In this article (24 sections)
In short: in an online store, a chatbot answers and an AI agent resolves. A chatbot (menu-based or powered by generative AI) responds to questions using information loaded in advance: hours, payment methods, return policy. An agent queries and operates your systems: it looks up the order, checks real inventory, starts the exchange, builds the offer and sends the payment link, and remembers what the customer bought before. The three differences that matter are actions, integrations and memory. A chatbot is enough if your inquiries are few and generic; an agent is justified when most messages depend on data about each customer or each order.
The general comparison between the two, for any type of company, is in AI agent or chatbot: the differences that matter. This article takes that comparison down to an online store: which inquiries each one resolves, examples by vertical, reference costs and how to move from one to the other. If you are weighing an ecommerce chatbot vs an AI agent, this is the store-specific version.
Three tools sold under the same name
In ecommerce, "chatbot" is used for very different things. It helps to separate them before comparing. And if you are shopping for an AI chatbot for an online store, the label on the product page will not tell you which of the three you are getting.
Rule-based chatbot | Generative AI chatbot | AI agent | |
|---|---|---|---|
How it understands | Buttons, menus or keywords | Natural language | Natural language |
Where the answer comes from | Text written in advance | Your FAQs, your pages and your policies | Your policies and your live systems: store, inventory, shipping, payments |
What it can do | Inform and hand off | Inform better and hand off | Inform, look things up, take actions and hand off with the case already built |
Memory | None | The current conversation | The conversation and the customer's history |
Example | "Type 2 for returns" | "Returns are accepted within 30 days, unworn and with tags" | "Your order 4812 is within the window. We have size 9 in black. I'll generate your return label" |
The generative AI chatbot is the one that causes the most confusion: it converses well and looks like an agent, but it only knows what is in your text. It can explain the return policy. It cannot tell you whether your order qualifies.
The three real differences
1. Actions or answers
A chatbot finishes its job when it answers. An agent finishes when the task is done or handed off.
In a store, the typical actions of an agent are:
- Looking up an order by number, email or phone.
- Checking shipment status with the carrier.
- Verifying inventory for a specific variant.
- Starting an exchange or a return and generating the label.
- Applying a coupon within a cap.
- Building a cart or generating a payment link.
- Logging a claim with details and photos for the team.
Each action is a tool someone has to build and connect. That is why an agent costs more than a chatbot: the AI model is the cheap part.
2. Integrations
A chatbot lives in the channel (web chat, SMS, Instagram, WhatsApp) and is fed by text. An agent also needs read connections and, sometimes, write connections to:
System | What the agent looks up or does |
|---|---|
Store platform (Shopify, WooCommerce, BigCommerce or custom) | Orders, products, customers, carts |
Inventory system or ERP | Real availability if it is not in the store |
Carrier or shipping software | Status and estimated delivery date |
Payment processor (Stripe, PayPal, Shop Pay) | Payment links, payment status, refunds |
Helpdesk or CRM | Case logging and handoffs |
Integrations set the ceiling of the agent. If your inventory lives in a spreadsheet that is updated on Mondays, the agent will answer with Monday's inventory. The same is true if you sell on Amazon or Walmart Marketplace and in retail locations as well: if inventory is not synced across channels, the agent inherits the lag.
3. Memory
There are two kinds of memory, and the practical difference is large.
- Conversation memory: remembering what the customer said three messages ago. Any generative AI chatbot has it.
- Customer memory: knowing what they bought, when, what size they wear, whether they had a complaint, whether they agreed to receive offers. That comes from your systems, and it is what makes post-sale and reorders possible.
With customer memory, the agent can say, "Last time you bought size 9 in this model, which runs small; want the 9.5?" Without it, every conversation starts from zero.
The same inquiry, three answers
Four real messages from any store and how each tool handles them.
"Where is my order?"
- Rule-based chatbot: shows a menu and asks for the order number; if the customer doesn't have it, it hands off.
