Which Shopify store questions to automate first, when the chatbot needs order data versus policy training, how to add WhatsApp and proactive cart recovery without pop-ups, and what it costs.
Shopify stores lose carts when product and shipping questions go unanswered. An AI chatbot trained on catalog, policies, and order APIs acts as a 24/7 sales associate sizing, recommendations, and WISMO on website chatbot and WhatsApp. See cart recovery chatbot tactics.
A Shopify AI chatbot answers product, shipping, returns and order-status questions on your storefront and on WhatsApp from your catalog, policies and order data, captures the shopper's details when a question needs a person, and hands off with the transcript. It is for store owners and support leads whose inbox is full of the same questions, and whose checkout loses shoppers who could not find the answer in time.
The decision is which questions to automate first, whether the agent needs live access to Shopify order data or can start on policies alone, whether to add WhatsApp, and how to keep the agent from inventing discounts. This guide covers each with a cost example built on Botcadence pricing. For the wider sector view see AI for ecommerce.
Rank questions by volume and by what happens if the answer is wrong. Automate the top rows first.
| Intent | Source | Agent action | Risk if wrong |
|---|---|---|---|
| Where is my order | Shopify order and fulfillment status | Look up by order number or email, answer with the latest status | Medium; wrong status creates a second ticket |
| Return and exchange eligibility | Returns policy with dates | Answer the rule, start the return if within the window | Medium |
| Size and fit | Size charts, product descriptions, reviews | Answer from the chart; suggest the next size where the chart says so | Low |
| Shipping cost and countries | Shipping policy page | Answer by destination | Low |
| Stock availability | Product data | Answer from live data or say when it is unknown | Medium |
| Coupon and promotion rules | Current offers page | Quote only published offers | High; invented discounts cost money |
| Product recommendations | Catalog and collections | Suggest from the catalog with links | Low |
| Refund approvals, chargebacks, damaged goods | None | Collect details and hand off | High; keep human |
Our cart recovery chatbot and sales chatbot pages cover the pre-sale rows; the logistics delivery updates guide covers the status row in depth.
Start where your data allows and move up one tier at a time.
| Tier | What the agent uses | What it can do | Setup |
|---|---|---|---|
| 0: Policies and FAQs | Uploaded policy pages, size charts, FAQ | Answer rules and product facts | About 5 minutes after documents are uploaded |
| 1: Catalog | Product pages and collections | Recommend and link to products, answer stock questions where data is current | Add the store URL or product feed |
| 2: Orders | Shopify integration for order lookup | Answer where-is-my-order by order number or email | Connect the store |
| 3: Actions | Connected workflows | Start a return, apply a published offer, create a ticket | Rules per action, with confirmation steps |
Tier 0 alone removes a large share of pre-sale questions in most stores. Tier 2 is where status questions stop reaching people. Tier 3 should be introduced one action at a time, each with a confirmation step and a rule for when a person must approve.
Botcadence document training handles tier 0 and 1, the Shopify connector handles tier 2, and workflow automation with Zapier or webhooks handles tier 3. We build Botcadence, so we are not neutral; the tier model applies to any vendor you evaluate.
The website chatbot catches shoppers at the moment of hesitation, on the product page or in checkout, and needs no opt-in. The WhatsApp chatbot reaches shoppers after they leave: order updates, back-in-stock notices, and a reply channel that stays open on their phone. Messages you send first on WhatsApp outside Meta's 24-hour customer service window must be approved templates sent to opted-in customers, and Meta charges per conversation; see the WhatsApp business chatbot guide.
A sample exchange on the website: a shopper on a jacket page asks, "Does this run small, and can I return it if it does not fit?" The agent answers from the size chart and the returns policy in one reply, with the return window and the link to start a return, and asks whether the shopper wants the next size up added to the cart. No person was involved, and the answer matched the policy page word for word because that page is the source.
Run the website first. Add WhatsApp once you have opt-in at checkout and a reason to message, such as shipping updates, rather than as a second place to answer the same question.
Botcadence web chat and WhatsApp start on the Free plan, with Starter at $39, Growth at $99 and Business at $199 a month, and Enterprise custom; the plan depends on reply volume, so check the pricing page.
A worked example, with every input labelled as an assumption. Say a store receives 1,800 support contacts a month. Say 55 percent fall into the first six intents in the table above: 990 contacts. Say a person spends 5 minutes on each: about 83 hours a month. Say a tier 2 agent resolves 75 percent of those 990 without a person: 742 automated, 248 escalated with the transcript, and the human time on those intents falls to about 21 hours, plus the minutes saved on each escalation because the agent already collected the order number. The remaining 810 contacts (refunds, damaged goods, complex questions) stay with people. Put your own numbers in each "say" and compare against the plan price and the hours.
The revenue side is harder to count honestly. Track orders placed within the same session after a chat answer, and treat it as a signal rather than a proof, since those shoppers may have bought anyway.
The agent can open a proactive chat when a shopper pauses on checkout or returns to a cart, answer the shipping or returns question that caused the pause, and quote a published offer. On WhatsApp, it can follow up on an abandoned cart only if the shopper opted in and the message is an approved template.
It cannot read a shopper's mind; a proactive prompt on every page is a pop-up, not help. Trigger it on hesitation signals (time on checkout, a return to the cart, a shipping-page visit) and keep the first message a question, not a pitch. Measure recovered orders, not messages sent.
Choose it when the same ten questions dominate the inbox, when shoppers ask on product pages and leave before you reply, when order status is a large share of tickets, or when you sell across time zones.
Hold off when policies are inconsistent between pages, when order data in Shopify is not kept current by fulfillment, or when your catalog changes daily without a feed. Fix those first; the agent will repeat whatever the data says.
The chatbot deflection glossary entry explains how to count resolution without flattering the numbers.
The Botcadence chat agents page covers the inbound and proactive website agents, the Shopify connector and the WhatsApp channel a store needs to answer shoppers before they leave.
Questions readers ask
Order lookup and inventory accuracy improve with API access; smaller stores can start with policy + FAQ training only.
Where is my order, return eligibility, size/fit, coupon rules, and stock availability automate these before custom stitching requests.
Proactive chat on checkout drop-off and WhatsApp follow-up (with consent) can nudge completion measure recovered revenue, not message count.
Ground answers in uploaded policy docs and disable promotional language unless connected to approved offer rules.
Book a free demo and we will build a working AI agent for the channel this article covers, tailored to your business goals. No commitment required.
Ask AI about this page
