There’s a story this week that has sparked debate about AI applied to business operations, and it’s worth taking a close look because it touches several important issues: customer service, automation, transparency, and the increasingly blurry line between efficiency and manipulation.
Telus is using artificial intelligence to alter the accents of its call center agents in real time through its Telus Digital subsidiary, and the reaction has been immediate. Labor groups are calling the practice deceptive, and the Canadian government is already considering requiring customers to be told when this technology is being used.
Let’s take it step by step, because the case is more interesting than it first appears.
What Telus is actually doing
Telus Digital, the Telus Corp.-owned division responsible for customer experience and call centers, has deployed AI technology that alters customer service agents’ accents.
The technology is supplied by a company called Tomato.ai. When a call center agent speaks, their audio is routed through Tomato.ai’s cloud AI model, which instantly processes the speech stream to adjust pronunciation patterns and reduce heavy accents. The enhanced voice stream is then sent to the customer, making the speaker sound clearer and easier to understand — while still retaining their own voice.
Technically, it’s not magic. The AI first encodes the speaker’s voice into a high-dimensional representation that captures both linguistic content and vocal characteristics. It then modifies only the pronunciation-related features before decoding the speech back into audio.
Telus’s argument is straightforward: reduce what it calls accent-related friction in calls handled by offshore agents, mainly in the Philippines and India.
Why the controversy has blown up
This is where things get complicated. There are three open fronts:
1. Transparency with customers. At a hearing before the parliamentary committee on industry and technology, Unifor telecommunications director Roch Leblanc asked the government to require companies to tell Canadians when AI is being used. He told MPs that the union knew of at least one of the three major telecom companies using AI to mask offshore agents’ accents, "altering how customers perceive who they are speaking with".
2. Jobs and invisible offshoring. Union leaders are concerned that accent masking and similar AI applications could make offshoring less visible, contributing to further cuts in domestic customer service roles. According to the union, around 20,000 telecom jobs have been lost over the past 10 to 15 years due to automation and offshoring.
3. Worker identity. United Steelworkers Local 1944 president Michael Phillips said he was aware that Telus is using the technology internally, with agents based in both Canada and abroad. A Telus employee in B.C. told him they had spoken with an agent in the Philippines. According to that employee, the overseas agent was laughing while turning the accent masker on and off — anecdotal, yes, but it makes the point.
The rest of the industry’s response has been telling: Rogers and Bell have said they have no plans to follow the same path.
The case for it (which also exists)
It’s not all black and white. Telus Digital defends the technology with data:
A TELUS Digital survey found that 55% of respondents believe that "nothing justifies a bad customer experience," and 54% would rather be stuck in a slow traffic jam than endure a bad customer experience.
And the use case makes sense in the abstract: speech enhancement technology allows global brands to expand their contact center operations to previously underserved locations, dramatically increasing the size of the available talent pool. This lays the foundation for organizations to provide genuine human support across multiple time zones.
There is also a worker-protection argument: the software, built by Tomato.ai, is designed to reduce what Telus describes as "accent friction." The company says it also helps protect staff from harassment.
That is a real point. Anyone who has worked in phone support knows accent-based harassment exists.
What this has to do with your business
Now for the part that matters if you’re reading this as an SMB or mid-sized company. Because even if you don’t have 10,000 offshore agents, this case tells you very concrete things about how to automate — and how not to.
Lesson 1: Transparency is not optional, it’s strategic
The problem with Telus is not the technology. It’s that they deployed it without clearly telling anyone. Labor groups have criticized Telus for what they see as a deceptive practice, demanding mandatory disclosure to customers about the use of such technology. This criticism reflects broader concerns about transparency and consent in the deployment of AI in customer service roles.
When you automate something that affects your relationship with the customer — whether it’s a voice bot, an email response system, or a chatbot — you have two options:
- Say it openly. “This call is being handled by our automated assistant.” “This email was routed automatically.”
- Hide it and hope nobody notices.
Option 2 works until it doesn’t. And when it stops working, it’s not the automation that breaks — it’s trust.
Lesson 2: Fixing symptoms is not the same as fixing problems
The accent is not Telus’s real problem. The real problem is that they have offshored so much that the perceived quality of service drops. Accent AI is a bandage on a deeper operational wound.
This happens a lot with poorly designed automation. Someone says: "automate email replies so they’re faster". But the real problem may be:
- Too many emails are coming in because the website FAQ is poor.
- The team lacks clear prioritization rules.
- Processes are missing before technology is even introduced.
Automating without understanding the problem leads to solutions like Telus’s: technically impressive, strategically questionable.
Lesson 3: AI should empower humans, not disguise them
There’s a huge difference between these two approaches:
| Approach | What it does | Result |
|---|---|---|
| Disguise | Hides who or what is on the other end | Short-term efficiency, high reputational risk |
| Empower | Removes repetitive work so humans can focus where it matters | Sustainable efficiency, happier teams |
At Studio SmartWork, we build automations of the second kind. A bot that answers the phone after hours and schedules meetings isn’t “tricking” anyone — it’s covering a time window you wouldn’t otherwise serve. An inbox that prioritizes emails and drafts replies doesn’t replace your team — it gives them back two and a half hours a day to do work that requires judgment.
The line is clear: AI should remove real friction, not simulate a reality that doesn’t exist.
Lesson 4: If you can explain it, it’s probably okay
A simple heuristic we use when designing automations for clients: Could you explain exactly what this bot does to your end customer without making them feel misled?
If the answer is yes, go ahead. If the answer is “it depends how I frame it,” that’s a red flag.
A bot that says “Hi, I’m the virtual assistant for [company], and I’m here to help you schedule a call” — perfect.
A system that passes offshore agents off as local agents — now you’re operating in a gray zone.
The regulatory context is shifting
This isn’t just an opinion. Growing public scrutiny around Telus’s approach has triggered calls for regulatory intervention. There is increasing demand for Canadian regulators to establish clear guidelines on the use of voice-altering AI technologies in customer-facing roles.
The European Union has already started moving with the AI Act, and it’s reasonable to expect stricter disclosure requirements in any customer-AI interaction over the next 12 to 24 months. Companies that build automation on transparency won’t have to rebuild later. Those that bet on hiding the ball will.
How to automate customer service without getting into this mess
To close, here are three practical principles we apply when a client asks us to automate customer service:
1. Always identify yourself. The bot says it’s a bot. Full stop. Interestingly, customers usually accept that far better than companies expect — because they know they can escalate to a human if needed.
2. Design a clear escape route. Any automation that touches the end customer should have a “talk to a person” option within one or two steps. Not three more menus.
3. Measure satisfaction, not just efficiency. It’s easy to fall in love with time-saved metrics. But if NPS drops while you automate, you’ve optimized the wrong thing.
The bottom line
The Telus story is not about bad AI. It’s about AI applied without really thinking through where the line is. The technology they’re using is genuinely impressive — the issue is the business decision wrapped around it.
AI automation in 2026 is no longer a question of can you? — almost everything can be done. The useful question is should you? And only someone who understands the technology, the business, and the customer can answer that.
The kind of efficiency that lasts is the kind the customer can see and appreciate. Everything else eventually ends up on the front page.