- AI chatbot: explains how to track a shipment and sends the generic tracking link.
- Agent: identifies the customer by phone number, finds the order, checks with the carrier and replies with the estimated date.
"I bought a size 8 and it's too small. Can I swap it for a 9?"
- Rule-based chatbot: sends the text of the return policy.
- AI chatbot: explains the policy naturally and tells the customer to email support.
- Agent: verifies the order date, confirms a 9 is in stock in that color, starts the exchange and sends the label.
"Will this charger work with my laptop, model X?"
- Rule-based chatbot: has no answer; hands off.
- AI chatbot: if compatibility is written on the product page, it finds it; if not, it may invent an answer with confidence. That is the main risk in this category.
- Agent: looks up the product's compatibility attributes and, if there is no data, says it doesn't know and hands off.
"Do you have something similar but cheaper?"
- Rule-based chatbot: not applicable.
- AI chatbot: suggests products it finds on your pages, without knowing whether they are in stock.
- Agent: searches for in-stock alternatives in the price range, builds the comparison and sends the payment link for the one the customer picks.
When each one makes sense
Your store's situation | Recommendation |
|---|---|
Fewer than 20 inquiries a day, almost all about shipping and payment methods | Rule-based chatbot or saved replies in your messaging app |
Many general questions, small catalog, no need to look up orders | Generative AI chatbot over your policies and product pages |
Most inquiries are "where is my order" and exchanges | Agent, at least for post-sale |
Products with compatibility, sizing or configuration questions | Agent connected to product attributes |
Replenishment products (consumables, food, cosmetics) | Agent with customer memory for reorders |
Several systems (store, ERP, your own logistics) | Custom agent |
Support team overwhelmed during sale events | Agent for pre-sale and post-sale, with clear handoff |
A simple rule: look at the last 100 messages your store received. If more than half require looking up an order, inventory or a shipment in order to answer, a chatbot will hand off almost everything and the savings will be small.
Which channels, for a US store
The channel is separate from the tool. A rule-based chatbot, an AI chatbot or an agent can all run on the same channels, and for a US store those are usually:
- Web chat on the store itself, where most pre-sale questions start.
- SMS, which works well for order updates and reorder reminders when the customer has opted in.
- Email, where the agent can draft or send replies to support tickets.
- Instagram, for stores that get product questions in direct messages.
- WhatsApp, which is relevant if you serve Hispanic or international customers who already use it as their main way of messaging a business.
One rule applies on WhatsApp. Since January 15, 2026, Meta bars general-purpose AI assistants from the WhatsApp Business API, but business agents for a company's own products and services are allowed. A store agent that handles your orders and your catalog falls in the second group. For the WhatsApp side in detail, see AI agent for WhatsApp: what it handles.
Examples by vertical
Vertical | Typical inquiries | What is enough | Where an agent pays off |
|---|---|---|---|
Apparel and footwear | Sizing, exchanges, shipment status | Chatbot for the size chart | Size exchanges with real inventory and an automatic return label |
Electronics and accessories | Compatibility, warranty, shipping | Little: almost everything depends on the model | Compatibility from attributes, warranty claims with photos |
Beauty and personal care | Skin type, ingredients, replenishment | AI chatbot for general advice | Reorders when the product is about to run out |
Pet food and supplements | Replenishment, flavors, scheduled shipping | Chatbot for product questions | Reorder reminders with a payment link and variant swaps |
Home and furniture | Dimensions, lead times, assembly | Chatbot for published dimensions | Lead times based on inventory and the factory, delivery coordination |
Auto and motorcycle parts | Compatibility by make, model and year | Almost never enough | Lookup by vehicle and available alternatives |
Wholesale | Price lists, minimums, account balances | Chatbot for general terms | Customer-specific pricing, account statements and order taking |
If you sell through WhatsApp with orders built in the conversation, the agent can also take the order: it reads what the customer writes, matches it against your catalog and prices, confirms it and logs it in your system.
