July, 2026
In 2001, when I launched the small digital agency called Lava that later became Wolf Interactive, I had what felt like a bold idea. We would make our own website multilingual.
The internet was young enough that this felt exciting. A small Australian business could present itself to a wider market, and digital was becoming more than an online brochure.
In practice, the idea was also slightly mad.
Money was tight. Every change meant involving a translation specialist, updating duplicated content and spending more time than the budget justified.
The ambition was good. The operating model was not.
The multilingual website did not last.
That experience is one reason I find the current multilingual AI moment so interesting. A capability that once required translation providers, duplicated content and considerable patience can now appear inside a chatbot or live-chat interaction almost instantly.
That is genuine progress.
But the operating-model question has not gone away. It has simply moved further into the service.
When the conversation gets ahead of the service
AI can make the conversation multilingual before the organisation itself is ready to provide a multilingual service.
A customer can ask a question in Vietnamese, Arabic, Mandarin, Cantonese or Punjabi and receive a fluent response. A live-chat platform can translate both sides of an exchange. A contact-centre agent can receive an English summary.
For straightforward interactions, this can be very useful. Opening hours, appointment times, order status and simple service information are sensible places to begin. The consequence of a minor translation issue is usually limited and recoverable.
The difficulty starts when a fluent conversation is treated as evidence that the whole service is now multilingual.
Most services do not begin and end inside one chatbot interaction. They move through portals, forms, documents, payments, records, contact centres and human escalation.
A chatbot might explain a hardship process clearly, only to send the customer to an English-only application. It may describe an insurance claim accurately while the policy wording and dispute process remain difficult to understand. Or it may hand the case to an employee who receives a short English summary with much of the context removed.
I think of this as the channel cliff: the point where translated guidance ends and the customer is expected to continue through a service that has not changed with it.
The conversation has improved. The customer’s ability to finish the task may not have.
The failure often happens after the chatbot
The risk is not necessarily in the translated conversation. It is what happens next.
This is where performance reporting can become a little too reassuring.
Chatbot use is increasing. More languages are being selected. Containment is improving. The numbers look tidy in a dashboard.
All of that may be true.
But containment is not completion.
A closed conversation does not tell us whether the customer understood the next step, completed the application, challenged a decision or received the support they needed.
When the journey breaks after the chatbot, the organisation may simply move the cost into repeat contact, manual assistance, complaints or rework.
Low demand for interpreters can also be misleading. It may mean formal support is not needed. It may equally mean customers have stopped asking.
People often find another way through. A child helps a parent navigate a portal. A grandchild explains an aged-care form. A community worker walks another household through the same process for the third time that week.
That support can be generous and practical.
It should not quietly become the service model.
If an essential service depends on a teenager interpreting a financial, legal or health obligation, the organisation has transferred the risk to the household.
The real test is service completion
The first question most organisations ask is:
What languages do our customers speak?
It is a reasonable starting point, but it does not tell us whether the service is working.
A more useful question is:
What were customers who speak those languages actually able to complete?
That shifts the conversation from language availability to service outcomes.
Take one important journey, such as hardship support, eligibility, a complaint or an essential-service interruption, and follow it from the first multilingual interaction to the final result.
Where does English reappear? Where does context disappear during a hand-off? What does the customer need to understand, agree to or challenge? Can the organisation reconstruct what was explained and acted on if the outcome is disputed later?
Not every interaction carries the same consequence.
Checking an appointment time is different from agreeing to a payment arrangement. Asking where to upload a document is different from consenting to care or accepting a legal obligation.
AI translation is particularly useful when the consequence is low and any error can be corrected easily.
More care is needed when the journey involves money, access, safety, consent, dispute or essential support. In those situations, translation quality, human escalation, record keeping and accountability need to be part of the service design.
A practical place to begin
This does not mean every document, channel and service needs to be translated immediately. Most organisations will need to sequence the work.
Start with one consequential journey and identify where multilingual support begins and ends. Look at the customer outcome, not only the interaction.
Then decide which parts can sensibly rely on AI translation, where human language support is required and what information needs to travel with the customer between channels.
The goal is not a perfect multilingual service overnight.
It is to avoid creating a polished front door that leads customers into the same barriers that were already there.
AI has made multilingual interaction faster and far more affordable. That creates an important opportunity for organisations serving increasingly diverse communities.
The risk is mistaking a translated conversation for an inclusive service.
The practical test is whether the customer can understand, act, challenge and finish.
That is the point where multilingual capability becomes service capability.
Executive note for leaders
Before approving multilingual AI in a customer channel, ask what the customer can actually complete.
Follow the journey beyond the chatbot into the form, portal, hand-off, record, payment, escalation and final outcome.
Find where English reappears, where context is lost and where someone outside the organisation is quietly expected to finish the task.
A fluent conversation proves the front door can speak.
A completed service shows the organisation can support the customer through to the outcome.
Subscribe
You can subscribe for free. Every edition is delivered directly to your inbox and published here on Substack.
If a piece raises a question, surfaces a pattern, or helps you think more clearly about a decision, I’d value the conversation.
Thanks for reading,
Stuart Gonsal MAICD
With occasional help from Springsteen, my Border Collie, who reminds me that clarity comes from movement 🐾.
Connect
LinkedIn – Follow for practical leadership on AI-era opportunity. ↗
Disclaimer
Everything shared in The Ripple Effect reflects my personal views and does not reflect those of my current or past employers, clients or partners. Any examples are illustrative, drawn from publicly known patterns or anonymised experience.