Skip to content

USE CASE

WhatsApp AI Chatbot for E-commerce Customer Support

A WhatsApp AI chatbot for e-commerce answers product questions, tracks orders and handles returns inside the chat, and hands everything else to your team. CognyX built one GPT-4 chatbot that serves an Indian e-commerce brand on WhatsApp, its website and Instagram; support tickets fell 60%. Delivery starts with a 2–6 week pilot, then 6–12 weeks to production.

Book a 30-minute review

At a Glance

Typical figures for a project with an India-based team. Third-party facts are in the sources list.

ItemWhat to expect
Typical cost$8,000–$35,000 to build; rates from $25/hr
Timeline2–6 week pilot, then 6–12 weeks to production
Running costAbout $200–$6,000 a month, plus Meta’s WhatsApp message charges
ChannelsWhatsApp, website chat and Instagram, from one bot
IntegrationsWhatsApp Cloud API, Instagram messaging API, your store’s order and returns APIs
LanguagesEach language your customers use, listed in discovery and tested in the pilot
Compliance postureGDPR ready controls; opt-in before any outbound WhatsApp message
HandoffTransfer to your support team with the full chat history

What the Bot Handles and What Stays With Staff

The bot takes the requests that follow a clear process and need data from your store. Your support team keeps anything that needs judgment or moves money.

Requests the bot handles

  • Product questions. Sizes, stock, delivery times and store policies, answered only from your catalogue and approved content, through retrieval-augmented generation (RAG).
  • Order tracking. It finds the order, reads its status and sends the carrier’s tracking link.
  • Returns. It checks the order against your return policy, collects the reason and files a return request for your team to approve.
  • Order updates (optional). Shipping messages to customers who have opted in, sent as approved templates.

The first three are the request types the case-study bot handles.

What stays with your team

  • Refunds and exceptions. Refunds, credits and requests outside the written policy go to a person.
  • Payment problems. The bot never asks for full card or bank account numbers; WhatsApp’s Business Messaging Policy bars it.
  • Everything else. Complaints, damaged or missing items, requests for a person and chats the bot cannot resolve go to your team with the chat history.

The rule: the bot answers and files; your team decides anything that moves money or bends a policy.

How a Conversation Works, From First Message to Resolved

Every chat follows the same steps, and every step is logged.

  1. The message arrives. A customer writes on WhatsApp, the website chat or Instagram, and a webhook passes it to the bot. The bot says it is an automated assistant.
  2. The bot works out the request: product information, an order status, a return or something else.
  3. It checks who it is talking to. Before it reads an order, it matches the customer, for example the WhatsApp number against the phone number on the order.
  4. It reads only what the task needs: one order, or the return rules for one item.
  5. It completes the task in the same conversation: an answer, a tracking link or a filed return request.
  6. It hands off when it should, with the chat history, on a complaint, a policy exception or a request for a person.
  7. Sensitive actions wait. Actions you mark as sensitive, such as changing a delivery address, go to a staff queue first. The audit log records what the bot read and changed, with sensitive fields redacted.

Integrations: WhatsApp, Instagram, Web Chat and Your Store

The bot connects to each channel and to your store through their APIs. On your website it runs in a chat widget. Each action has a defined schema, input checks and an allow-list, so the bot can run only the actions you approve.

WhatsApp Business Platform (Cloud API)

Meta’s Cloud API sends text, media and interactive messages, and delivers incoming messages and status updates to your server through webhooks. Three rules shape the design:

  • The customer service window. Each customer message opens or resets a 24-hour window. Free-form replies, from the bot or a person, are allowed only while it is open.
  • Templates for everything else. Outside the window, you can send only approved templates. Meta reviews each one, which can take up to 24 hours, as marketing, utility or authentication.
  • The category matters. A shipping update is utility if it has no promotion. A cart reminder is marketing, even if the customer asked for it.