What changes in a store during peak season
Ecommerce has a rhythm that most service businesses don't. Volume is flat for weeks and then spikes around Black Friday, Cyber Monday, the holidays or your own sale events. The questions in those weeks are also different: where is my package, can I still return this after the holidays, is this in stock in my size, does this coupon stack with the other one.
That rhythm is the practical test for the choice. A chatbot with menus holds up while volume is low, and then hands off the same questions to a team that is already stretched. A generative AI chatbot can explain your holiday return policy, but it cannot check whether a particular order qualifies. An agent can do the lookups on its own, so the team only receives the cases that really need a person, with the order already identified and the history attached.
Two practical consequences follow. First, test with a peak-season scenario before you decide, not with a calm week. Second, make sure the agent and the store read prices and promotions from the same source, so that it never quotes a sale price that ended yesterday. If the agent hands off to a person, it should do it with the case already summarized, so the customer doesn't have to repeat anything.
Reference costs
Market reference ranges as of October 2026. Platform prices change often: verify each vendor's pricing before deciding.
Option | Upfront cost | Reference monthly cost | What it includes |
|---|---|---|---|
Saved replies and an away message in a messaging app | USD 0 | USD 0 | Fixed text from the app |
Rule-based chatbot on a messaging platform | USD 0 to 500 | USD 15 to 150 | Menus, handoff, multiple channels |
Generative AI chatbot over your text | USD 0 to 1,500 | USD 30 to 400, depending on volume | Natural-language answers over your policies and product pages |
Agent on a support platform | USD 0 to 3,000 setup | Subscription plus, with some vendors, a charge per resolved conversation, on the order of one dollar per resolution | Predefined actions with standard integrations |
Custom agent | At Atomix, from USD 3,000 per project | At Atomix, maintenance from USD 300, plus AI model usage, hosting and channel messages | Actions, integrations and rules specific to your business |
On every option that uses WhatsApp, add the cost of messages the store initiates (marketing and utility templates), which Meta charges per message and which varies by country. SMS carries its own per-message cost.
How to do the math
To compare, calculate the cost per resolved conversation, not the subscription:
- Count how many post-sale and pre-sale conversations you receive per month.
- Estimate how many each option would resolve without handing off. A rule-based chatbot resolves the generic ones; an agent also resolves the ones that depend on data.
- Multiply the handed-off conversations by the time your team takes and the cost of that time.
- Add the cost of the tool.
At low volumes, the chatbot almost always wins. At medium and high volumes, the difference is made by the conversations the chatbot hands off and the agent closes. At high volumes, the per-resolution charge on some platforms can exceed, within a few months, what it costs to build an agent of your own.
The risks of each one, in a store
Rule-based chatbot: the customer who can't find their case in the menu abandons or writes through another channel, more annoyed. During sale events, the chatbot hands off everything and your team gets overwhelmed anyway.
Generative AI chatbot: it answers with confidence things it doesn't know. In ecommerce that shows up as invented compatibility claims, delivery times that don't exist or expired promotions. It is controlled by limiting its answers to your text and having it say "I don't know" when it can't find the data.
AI agent: it can execute a wrong action, such as an exchange outside the window or a discount that didn't apply. It is controlled with caps, human confirmation on actions involving money at first, and a weekly review of conversations. The other risk is dependence on integrations: if the carrier's API goes down, the agent has to know how to say so instead of making something up.
Signs your chatbot has fallen short
- More than half of conversations end up handed off.
- Your team copy-pastes order status by hand.
- Customers type "I want to talk to a person" in the second message.
- During sale events, response time goes from minutes to hours.
- You have customers who buy the same thing every month and nobody writes to them.
How to move from a chatbot to an agent
You don't have to throw away what you have.
