Last year, a grieving air traveler asked Air Canada’s chatbot about bereavement fares. The robot invented a refund policy that didn’t exist. The customer acted accordingly, the airline ended up in court and the story went viral. The court rejected Air Canada’s argument that its chatbot was a “separate legal entity” responsible for its own actions and ordered the airline to pay damages.
This incident now serves as a warning for all brands scaling AI in their customer communications. And new research from customer communications platform Sinch suggests this is far from an isolated case.
According to one study, some 74% of companies have already been forced to remove a deployed AI agent due to governance failures. Fishingch, “AI Production Paradox” report. Here’s the problem: Companies with the most mature safeguards, those that invested most heavily in compliance, security protocols, and monitoring, fell back at an even higher rate of 81%. Teams that do the most to avoid failure fail more often, not less.
“If governance were the solution, more mature teams would back down less, not more,” said Daniel Morris, chief product officer at Sinch. “Engineering teams spend most of their time building and maintaining security systems instead of focusing on improving the customer experience. It’s the gate tax that slows organizations down.”
The impact of the guardrail tax
For marketing teams, this safeguard tax has a direct cost. Every hour engineering spends rebuilding security infrastructure is an hour not spent improving the revenue-driving customer experience.
Air Canada is not alone. A car dealership’s chatbot agreed to sell a Chevrolet Tahoe for $1 after a joke. A AI Support Bot at coding startup Cursor invented a nonexistent login policy, triggering a wave of customer cancellations. A the delivery company bot insulted a customer and wrote a poem trashing his own employer. Each incident went viral. Each damaged a brand. And each of them helps explain Sinch’s findings that three out of four companies have already canceled a deployed AI agent.
Sinch surveyed 2,527 business decision-makers across 10 countries and six industries. The findings that matter most to marketers:
- 62% of companies already have AI communications agents in production and 88% plan to deploy one within 12 months. The pressure to deploy is intense.
- 74% were forced to restore a deployed agent due to governance failures. Three out of four marketing organizations have already felt the consequences of an AI deployment that had to be rolled back.
- 84% of teams spend at least half of their engineering time rebuilding security infrastructure from scratch. This is an engineering capability that could be geared toward personalization, channel expansion, and campaign optimization.
- When an AI agent fails, 35% of the impact ends up in the support queue. Almost as much, 34%, depends on brand perception – and this one is harder to fix.
According to the study, infrastructure quality is the most important predictor of deployment success, surpassing model choice, team size and budget. Yet most organizations say their current vendor falls short of expectations in at least one critical area.
AI customer communications agents manage customer conversations at scale: chatbots on websites, voice agents in contact centers, automated text and email responders, and omnichannel platforms that route and respond across all channels. They range from simple FAQ bots to sophisticated agents that authenticate users, process transactions, and personalize responses based on customer history.
Sinch’s research tracks agents already in production, not pilots or internal experiments. These are systems that marketers rely on every day, where an outage means frustrated customers, longer wait times, and brand damage that spreads within minutes.
Choosing the wrong foundation is the real risk
Jayashree Iyangar, global head of CX data and AI at HGS, a digital experience company, said the results match what she sees on the ground. Marketers are past the pilot phase, she noted, and the real challenge lies in operations.
“The key question is how AI can be seamlessly orchestrated across multiple channels, not whether it can be deployed across just one,” Iyangar said.
She pointed out that the risk profile varies significantly depending on the use case. A marketing chatbot that manipulates a promotional offer carries less leverage than a service agent who mishandles a sensitive billing issue. “Human monitoring remains central in service environments where the risk of negative impact on customers is higher,” she said. “This is also where we are seeing more cases of AI rollback.”
Your customers are searching everywhere. Make sure your brand introduces himself.
The SEO toolkit you know, plus the AI visibility data you need.
Start free trial
Start with
His view on infrastructure echoes Sinch’s key findings. “A significant amount of effort goes into creating and maintaining security systems rather than improving the customer experience,” she said. It sees organizations consolidating around centralized AI governance teams that manage trust, compliance and security separately from the AI use cases themselves.
Three Steps Marketers Can Take Right Now
For marketing teams, the study highlights three practical steps.
- Let infrastructure guide your vendor decision. Infrastructure quality predicts deployment success more than any other variable in Sinch data. When evaluating vendors, ask about guardrail engineering, cross-channel orchestration, and the extent to which your team will absorb the security burden. The right platform handles the bulk of the security work, so your team can focus on the customer experience.
- Include the guardrail tax in your roadmap. Security systems are not a one-time installation cost. They consume ongoing engineering resources that would otherwise be devoted to CX improvements. Plan for this reality from the start rather than seeing your schedule slip when restorations arise.
- Push for a separate governance function. Iyangar’s observation about centralized AI governance teams directly aligns with the data. Separating AI use cases from governance engineering reduces overhead. Marketing should not own a security infrastructure. It should partner with a dedicated governance function that manages trust, compliance and security, allowing marketing to focus on work that directly impacts customers.
