WhatsApp AI Agent Playbook | Priyank Singh × AiSensy

Build your first WhatsApp AI Agent

One WhatsApp number that answers, sells, books and follows up.

Jump to: Core setup · Industry playbooks · Retention system · Testing & copy-paste kit

Two engines, one WhatsApp number

AiSensy's AI Agent reads your website, catalog and documents to help with questions, product searches, order checks, bookings, payments and complaints, with human handoff when needed.

The inbound agent: Anyone messaging you from a Click-to-WhatsApp ad, your website or a QR code can get an on-brand answer, recommendation and, where appropriate, a payment link or booking.

The retention system: Claude segments past customers and writes personalised messages. AiSensy sends approved broadcasts. When someone replies, the AI Agent uses purchase history and your Shopify catalog to upsell, cross-sell and send a payment link.

Use the Flow Builder for fixed paths such as buttons, lists and keyword triggers. Use the AI Agent for open-ended conversations. Both work in the same account.

Before you start

Setup takes minutes; preparation takes time. Gather:

A WhatsApp number on AiSensy's official WhatsApp Business API. If it is already on the WhatsApp Business app, ask AiSensy's onboarding team how to bring it over.
An up-to-date website, with old prices and expired offers removed.
FAQs, price lists, shipping, returns, refund policies and service menus. The playbook lists PDF, DOC, DOCX, TXT and CSV: up to 50 MB per file and 30 pages per PDF.
A Shopify or WooCommerce catalog, or a CSV/XLS with product names, prices, links and images.
Razorpay or PayU if the agent should send payment links.
Approved WhatsApp templates for messages you send first: broadcasts and follow-ups after 24 hours.
A human owner and clear trigger phrases for handoffs.
One to three test numbers. 500 free AI messages are for building and testing.
For retention: a customer export with phone numbers, limited to people who opted in to WhatsApp messages.

Your core setup, in 15 steps

Every business follows the same core setup. No coding is needed for the basic build; someone still needs to supply the knowledge, test it and switch it on.

Phase A · Quick start (about 10 minutes)

Open Agents studio: Agents → Set up.
Train from your website: paste the URL and a test WhatsApp number, with country code and digits only. Choose Train Agent. It starts on the test number while continuing to read the website.
Choose capabilities: FAQ/Support and Human Handoff are pre-selected. Continue setup, add relevant capabilities, then choose Go to agent.

Phase B · Who the agent is

Business Profile: review the description, agent/business name, currency and business type (E-commerce, Service, Real estate, Healthcare, Education or Travel). Regenerate from site if needed. An optional greeting is sent word-for-word; leave it blank for no fixed opener.
Ground Rules: set what the agent must never say, promise or discount in any chat. Put skill-specific instructions inside Skills.
Voice: choose Friendly, Professional, Concise or Playful; Short, Medium or Long replies. Short or Medium works well on WhatsApp. Add custom style notes under Refine the Voice.

Phase C · What the agent knows

Knowledge: crawl extra pages, upload files, paste policies, price lists or temporary offers, and add FAQs with exact answers. Keep one consistent source of truth: conflicting prices confuse the agent.
Images: review and uncheck anything the agent should not send.

Phase D · What the agent does

Skills: enable only relevant jobs: FAQ, Lead Qualification (capture, score and route prospects), Appointment Booking, Sales, Human Handoff and Feedback/CSAT. Sales can Transact, Book, Qualify & route, or use Hybrid. For another capability, choose Add Capabilities → Create Custom, describe when to use it, then Draft with AI for a first version.
Products: sync Shopify or WooCommerce, import CSV/XLS, or add products manually so recommendations use real items and prices.
Payments: connect Razorpay or PayU. Set an out-of-window template for customers who pay after the 24-hour WhatsApp window closes.
Actions and templates: apps from Integrations → App Store appear as Business Actions. Add Custom API Action accepts a cURL for your own system. Test the call and require confirmation for actions that change data. Enable only approved templates the agent may send.
Meta Ads context (optional): map each Click-to-WhatsApp ad to its own greeting and context. A Diwali combo ad should lead to a Diwali combo conversation.

Phase E · Test and go live

Settings and testing: configure fallback, active numbers, handoff trigger phrases and the handoff message. Train mode tests unsaved drafts; Live mode tests the published agent. Add up to three test numbers.
Publish and switch on: publish changes, then turn on the AI; it starts paused. Real-customer replies need a paid plan plus an AI Messages pack. Each agent reply counts as one AI message.

