Conversational customer service is support delivered as one continuous, two-way conversation, by a person, an AI agent or both, on channels such as chat, email, WhatsApp, SMS and social DMs. The conversation carries its context: who the customer is, what they ordered and what was already said, so nobody repeats themselves and the issue moves to a resolution. If you want ready-made wording, see our templates hub.
What is conversational customer service?
Conversational customer service treats each customer issue as one conversation that stays open until the problem is solved. It is not a chain of disconnected tickets. A person, an AI agent or both can reply inside it, and the history travels with every reply, so the customer never starts again.
The term is broad and every vendor stretches it. For an online store it has a concrete meaning: the customer’s order, delivery and return travel with the conversation. This table shows how it differs from working ticket by ticket.
| Ticket by ticket | Conversational | |
|---|---|---|
| Unit of work | A ticket | A conversation about one issue |
| Context | The customer re-explains | History travels with the thread |
| Tone | Form reply | Two-way, plain language |
| Who replies | One queue | AI, human or both in one thread |
| Success measure | Ticket closed | Issue resolved |
What is conversational AI for customer service?
Conversational AI is software that understands a customer’s message in plain language, keeps track of the exchange over several turns and replies in kind, instead of walking the customer through fixed buttons or keyword triggers. In e-commerce support the useful version also reads the order and acts on it.
The same idea goes by other names: conversational AI for customer support, for customer engagement, or simply conversational AI customer service. They describe one thing. The table separates three kinds of tool that often get the same label.
| Scripted bot | Conversational AI | AI agent | |
|---|---|---|---|
| Input | Buttons or keywords | Free text | Free text |
| Memory across turns | None or limited | Yes | Yes |
| Data | FAQ only | FAQ and knowledge base | Live order, catalogue and policy data |
| Actions | None | A few | Takes actions within rules |
| Failure mode | Dead end | Confident wrong answer | Governed by policies, hands back |
The last row links to the risks below. For the full comparison, read our chatbot comparison. For the question of AI versus a human team, read AI agents compared with people.
How does one conversation carry context across channels?
The conversation belongs to the customer, not the channel. A customer who emails about a late parcel and then messages on WhatsApp should land in the same thread with the same history, so whoever replies, person or AI, starts from what is already known.
That is normal behaviour, not an edge case. CM.com surveyed over 7,000 consumers (a CM.com survey published in May 2023) and found that 46% use multiple channels at once to get a faster response. Expect duplicates across channels, and aim for one thread per issue.
In a store, context means five things:
- Who the customer is, and their order history.
- The order and its fulfilment status.
- Earlier messages in the thread.
- The policy that applies, such as the return window or refund rules.
- Actions already taken, such as a reshipment or a refund.
The thread and its history live in your help desk. An AI agent should read that thread plus live store data. Engaige does not merge channels or route tickets: the help desk keeps the thread, and Engaige works inside it.
What does a conversational thread look like in practice?
A conversational help desk thread is one conversation in which each reply builds on the last, even when the customer switches channel. Below is a short, annotated example from a fictional store, not a real customer conversation. The placeholders in square brackets are filled from the order and the policy.
Email, Monday
Customer: Hello, where is my order [#1234]? It was due today.
Reply
Hello [first name], order [#1234] left our warehouse on [date] with
[carrier] and shows as [in transit] at [last scan location]. The
expected delivery date is [Wednesday]. I will update you here if that
changes.
WhatsApp, Wednesday (same conversation)
Customer: My parcel says delivered but it is not here.
Reply
Hello [first name], I can see order [#1234] was scanned as delivered
on [date] at [time]. Could you check with neighbours or a safe place
at the address? If it has not turned up by [Friday], reply here and I
will arrange the next step under our [missing parcel policy].
