You already run the numbers on everything else — cost per acquisition, per-order margin, refund rate. Support is usually the last line item you look at, right up until ticket volume starts eating your evenings. This guide covers what a customer service AI chatbot actually does for an ecommerce store, where it fails, and the real per-ticket math — the parts most roundups skip.
What a customer service AI chatbot actually does
At its core, a chatbot converses on one channel and resolves requests that can be answered from structured data. Order status, "where is my package," return-policy questions, and simple product details are the bread and butter. These resolve reliably because the answer already lives in your store's records.
The dividing line worth learning is chatbot versus agent. A chatbot can tell a customer how to request a refund; a true agent can issue the refund in your store. Analysts call the acting version agentic AI — "a system based on generative AI foundation models that can act in the real world and execute multistep processes," per McKinsey's definition as quoted in industry coverage.
Most tools marketed as "AI chatbots" for ecommerce sit on the conversation side of that line. If you want the fuller picture of where chat software ends and action-taking begins, our guide to conversational AI for ecommerce walks the same distinction in detail.
How support AI is priced now
Here is the structural shift that matters for your P&L: support AI is increasingly priced like an outcome, not a seat. You pay when a conversation is fully resolved without a human, not a flat monthly fee per agent.
Gorgias, an ecommerce-focused helpdesk, charges per resolved conversation — "Each resolved conversation costs $0.90 on most plans," with starter plans beginning at a dollar per resolved conversation, per its pricing explainer. A conversation handed off to a human is not billed as an AI resolution. Notably, Gorgias declines to promise an automation rate, stating your actual rate "emerges from usage over time."
Zendesk follows the same logic: its AI agents are "included in every Suite and Support plan, with pricing based on the successful outcomes they deliver," with Suite plans starting at fifty-five dollars per agent per month billed yearly, per Zendesk's pricing page. The pattern tells you something: both vendors build in a handoff path, which is an admission in the business model that these bots do not handle everything.
What automates well — and what doesn't
The honest split matters more than any feature list. Some support work is a clean fit for automation today; some is a trap.
Fits well today
- Tier-1, structured tickets. Order status, tracking, and returns-policy questions resolve from data your store already holds. This is the layer the entire outcome-priced category is built on.
- Triage and routing. Even when the bot can't resolve a ticket, it can tag, prioritize, and route it to a human faster than a person sorting an inbox by hand.
- After-hours coverage. A bot answers at 2 a.m. without overtime. For the easy half of your volume, that is round-the-clock response at no marginal labor cost.
Gartner projects that by 2029 "agentic AI will autonomously resolve 80% of common customer service issues without human intervention," per its March 2025 release. Read the qualifier: common issues. The easy half is the automatable half.
Fails poorly today
- Ambiguous, high-stakes edge cases. The canonical warning is Air Canada's chatbot, which told a customer he could claim a bereavement discount after flying — contradicting the airline's real policy. A British Columbia tribunal ordered the airline to pay CA$812.02 and rejected its argument that the chatbot was a "separate legal entity," per CBC News.
- Novel judgment. A bot can run a playbook; it cannot decide whether to make a goodwill exception for a furious repeat customer who drives referrals.
- Anything consequential without review. This is why vendors converge on approval gates — the bot proposes, a person signs off on the costly stuff.
That Air Canada ruling is the sentence to tape above your desk: your store owns what your AI tells customers. The vendor's bot making the mistake does not transfer the liability.
The real cost math
Marketing copy loves "cut support costs by 80%." Here is the arithmetic for an operating store so you can judge it yourself.
Say you run a store doing 340 orders a month at a $31 average order value, and that generates about 300 support conversations monthly — mostly order status, tracking, and returns. Assume each conversation takes a human eight minutes, and the bot fully resolves half of them. That is a working assumption for the math, not a guaranteed rate, since even Gorgias won't promise one.
The labor baseline, with rate sources: an offshore virtual assistant runs roughly six to ten dollars an hour at a mid level, per DDIY's 2026 Filipino VA rates, while a US-based assistant runs about twenty-eight to sixty-five dollars fully loaded, per CallForce's 2026 rate guide.
Now the two options, using a mid-offshore rate of eight dollars an hour and the Gorgias rate of ninety cents per AI resolution:
- Human handles everything: 300 conversations × 8 min = 40 hours. At $8/hr that is $320/month offshore; at $40/hr it is $1,600/month US.
- Bot clears Tier-1, human takes the rest: 150 AI resolutions × $0.90 = $135, plus 150 remaining × 8 min = 20 human hours. That is ~$295/month offshore-hybrid or ~$935/month US-hybrid, plus your helpdesk subscription.
The honest reading: against a US labor baseline, the bot is dramatically cheaper on Tier-1 volume. Against a six-to-ten-dollar offshore VA, the pure dollar gap on 300 tickets is small — the stronger arguments become instant 24/7 response and zero management overhead, not price.