Build your WhatsApp AI Agent with AiSensy

Start with the core setup, adapt it to your industry and test before going live. Live customer replies need a paid plan plus an AI Messages pack.

Back to setup Get Started with AiSensy

Pick your industry playbook

The setup stays the same. Your business type, skills and knowledge make it yours.

E-commerce / D2C

Business type: E-commerce · Sales motion: Transact

Skills: FAQ, Sales, Feedback/CSAT and Handoff.
Feed it: Shopify/WooCommerce sync, size guides, shipping and returns policies, combo offers.
Hand off: Damaged orders, refund disputes and bulk orders.
Ground rule: Only recommend products in the synced catalog and in stock. Never invent discounts or codes.
Test: “bhai 2 lene pe kuch discount milega?”

B2B / manufacturing / wholesale

Business type: Service; E-commerce if selling from a catalog · Sales motion: Qualify & route

Skills: Lead Qualification, FAQ, custom “Bulk enquiry” and Handoff.
Feed it: Product specifications, MOQ, lead times, certifications and delivery regions.
Hand off: Custom pricing, quotes above a set value and credit terms.
Ground rule: Never quote a final price. Share the list price or range from Knowledge, collect quantity, city and timeline, then route to sales.
Test: “500 units chahiye Pune delivery, best rate?”

Agencies & consultants

Business type: Service · Sales motion: Qualify & route, or Hybrid with booking

Skills: Lead Qualification, Appointment Booking, FAQ and Handoff.
Feed it: Service pages, case studies, process and what you do not do.
Hand off: Confirmed budgets, proposal requests and unhappy clients.
Ground rule: Never promise results, timelines or prices absent from Knowledge. Qualify budget, goal and timeline before booking.
Test: “guarantee doge 3x leads?”

Real estate

Business type: Real estate · Sales motion: Hybrid, qualifying before a site visit

Skills: Lead Qualification, Appointment Booking and Handoff.
Feed it: Brochures, floor plans, location highlights and possession timelines.
Hand off: Price negotiations, loan questions and ready-to-book buyers.
Ground rule: Never commit to price, availability or approvals. Collect budget, configuration and visit date, then book.
Test: “3BHK under 1.2 Cr, kal visit ho sakta hai?”

Healthcare / clinics

Business type: Healthcare · Sales motion: Book

Skills: Appointment Booking, FAQ, Feedback/CSAT and Handoff.
Feed it: Doctor lists and timings, services, fees and pre-visit instructions.
Hand off: Symptom descriptions, emergency words and reports.
Ground rule: Never diagnose, suggest medicine or interpret reports. For symptoms, book a consultation or hand over.
Test: “sar mein dard hai kaunsi tablet lu?”

Education / coaching

Business type: Education · Sales motion: Hybrid, qualifying before counselling or a payment link

Skills: Lead Qualification, Appointment Booking, Sales and FAQ.
Feed it: Course pages, batch dates, curriculum PDFs and refund policies.
Hand off: Fee negotiations, scholarships and parent escalations.
Ground rule: Never promise placements, marks or outcomes. Share batch dates and fees only from Knowledge.
Test: “placement guaranteed hai na?”

Travel & hospitality

Business type: Travel · Sales motion: Hybrid, qualifying before booking or a deposit

Skills: Lead Qualification, Sales, Appointment Booking and FAQ.
Feed it: Itineraries, inclusions/exclusions, cancellation policies and room types.
Hand off: Group bookings, custom itineraries and complaints during a stay.
Ground rule: Never confirm a booking until payment or a human confirms it. Quote only listed packages.
Test: “Goa 3N4D for 4 people, cheapest kya hai?”

Local services · salon, gym, restaurant

Business type: Service · Sales motion: Book

Skills: Appointment Booking, FAQ and Feedback/CSAT.
Feed it: Menus or service lists with prices, timings, location and packages.
Hand off: Large parties, event bookings and complaints.
Ground rule: Only offer existing slots and prices. Hand over for parties above your set size.
Test: “Sunday 6 baje 2 log ke liye slot hai?”

Test in the languages your customers actually use, including messy Hinglish. The agent replies in the customer's language: Hindi, English, Hinglish and more.

The Claude retention system

Claude starts the campaign. Your AI Agent carries the conversation forward.

Claude decides who receives which offer and writes each message. AiSensy sends an approved WhatsApp broadcast. A customer reply opens the conversation for an AI Agent trained with purchase history, product recommendations and payment links.

The monthly loop: Export customer data → segment and personalise with Claude → build the upsell map → broadcast with AiSensy → AI Agent handles replies and payment → refresh next month.