Instagram

Instagram messaging runs through Meta’s Instagram API with Instagram Login, on a professional account. The bot can reply only after the customer writes: within 24 hours for most actions, and up to 7 days after a first message from a Click-to-Direct ad. Your team can tag replies to use a separate 7-day Human Agent window.

Store, order and returns systems

The bot reads orders and files returns through your store’s API. On Shopify, for example:

  • The Admin API Order object returns only the last 60 days of orders by default; older orders need Shopify’s approval.
  • Each fulfillment carries the tracking company, number and URL.
  • The returnRequest mutation creates a return with the status REQUESTED, which the merchant must approve or decline.

For other platforms or a custom order system, we use that system’s API and confirm access in discovery.

If we built the case-study bot today

We would use the OpenAI Agents SDK with GPT-6.1 Terra in place of GPT-4, and connect the order and returns systems through the Model Context Protocol (MCP).

Data Protection and Safety

Our controls are GDPR and SOC 2 ready. On a store bot, the design starts from three questions: who has opted in, what customer data the bot needs, and how a customer reaches a person.

Opt-in before you message first

WhatsApp requires opt-in before a business messages a customer. It can be a general opt-in, not only for WhatsApp, if it meets local law, but it must say clearly that the customer is opting in to messages and name your business. Meta suggests separate opt-ins for order updates and offers, and clear opt-out instructions.

A clear path to a person

WhatsApp’s Business Messaging Policy allows automation in the 24-hour window, but requires prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent. On Instagram, Meta asks automated chats to say they are automated where the law requires it, and recommends it everywhere. The bot says so at the start of each chat.

Keep customer data to the minimum

The bot asks for, and reads, only what the task needs. Shopify’s app rules work the same way: process only the minimum personal data, with stricter rules for name, address, phone and email. WhatsApp’s policy bars asking for full card, financial account or ID card numbers, and requires a published privacy policy. We add scoped credentials, per-tool rate limits, PII detection, encryption, role-based access and redacted audit logs.

Keep the bot on your store

WhatsApp’s Business Solution Terms bar AI Providers from offering AI, such as general-purpose assistants, on the platform when the AI is the primary product rather than incidental; Meta decides what counts. We keep the bot to your products, orders and returns. Answers come only from approved sources, and filters block off-topic output.

Results and How We Measure Them

The case study

We built one GPT-4 chatbot that serves an Indian e-commerce brand’s customers on WhatsApp, the website and Instagram. It handles product questions, order tracking and returns in the same conversation.

Support tickets down 60%. Read the multi-channel e-commerce chatbot case study.

What we measure

We set the success metrics in discovery, before we build:

  • Ticket volume: helpdesk tickets by type, before and after launch (the case-study metric).
  • Resolution rate: chats completed without staff.
  • Handoff rate: how many chats go to staff, and why.
  • Cost per completed task, including Meta’s message charges.

Real past chats become the test set, run again after each prompt or model change. The go or no-go decision rests on these numbers, not on a demo.

Cost and Timeline

Typical ranges with an India-based team, from our AI agent development cost guide:

ItemTypical range
Build: FAQ chatbot, no tool use$5,000–$12,000
Build: single-task support agent (FAQ and triage)$8,000–$15,000
Build: agent that reads and writes to your systems (orders, returns)$15,000–$35,000
Build: published range, all regions$5,000–$180,000
Running cost, AI side$200–$2,500 a month (single-task); $1,500–$6,000 (with tool use)
Upkeep of prompts and knowledge base10–30% of build cost per year
CognyX ratesFrom $25/hr
  • Example: a support agent that handles 10,000 conversations a month costs about $1,680 a month to run, or $0.17 per conversation. LLM tokens are about 28% of that.
  • WhatsApp charges are separate. Meta bills per delivered message, by category and market. Utility templates, such as shipping updates, are charged even inside the 24-hour window. Service replies are free for the first 1,000 a month per business phone number.
  • Billing in India. If you pay Meta directly and your Sold-To country is India, Meta requires your messaging accounts to move to INR billing by 31 December 2026.
  • Final price depends on channels, languages and integrations. Well-scoped chatbot builds suit a fixed-price project.