Two things scale differently, though. Doubling your ticket volume doubles VA hours and eventually forces a second hire plus the time to manage them; per-resolution fees just rise linearly with no hiring step. And neither option removes the human — the bot concentrates your team's attention on the hard half of the inbox.
When a chatbot isn't enough
A single-surface chatbot answers inside one tool. It can tell a customer their order shipped, but it can't cross-check the supplier's fulfillment status, reconcile that against your ad spend for the week, or draft the apology-plus-discount email and log the outcome — the connected work a human support lead actually does.
That cross-tool scope is the difference between a chatbot and what the market now calls an AI employee. The umbrella concept is real, but so is the hype: Gartner predicts "over 40% of agentic AI projects will be canceled by the end of 2027" and warns of "agent washing" — rebranding plain chatbots as agents — estimating "only about 130 of the thousands of agentic AI vendors are real," per its June 2025 release. Both numbers belong in the same breath: the category is genuine and the most over-labeled on the market.
PodVector AI's Victor is an AI employee built for ecommerce and print-on-demand stores, and support is one slice of its job. Victor drafts customer-support email and sends it only after you approve — the seller stays the decision-maker of record, which is the same approval-gate pattern the serious vendors above all land on. Every write action Victor takes is approval-gated; it proposes, you confirm, then it executes.
The cross-tool part is what separates it from a widget. The same system that drafts a support reply can look up the order in Shopify, check supplier status in Printful or Printify, compute the true per-order profit on that order, and save the outcome to a report in your own Google Drive. If you're mapping where this category is headed, the AI employees for ecommerce guide is the hub, and the piece on whether to hire AI developers covers the build-versus-buy decision once you've outgrown a single bot.
How to choose, as an operator
Four questions cut through the vendor noise:
- Does its work product live in your accounts? If more than forty percent of agentic projects die by 2027, you want artifacts — macros, flows, reports — that survive the tool. Prefer software whose output sits in your Shopify, your Klaviyo, your Drive.
- Is there a real handoff path? Unattended-by-design is a red flag, not a feature. You want the bot to escalate what it can't resolve and gate anything consequential.
- How does it price at your volume? Run the per-resolution math against your current support hours before you sign.
- What else does it touch? A support-only bot is one tool; if you're drowning in coordination across ads, orders, and email, a connected AI employee may do the routing that is otherwise your unpaid job. Managing that handoff between software and people is its own skill — our note on AI employee scheduling covers how operators keep it sane.
The defensible headline metric is time, not revenue. Vendor revenue-lift claims are context numbers; the universal, checkable outcome is that structured support work moves off your calendar — and what that does to your P&L depends on what you do with the reclaimed hours.
See it on your own store
Victor connects to Shopify, Meta Ads, Google Ads, Printify, Printful, Gelato, and Klaviyo, drafts approval-gated support email, and computes true per-order profit — every action run past you first. Try PodVector AI free and point it at your live store.
FAQs
Will an AI chatbot replace my support team?
No. Outcome-priced support AI is built on the handoff — you are billed only for what the bot fully resolves, and the rest routes to humans. Per the Gorgias pricing model, conversations handed to a person aren't charged as AI resolutions, per its pricing explainer. The team shrinks per ticket; it does not vanish.
Who is liable if the chatbot gives a customer wrong information?
Your store. A British Columbia tribunal held Air Canada liable for its chatbot's misinformation and ordered it to pay CA$812.02, rejecting the "separate legal entity" defense, per CBC News. Ground the bot in your real policies and data, and keep a review path for anything consequential.
How much does a customer service AI chatbot cost for an ecommerce store?
The category has moved toward per-resolution pricing — around ninety cents per resolved conversation on Gorgias, per its pricing explainer, or per-outcome pricing bundled into Zendesk Suite plans that start at fifty-five dollars per agent per month billed yearly, per Zendesk's pricing page. Run those rates against your monthly ticket count and current labor hours before you commit.
What's the difference between a chatbot and an AI agent?
A chatbot converses and answers on one surface; an agent takes multi-step actions across tools toward a goal. The line in every analyst definition is action-taking — generating text versus executing a workflow, per McKinsey's definition. A widget that can only explain your refund policy is a chatbot; software that can issue the refund is an agent.
Can one tool handle support and the rest of my operation?
Some can. A cross-tool AI employee like Victor drafts support email, checks the matching order and supplier status, and logs outcomes — all approval-gated — rather than living only in the inbox. Whether that beats a dedicated support bot depends on how much of your week goes to coordinating between tools; the AI employees for ecommerce guide lays out the trade-off.
How long before the chatbot actually resolves a meaningful share of tickets?
Expect a ramp, not a switch. Gorgias states the automation rate "emerges from usage over time" — the bot needs your policies, macros, and catalog before its resolution share climbs, per its pricing explainer. Budget review time during the ramp; that is the new cost that replaces execution time.