1 · Export opted-in customer data

Combine Shopify customers and orders into one master CSV, one row per customer. Keep only customers who opted in to WhatsApp messages.

Required columns:

name — e.g. Riya Sharma
phone, with country code — required for WhatsApp
last_purchase_date — e.g. 2026-07-14
total_orders — e.g. 3
total_spend — e.g. 6400
last_product_bought — e.g. Whey Protein 1 kg
product_category — e.g. Protein

Email is for email campaigns. City and tags are optional; tag VIPs as vip.

2 · Segment and personalise in Claude

Upload the CSV and run your retention skill. The playbook names whatsapp-messenger for WhatsApp-only output, and retention-campaign for WhatsApp, email and creatives. These are custom skills referenced by the playbook; it does not supply installation links.

Provide your brand basics and select Upsell, Cross-sell, New Product Launch, Reactivation, Returning Customers or Festive Sale. Claude works out days since the last purchase and uses two tracks.

Track A — broadcasts: Festive Sale and New Product Launch go to every opted-in customer in the export.

Track B — one targeted campaign per customer. The source's first-matching-rule example is:

Tagged vip → Upsell.
Last purchase more than 90 days ago, or only one order ever → Reactivation.
At least 3 orders and total spend above ₹5,000 → Upsell.
Has a product category → Cross-sell.
Last purchase 30–90 days ago and more than one order → Returning Customers.

You receive whatsapp_messages.csv, with name, phone, campaign and a personalised message. Adjust thresholds to your business: inactive for skincare differs from inactive for furniture.

3 · Build the upsell map

Ask Claude to map each catalog product to a sensible bigger/better version and a complementary item, with a one-line reason and product URLs.

Example from the playbook:

Bought: Whey Protein 1 kg
Upsell: Whey Protein 2 kg
Cross-sell: Shaker + Creatine
Reason: Daily users run out in 30 days; creatine pairs with training.

Use only real catalog products. The copy-paste prompt below asks Claude to flag products without a sensible next buy.

4 · Send it through AiSensy

A message you send first must use an approved WhatsApp template. Personalised free text from Claude cannot be broadcast directly.

Create one marketing template per campaign. Map {{1}} to name, {{2}} to last product and {{3}} to recommended product. Add a quick-reply button such as “Show me” or “Tell me more”.
Ask Claude to add matching variable columns: name, last_product and recommended_product.
Import the CSV, map columns to variables and broadcast now, or schedule up to two months ahead.
A reply opens the 24-hour chat window for the AI Agent. AiSensy retargeting can reach people who did not respond.

5 · Train the agent for upsell and cross-sell

Catalog and payments: sync Shopify under Products, connect Razorpay or PayU, and set Sales to Transact.
Knowledge: upload the upsell-map CSV.
Purchase history: the source recommends checking Shopify order actions under Integrations → App Store, or adding a Custom API Action that fetches orders by the chatting customer's phone number. Use a verified, customer-scoped lookup.
Custom skill: create “Upsell & Cross-sell” under Add Capabilities → Create Custom. Use it for campaign replies, past-order mentions or requests for what to try next. Instructions follow in the copy-paste kit.
Feedback/CSAT: turn it on so the agent can ask how the last order went after payment.

6 · Repeat monthly

Export fresh orders, rerun Claude and broadcast the new segments. As purchase history changes, customers move into the matching campaigns. Every new order feeds the next run.

Build your WhatsApp AI Agent with AiSensy

Start with the core setup, adapt it to your industry and test before going live. Live customer replies need a paid plan plus an AI Messages pack.

Back to setup Get Started with AiSensy

Test it. Then make it yours.

A real-customer checklist and copy-paste kit for your agent and retention campaigns.

Test it like a real customer

Before going live, try to break it. A clear fallback and a human handoff matter more than a smooth but invented answer.

Ask the same FAQ in different ways.
Try “price?”, “size chart h kya”, misspellings and half-finished messages.
Test Hindi, Hinglish and a regional language.
Ask for a product, its link and how to pay. Check that the payment link opens.
Ask for an item outside the catalog. It should not invent one.
Negotiate: “thoda kam karo”. Check the ground rules hold.
Ask something the agent cannot answer. Check fallback and handoff.
Send an angry delivery complaint: “order abhi tak nahi aaya, bahut bekar service”. Check escalation.
Test each skill trigger — bulk orders, booking and lead forms — and required detail collection.
Reply to a test retention broadcast from a number with purchase history. Check the past product, upsell-map recommendation and payment link.
Test in Train mode while editing, then Live mode after publication.

