Omnichannel AI and multilingual support: what actually works in customer service
A customer in Riyadh messages your business on WhatsApp in Arabic. Ten minutes later they switch to English, mid-message. Tomorrow they follow up by email, and expect whoever answers to already know what happened. Most support platforms cannot hold that thread. Two separate problems, switching channels and switching languages, get treated as one problem, and both get handled badly.
"Omnichannel" and "multilingual" are two of the most overused words in customer service software. Here is what they actually require from an AI platform, what real research says about why they matter, and how to tell a genuine capability from a marketing claim.
What omnichannel AI actually means
Omnichannel AI is a single system that keeps one shared memory of a customer across every channel, so a conversation that starts on WhatsApp and finishes on email is one conversation, not three. Multichannel just means a business shows up on multiple channels, without that shared memory. The difference sounds small. It is not.
McKinsey's research on customer care found that in the strongest AI-powered care operations, around 65% of tasks and 50 to 70% of contacts get automated, and the businesses that do it well build "a true omnichannel experience that provides a consistent and seamless experience across interactions" (McKinsey, 2022). Notice what that definition rests on: consistency, not just presence.
Here is where most tools fall short. Many were built for one channel first, web chat or email, then had WhatsApp or Instagram bolted on as a separate integration, each with its own memory. The customer repeats themselves. That is multichannel wearing an omnichannel label.
Why "multilingual" is a bigger claim than most platforms can back up
A platform that is genuinely multilingual does not just translate words. It understands intent, checks policy, and can act on a request in whichever language the customer used, including switching languages partway through a conversation without restarting.
It helps to know what is happening under the hood. Researchers studying how large language models process requests in other languages found that many models effectively think in English internally, converting the request into English in the middle layers before generating a reply back in the customer's original language (Zhao et al., NeurIPS 2024). That is useful context for a buyer: a model can sound fluent in a language while reasoning in a different one behind the scenes, which is part of why formal language support and real dialect fluency are not the same thing.
This matters directly in the GCC, where customers write in Modern Standard Arabic, Gulf or Levantine dialect, and Arabizi (Arabic written in Latin letters), sometimes in the same message. A system that only handles textbook Arabic will misread a large share of real customer traffic.
Why this shows up most clearly in travel and airlines
Few industries force omnichannel and multilingual support together as hard as airlines do. A delay is a delay in any language, but a traveler's patience for repeating themselves, or getting a reply in the wrong language, runs out fast.
IATA's 2025 Global Passenger Survey found that more than half of travelers, 54%, now want to deal directly with the airline rather than a third party, and that mobile has become the default channel for managing a trip from booking through baggage claim. When a flight changes, that same traveler expects the airline to reach them on whichever channel, and in whichever language, they are already using. Not a phone queue.
Independent tracking backs this up. The American Customer Satisfaction Index's 2026 airline research shows satisfaction among travelers outside loyalty programs climbing steadily, from 70 in 2022 to 74 in 2026, as airlines invest in more consistent, connected communication across the journey (ACSI, 2026). Airlines that keep the conversation continuous across channels are winning back trust that ticket price alone cannot buy.
And language is not a checkbox added after the fact. CSA Research's study on buying behavior, run for years across dozens of countries, found that 76% of consumers prefer to buy when information is available in their own language, and 40% will not buy from a site in another language at all (CSA Research, 2020). For a GCC travel business, that is not a translation nice to have. That is most of the customer base.
The same logic holds outside travel too, insurance claims after a regional weather event, banking during a public holiday, but airlines make the stakes visible fastest, because delays are public and travelers are already anxious.
How AWX handles omnichannel and language today
AWX runs as one agent with one shared memory across every channel it connects to: WhatsApp, email, Instagram, Facebook, SMS, web chat, and Slack. A customer who starts on WhatsApp and finishes on email never repeats themselves.
On language, AWX's documented capability today is bilingual Arabic and English, including switching between the two partway through a conversation without restarting the thread. That is a deliberately specific claim, not a broad "multilingual" one. Arabic and English switching is the exact pairing that matters most for AWX's GCC customer base, not a placeholder for "handles everything."
Whichever language or channel a request arrives on, the same rule applies. AWX resolves what it is confident about and hands the rest to a person, with the full conversation history and its reasoning attached, so nothing gets lost between languages, or between channels.
Criteria for selecting a generative AI provider for multilingual, omnichannel support
Ask any vendor these questions before you sign, including AWX:
Which specific languages and dialects, exactly?
Multilingual support" as a marketing line means nothing. Ask for the list, and ask whether it covers dialect and informal writing, not just the formal, textbook version of a language.
Does the system keep one memory across every channel, or does each channel start fresh?
Ask to see a conversation move from one channel to another during the demo.
Can a customer switch languages partway through a conversation without starting over?
This is where most "multilingual" claims quietly fail.
Where is customer data processed and stored, and does that fit your region's requirements?
Ask directly. A vague answer is itself an answer.
What happens when the AI is not confident?
Does it guess, or hand off to a person with context attached?
Ask for a live demo in the actual dialect and channel mix your customers use, not the vendor's default script.
See it hold one conversation, across channels and across languages
The fastest way to test any omnichannel or multilingual claim is to watch it handle a real conversation that switches both. Start Your Free Trial, no credit card, and send AWX a message that starts in Arabic on WhatsApp and finishes in English by email. Watch what it remembers.
Frequently asked questions
What is the difference between omnichannel and multichannel AI?
Multichannel means a business is present on several channels. Omnichannel means those channels share one memory, so a customer's history and context carry over automatically, without them repeating themselves.
Do AI language models understand every language equally well?
No. Performance varies by language and by how much training data exists for it, and research shows many models internally reason in English even when replying in another language (Zhao et al., NeurIPS 2024). Ask any vendor for specifics, not a language list.
Is a multilingual AI chatbot the same as a conversational AI agent?
Not necessarily. A multilingual chatbot may translate a reply without acting on the request behind it. A conversational AI agent checks policy and account data, then completes the task, a refund, a booking change, a claim update- in the customer's language, rather than just replying in it.
Why does multilingual support matter more for travel and airlines?
Because travel disruptions are time-sensitive and public, and travelers increasingly expect to manage bookings, delays, and rebooking directly with the airline, in their own language, across whichever channel they are already using (IATA, 2025).
