What happened
Telstra launched its virtual support assistant 'Codi' in October 2017, billing it as artificial intelligence that had already engaged 'hundreds of thousands' of customers. Within months, customers were blasting it on Telstra's own social media channels as a 'virtual moron-idiot', 'a complete and utter waste of my time', and 'the worst idea in the history of customer support'.
Codi failed at the basics. Asked for a customer's name, it could not distinguish 'Paris' the person from Paris the city, and instead of transferring to a human operator it offered information on international roaming; the customer had asked for a human 15 times without success. The most common complaint was the bot failing to understand a question and getting stuck in a loop; it also appeared to crash, repeating 'Oops… something went wrong. Can you please retype your message?' multiple times.
The backlash landed just as Gartner was predicting a quarter of customer service operations would use virtual customer assistants by 2020, with adopters reporting up to 70% reductions in inquiries. Telstra's version delivered the opposite of the promised experience — the bot added a layer between customers and humans that had to be defeated before help arrived.
Why it happened
Codi lacked the core competency the job required: recognising when it had not understood and handing off to a human.
Intent-matching failures — a name read as a city — produced absurd routing that customers screenshotted and shared.
Deploying the bot on public social channels meant every failure became visible marketing for the company's worst support moments.
The lesson
A support chatbot must fail gracefully to a human: when escalation is the hardest task, every misunderstanding becomes a public protest on the brand's own page.
Aftermath
The episode became a widely cited example of chatbot overreach in customer service, arriving at the precise moment the industry was selling virtual assistants as the future of support operations.
FOLLOW THE EVIDENCE