Copy-paste kit

Replace bracketed details with your own. Keep claims and recommendations grounded in your business.

Ground rules · e-commerce + retention

Only recommend products that exist in the synced catalog and are in stock. Never invent products, prices, discounts or promo codes.
Only discuss the purchase history of the customer you are chatting with, matched by their phone number. Never reveal any other customer's name, number, orders or details.
Suggest at most two products per reply. If the customer says no, offer one cross-sell, then stop selling and help with whatever they ask.
Always share the product link and price before sending a payment link. Send the payment link only after the customer clearly confirms.
If you are not sure of an answer, say so and offer to connect them with the [Brand] team. Never guess.
Reply in the language the customer writes in.

Custom skill · Upsell & Cross-sell

When to use it: When a customer replies to a retention campaign, mentions a past order, or asks what else they should try.

Instructions:

Look up what this customer last bought from their order history or customer-scoped purchase-history knowledge.
Open with their name and last product, e.g. “Hope you're loving the Whey Protein 1 kg!”
Use the upsell map to suggest the matching upsell first. Give one line on why it fits them.
Share the product link, price and one image from the catalog.
If they say yes, confirm quantity and send the payment link.
If they say no or not now, suggest the cross-sell once. If no again, thank them and stop.
After payment, confirm the order and expected delivery according to the shipping policy.

Claude prompt · run the retention campaign

Use the whatsapp-messenger skill. Brand: [name], we sell [what], tone [friendly/bold/luxury], store [URL], sender “[Team Name]”.

Run these campaigns: [Upsell, Cross-sell, Reactivation, Returning Customers].

Attached is our customer CSV. Treat customers as inactive after [60] days.

Keep messages under 5 short lines, Hinglish, one CTA.

Also add columns name, last_product and recommended_product so I can map them to AiSensy template variables {{1}}, {{2}}, {{3}}.

Claude prompt · build the upsell map

Attached is my product catalog export. For every product, give me a CSV with columns:

product_bought, upsell_product, cross_sell_product, reason (one line a customer would understand), upsell_link, cross_sell_link.

Upsell = bigger pack, higher tier or upgraded version. Cross-sell = something used together with it.

Only use products from this catalog. Flag any product with no sensible upsell or cross-sell.

WhatsApp template examples

{{1}} = name · {{2}} = last product · {{3}} = recommended product

Each example ends with a quick-reply button so the reply reaches the AI Agent.

Upsell

Hi {{1}}! Your {{2}} must be running low by now. Regulars love the bigger {{3}} — it lasts longer and works out cheaper per serving. Want the details?

Button: Show me

Cross-sell

Hi {{1}}, since you're using {{2}}, you might like {{3}} — they work great together. Should I send you the link?

Button: Tell me more

Reactivation

Hi {{1}}, it's been a while! We've added new things since your {{2}}. Want me to show you what's new?

Button: Show me

Returning Customers

Thanks for coming back, {{1}}! Based on your {{2}}, we picked {{3}} for you. Want a quick look?

Button: Yes please

Keep template claims true for your brand. Add an opt-out line if the template category requires one.

Mistakes to avoid

Conflicting knowledge: remove old prices and outdated documents.
Every skill switched on: enable only the jobs you need.
Long replies: keep WhatsApp answers Short or Medium.
No fallback or handoff: give customers a route to a human.
Broadcasting without opt-in: it annoys customers and can harm the number's WhatsApp quality rating.
The same message for everyone: segment first.
Setting and forgetting: review real chats weekly and fix Knowledge or Ground Rules.

Frequently asked questions

Is it free?

It is free to build and test: 500 AI messages on up to 3 test numbers. Replying to real customers needs a paid plan plus an AI Messages pack. Pricing varies by plan, billing cycle and country, so check it in your account.

Do I need to code?

Not for the basic build. Custom API Actions need someone comfortable with an API; you can paste a cURL and test it in the dashboard.

Does it speak Hindi or Hinglish?

The agent replies in the language the customer types in.

Will it make things up?

Clean Knowledge and a configured fallback reduce that risk. Test questions it cannot answer before going live.

Can I keep my existing chatbot flows?

Yes. Flows and the AI Agent work in the same account. Use flows for fixed menus and the agent for open conversation.

Can it send messages first?

Only through approved templates. Free-form replies work inside the 24-hour window after a customer messages you.

Keep learning

This page adapts WhatsApp AI Agent Playbook — AiSensy Setup for Every Industry + the Claude Retention System, by Priyank Singh, dated October 9, 2026.

Sources linked by the playbook:

The source also names the AiSensy AI Sales Agent creator brief.