Handoff note (internal)
[Reship or refund] for order [#1234] needs approval from [team
member] under the [missing parcel policy] before it is sent.What carries over:
| Line | What carries over |
|---|---|
| Second message, new channel | No “what is your order number?”; the thread and order are already known |
| Second reply | The delivered scan is cited, and one question is asked, the one the policy needs |
| Second reply, policy | The store’s missing-parcel policy is cited by name |
| Closing line | A clear next step and when it happens |
| Handoff note | A reship or refund waits for a person’s approval where the policy says so |
It is a typical WISMO case that turns into one of the missing parcels cases. For longer threads and phrase-level wording, see our conversation examples, chat examples and email templates.
Which channels suit conversational customer service?
Text channels do. Email, live chat, WhatsApp, SMS and social DMs let the customer reply when it suits them and give whoever answers, human or AI, time to read the order and check the policy before replying. Phone is conversational too, but it runs in real time.
Engaige is text-first by design: email, chat, WhatsApp, SMS and social DMs, no voice AI. A live call gives an AI seconds to reason; text gives it time to read the order, check your policy and act. If phone is a big share of your mix, a voice specialist is the honest answer, and our guide to voice AI is the place to start.
The evidence is mixed, so read it as it is. In the same CM.com survey, a vendor survey, phone is still the most preferred channel, at 69% of consumers. But around 80% say they call only when something is urgent. 39% prefer email, and 37% expect customer service to be reachable on messaging platforms, rising to about 50% of under-35s.
| Channel | Why it suits conversation | Watch-out |
|---|---|---|
| Asynchronous, room for long context | Slower expectations | |
| Live chat | Fast, on the site | The customer may leave |
| WhatsApp and SMS | Mobile, asynchronous, the thread persists | Opt-in rules apply |
| Social DMs | Customers already there | Public brand pressure; move to DM |
| Phone | Real time, good for urgent or complex cases | Little time to check data |
What are the benefits and risks of conversational customer service?
Done well, customers stop repeating themselves, simple questions get answered at any hour and agents get the hard cases with context attached. Done badly, an AI that sounds confident gives a wrong answer, and a small-claims tribunal has held a company responsible for what its chatbot said (see Liability below).
Benefits
- No repeat explanations. The history travels, so the customer says it once.
- Answers at any hour. Simple, policy-based questions do not wait for the morning.
- Agents work on the hard cases. They get them with the thread and order attached.
- One place for the whole history. Anyone who picks it up sees what was said and done.
Risks
Liability. In Moffatt v. Air Canada (2024 BCCRT 149, February 2024; see this case summary), the airline’s website chatbot told a grieving passenger he could claim a bereavement fare retroactively within 90 days. The airline’s policy said otherwise. Air Canada argued it could not be held liable for what the chatbot said.
The tribunal called the idea that the chatbot is a separate legal entity responsible for its own actions “a remarkable submission”, and found Air Canada did not take reasonable care to ensure its chatbot was accurate.
It ordered the airline to pay CA$812.02 in damages and fees, as MobileSyrup reported. It is a small-claims tribunal decision, not binding precedent, but the reasoning is easy to imagine in your own market. This is not legal advice.
Customer trust. A Gartner survey of 5,728 customers (December 2023, published July 2024) found that 64% would prefer companies didn’t use AI for customer service, and 53% would consider switching. The top concern was that it gets harder to reach a person, followed by job displacement and wrong answers.
Disclosure. Salesforce’s customer research reports that 72% of customers say it is important to know if they are communicating with an AI agent.
Article 50(1) of the EU AI Act goes further: AI systems that interact directly with people must be designed so people are informed they are interacting with an AI system, unless that is obvious. According to the Act’s own text, Article 50(1) applies from 2 August 2026. Again, this is not legal advice.
| Risk | What goes wrong | How to reduce it |
|---|---|---|
| Wrong answer about policy (Moffatt) | The AI states a policy you do not have | Answer from your actual policy and show which one was used |
| No way to a person (Gartner concern) | The customer is stuck with the AI | A visible handoff with the thread attached |
| AI not disclosed (Salesforce 72%, AI Act Article 50) | The customer thinks they are talking to a person | Say it is an AI agent |
| Acting without approval | A refund or reship nobody checked | Set what the AI may do alone and where a person checks first |
| Robotic tone | Replies feel like a form | Train it on your team’s real replies |
How do you measure conversational customer service?