- Reuse the channel. Your web chat, your Instagram account and your number stay. If you use WhatsApp, an agent connects through the official WhatsApp Business API rather than the standalone app, so confirm that with your provider and pick one that supports it.
- Reuse the text. The chatbot's answers are the starting material for the policies the agent applies.
- Start with one action. It almost always makes sense to start with "where is my order," which is the highest volume and the lowest risk.
- Combine. Menus or fixed answers for the trivial, an agent for what needs data. The customer doesn't have to notice the switch from one to the other.
- Measure against a baseline. Resolution rate without handoff, response time and conversations per team member, before and after.
What an agent does at each stage of post-sale and reorders is in AI agents for ecommerce. If your goal is to be visible to shoppers' own assistants rather than to answer your customers, that is a different problem, covered in prepare your store for AI shopping agents.
What to ask a vendor before you sign
Many products are presented as an "agent" and are generative AI chatbots. These questions separate one from the other in a meeting:
- Which systems does it connect to, and what does it do in each one? Ask for the concrete list: reads orders, creates exchanges, generates payment links. "It integrates with your store" is not enough.
- What happens when it can't find the data? The right answer is that it says so and hands off. If the vendor has no clear answer, the product will make things up.
- How are caps defined? Maximum discounts, refund amounts, exchange windows. They should be rules you control, not loose instructions in a block of text.
- What does my team see when it hands off? Ideally the case summary, the identified order and what was already tried, inside the tool your team already uses.
- How is it billed as volume grows? Flat subscription, per conversation, per resolution or per message. Ask for a simulation with the volume of a sale-event month, not a quiet one.
- Who owns the conversations and the data? You should be able to export them if you switch vendors.
- Can I see it working with my data before I pay? A demo with generic questions says nothing about how it will answer with your catalog and your orders.
If a vendor answers all seven well, what it offers resembles an agent. If it fails the first and the second, it is a chatbot, even if it costs like an agent.
Frequently asked questions
What is the difference between a chatbot and an AI agent in ecommerce?
The chatbot answers with information loaded in advance. The agent queries your systems (orders, inventory, shipping, payments) and carries out actions such as starting an exchange, applying a coupon or sending a payment link, and it remembers the customer's history.
Is a ChatGPT-powered chatbot an AI agent?
Not necessarily. A chatbot that uses a generative AI model converses better, but if it isn't connected to your orders and inventory and can't carry out actions, it is still a chatbot. What turns it into an agent is the tools and the integrations.
Do I need an AI chatbot for my online store or an agent?
If you get few inquiries and they are general, a chatbot or saved replies are enough. When most messages are about orders and exchanges, an agent makes sense, at least for post-sale. The check is to read your last 100 messages and see how many need data from your systems to answer.
How much does an AI agent for an online store cost?
Support platforms with an agent charge a monthly subscription and, with some vendors, a charge per resolved conversation, on the order of one dollar per resolution as a reference. A custom agent at Atomix starts at USD 3,000 per project, plus monthly maintenance from USD 300 and AI model usage and channel messages.
Can an agent make more mistakes than a chatbot?
It can be wrong in a different way: a chatbot fails by not answering, while an agent can execute an incorrect action. That is why actions involving money have caps or human approval at first, and conversations are reviewed every week.
Can I have both a chatbot and an agent?
Yes, and it is the most common setup. Fixed answers or menus for the trivial, and an agent for what needs data or actions, in the same web chat or the same WhatsApp number.
Which channels can a store agent use in the US?
Web chat, SMS, email, Instagram and WhatsApp. WhatsApp matters mostly if you serve Hispanic or international customers. Meta allows business agents for a company's own products on the WhatsApp Business API, while general-purpose AI assistants have been barred since January 15, 2026.
If you want to compare, using your own messages, what a chatbot would resolve and what an agent would resolve, we build a free working demo in 24 to 48 hours with data from your store. You can book a call or see how we build AI agents for business.