last chat

71

✨ Suggested rescue for Priya

"Hey Priya, we noticed you haven't ordered in a bit — and we miss you! Your favorite maroon kurta just got restocked. Enjoy 20% off on your next order: msghub.co/r/priya20"

What churn looks like in your business

🏥 MEDICAL

Chronic patient missed 2 follow-ups — send an AI-drafted check-in.

🍽️ RESTAURANT

Weekly customer hasn't ordered in 3 weeks — push their usual combo with a nudge.

🛒 E-COMMERCE

Subscriber stopped opening emails — switch channel to WhatsApp with a personalized pick.

📈 INDIAMART

Quoted buyer went silent — auto-follow-up with a revised sample pricing.

🎯 GOOGLE / META ADS

Paid lead never showed up to demo — AI sends a recovery message with a new slot.

Every message, routed in real-time

"Hi" — quick greeting

Haiku

"Show me cart abandoners who spend > ₹5000"

Sonnet

"Build a churn prediction flow with 3 branches"

Opus

Haiku

₹0.02

fast & cheap

Sonnet

₹0.18

balanced

Opus

₹1.40

heavy reasoning

You're saving ₹1.2L/month

vs running every chat through the premium model.

08 · SMART ROUTING

The right AI model. For the right message.

Not every message needs the biggest, smartest, most expensive model. A simple "hi" shouldn't cost the same as a multi-tool agent call. Smart Routing analyzes each incoming message, classifies the difficulty, and sends it to the cheapest model that can handle it well.

Four providers (Claude, GPT, Gemini, OpenRouter), three tiers (fast / balanced / reasoning), one brain that picks for you. If one provider goes down, it automatically fails over to another. You set a monthly budget — Smart Routing stays under it.

What it saves you

  • Up to 80% lower AI costs by sending greetings to the fast tier.
  • Built-in failover — if Claude is down, it picks OpenAI seamlessly.
  • Prompt cache with a 1-hour TTL so repeat prompts are ~10× cheaper.
  • Per-tenant budgets — cap monthly spend, no surprise bills.

What it means for you

🏥 MEDICAL

FAQs on clinic hours go to fast tier. Symptom triage routes to reasoning tier for safety.

🍽️ RESTAURANT

Menu & hours: fast tier. Custom catering quote: reasoning tier. Savings: huge.

🛒 E-COMMERCE

Order status: fast. Complaint resolution: balanced. Custom bundling advice: reasoning.

📈 INDIAMART

Volume enquiries go cheap. Tax / compliance / export queries auto-route to reasoning.

🎯 GOOGLE / META ADS

Every paid click gets a reply in ≤3s — without burning your AI budget on small talk.

Each feature is great alone.
Together, they compound.

Conversation Intelligence spots unanswered questions. Prompt Suggestion rewrites your bot to answer them. Flow A/B tests the new version. Ask Analytics reports the win. Smart Routing keeps the cost low. Churn Rescue catches customers before they leave. The Agent runs the whole loop for you.

The loop, simplified

1

Listen

Conv. Intelligence surfaces gaps

2

Improve

Prompt Suggest + Intent Training

3

Test

Flow A/B auto-promotes winner

4

Rescue

Save slipping customers

5

Scale

Smart Routing keeps cost flat

Under The Hood

What happens in the
0.8 seconds after your customer hits send

Watch MsgHub's agent think, call a tool, verify safety, and reply — step by step. Every agent conversation, demystified.

Elapsed 000ms

What the customer sees

One simple reply. Five invisible steps.

What the AI actually does

1

Receive & sanitize

Message lands. Platform strips prompt-injection attempts and untrusted markup.

12ms
// incoming webhook
msg: "Where's my order #8821?"
from: "+91 98••• 12345" /* PII masked in logs */
✓ Prompt-injection scan ✓ HMAC verified ✓ Rate limit OK
2

Think & pick tool

Smart Routing picks the cheapest model that can answer. Agent reads intent and available tools.

320ms
model: "claude-haiku-4-5" // routed — fast tier
intent: "track_order" // confidence 0.94
tools_available: [lookup_order, create_ticket, escalate_human]
decision: "call lookup_order"
⚡ Tier: fast 🧠 Cache hit · 10× cheaper
3

Call lookup_order()

Tool args validated with Zod. Query is tenant-scoped — no cross-customer leakage possible.

340ms
lookup_order({
  order_id: "8821",
  tenant_id: "tnt_acme"  // injected, not from AI
})
✓ Zod type-check ✓ Tenant-scoped ✓ RLS enforced
4

Tool result

Your database replies in real time — live, not cached, not hallucinated.

520ms
{
  status: "in_transit",
  carrier: "BlueDart",
  awb: "8821",
  eta: "2026-04-19T15:00:00Z",
  track_url: "bluedart.in/AWB8821"
}
✓ Live DB data ✓ Tenant match
5

Compose & send reply

AI writes a natural reply using the live data. Final PII scan, then WhatsApp delivers.

820ms
// final output to channel
"Hi Priya! 📦 Your order #8821 is in transit with
BlueDart — ETA tomorrow by 3pm. Track live: …"
✓ No PII leak ✓ No hallucination (grounded in tool result) 🚀 Delivered

Every agent conversation in MsgHub runs through this exact loop. Every tool call is logged. Every destructive action needs confirmation. Every AI output is scanned before it reaches a customer. No black box.

Bring your business into the AI era.

No data team. No procurement cycle. Just one platform where AI actually moves the needle — from day one.