Measure whether issues end, not whether messages were sent. Track resolution rate, how often customers contact you again about the same issue, how often the AI hands back to a person and why, first response time per channel, and satisfaction after the conversation closes.
| Metric | What it tells you | Watch-out |
|---|---|---|
| Resolution rate | How many issues end without further work | Keep “resolved end to end” apart from “worked on” |
| Repeat contact on the same issue | Whether context was lost | Needs the issue tracked across channels |
| Handoff rate and reasons | Where a policy is missing | A low rate can hide wrong answers |
| First response time by channel | Whether people wait | Fast but wrong is worse than slower and right |
| Satisfaction after close | How it felt to the customer | Survey bias towards extremes |
The cost side is covered in our guide to ROI.
Where does Engaige fit?
Engaige (that’s us) isn’t a chat app or an inbox: it’s the e-commerce AI support agent that works inside the help desk your conversations land in. It does conversational customer service on text channels, with Engaige AI configuring the agent and showing which policies it used. At Otrium, 60% of 120,000 annual tickets are resolved end to end.
The angle here is context and tone. Engaige reads the order and the thread before it writes, so a second message on a new channel starts from what is known instead of from a question about the order number. On Shopify, that means live Admin API data and your storefront content, and every answer and action shows why, with the source, so you can check it against your policy.
Tone is the other half. Otrium’s AI agent Oliver, built with Engaige, resolves repetitive tickets in a natural, conversational tone, and the team did not want it to sound robotic.
Engaige offered control, flexibility, and the ability to really incorporate AI in a more human way.
In short, Engaige drafts inside your help desk first, ticket types go autonomous one at a time, and every reply your team sends becomes training data. The full loop sits in the templates hub. Read the Otrium case or see a demo on your own tickets. One caveat: Engaige is e-commerce-specialised, not a generic horizontal tool, and a team where phone dominates should look at a voice specialist.
Frequently asked questions
What is conversational customer service in simple terms?
It is support handled as one continuous conversation per issue, by a person, an AI agent or both. The history, the order and earlier actions travel with the thread, so the customer never has to explain again, and the conversation stays open until the issue is solved.
What is the difference between conversational AI and a chatbot?
A scripted chatbot follows buttons or keywords and often hits a dead end. Conversational AI reads free text and remembers the exchange. An AI agent goes further: it reads live order and policy data and acts within rules. The full comparison is in our chatbot-versus-agent guide.
Is conversational customer service only for live chat?
No. It applies to email, live chat, WhatsApp, SMS and social DMs, and to phone too. Text channels suit it best because whoever answers has time to read the order and check the policy. The point is one thread per issue, whatever the channel.
Can a store be held liable for what its AI agent says?
Yes, it can happen. In Moffatt v. Air Canada (2024 BCCRT 149), a British Columbia tribunal found the airline did not take reasonable care to keep its chatbot accurate and ordered it to pay CA$812.02 in damages and fees. It is a small-claims decision, not binding precedent, and not legal advice, but answer from your real policy.
Should customers be told they are talking to an AI?
Yes. 72% of customers say it matters to know if they are communicating with an AI agent (Salesforce), and Article 50(1) of the EU AI Act expects people to be told, unless it is obvious, from 2 August 2026. Say so in the conversation, and check the rules that apply to you.
How do you measure whether conversational support works?
Look at whether issues end. Track resolution rate, repeat contact on the same issue, how often the AI hands back and why, first response time per channel and satisfaction after close. Keep “resolved end to end” separate from “worked on”, or the numbers will flatter you.
Does Engaige handle phone calls?
No. Engaige is text-first by design, with no voice AI: a live call gives an AI seconds to reason, while text gives it time to read the order, check your policy and act, so resolution quality is higher. For a phone-heavy mix, a voice specialist is the honest answer.
Do I need to replace my help desk to offer conversational customer service with AI?
No. Engaige works inside the help desk you already run. It drafts replies there, learns from every reply your team sends, and Engaige AI proposes what to change while you approve it. Keep your help desk, replace the AI.