Timeline

  1. Discovery and design, about 2 weeks: map your request types, channels and languages, set the KPIs, and confirm API access to WhatsApp, Instagram and your store.
  2. Pilot, 2–6 weeks: a working bot on one channel and the main request types, tested on real data against the KPIs.
  3. Production, 6–12 weeks: the other channels, security hardening, edge cases, stress tests and full connection to your order and returns systems.
  4. Launch and optimize, ongoing: monitoring, prompt and knowledge-base updates, and cost tuning.

Frequently Asked Questions

Can one AI chatbot serve WhatsApp, Instagram and our website?

Yes. CognyX built one GPT-4 chatbot that serves an Indian e-commerce brand on WhatsApp, its website and Instagram. Each channel has its own rules. On WhatsApp, free-form replies are allowed only in the 24-hour window after a customer message. On Instagram, the bot replies only after the customer writes, within 24 hours for most actions.

What does Meta charge for WhatsApp Business messages?

Since 1 July 2025, Meta charges per delivered message from the business, by message category and the customer's country code. Since 1 October 2026, service messages (free-form replies) are charged after 1,000 free ones per business phone number each month. Chats that start from a Click-to-WhatsApp ad can be free for up to 7 days.

Can the chatbot message customers first, for example with shipping updates?

Yes, if the customer has opted in to messages from your business and the message uses a template that Meta has approved. Review can take up to 24 hours. Order and shipping updates are utility messages when they carry no promotion. Meta treats cart reminders as marketing, even when the customer asked for them.

Which languages can a WhatsApp AI chatbot handle?

We build multilingual chatbots. In discovery we list the languages your customers write in, then test each one on real questions in the pilot before go-live. Outbound templates need one version per language: Meta does not translate template text, and each version counts against your template limit.

How long does it take to launch a WhatsApp AI chatbot?

A typical project has about 2 weeks of discovery and design, a 2–6 week pilot on real data, then 6–12 weeks to production. The pilot proves the bot on one channel and the main request types. Production adds the other channels, security hardening, edge cases and full connection to your order and returns systems.

How much does a WhatsApp AI chatbot for e-commerce cost?

With an India-based team, a single-task support agent typically costs $8,000–$15,000 to build, and an agent connected to your order and returns systems $15,000–$35,000. Running cost is about $200–$6,000 a month, before Meta's message charges. CognyX rates start from $25/hr.

Start with a 30-minute review

Bring your request types, monthly chat volume, channels, languages and store platform. In a free 30-minute architecture review, our engineers check which requests the bot can take, which systems it can reach, and the build and running cost at your volume. You own the code and IP, and an NDA applies from day one.

Book a 30-minute review

Our AI engineers check your request types, channels and volume, and scope a 2–6 week pilot.

Sources

Published by CognyX AI, an AI development and consulting company in Ahmedabad, India. Ranges are typical figures as of November 2026; your cost depends on scope, volume and channels. Meta’s message charges are not included in our build or running-cost ranges. This page is not legal advice: confirm your consent and data protection duties with your compliance team.

CognyX shipped a production-ready AI agent that integrates with our LMS and support stack. Clean architecture, secure data handling, on-time delivery and fewer support tickets.
Jerry Xiao, AI Code Academy

Recent Case Studies

See how we've shipped production AI for teams that need AI chatbot development — with the architecture and measured outcome behind each build.

  • E-commerce

    Multi-Channel AI Chatbot for an Indian E-commerce Brand

    Handles product queries, order tracking, and returns — reducing support tickets by 60%.

  • EdTech

    Autonomous AI Agent for AI Code Academy

    Clear uplift in engagement and fewer support tickets, delivered on time.