The chatbot your customers meet is one your board has never used
A courier's WhatsApp bot could not hear a simple question. I suspect nobody senior has ever been on the receiving end.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
On 26 October 2026, at 09:43, a WhatsApp message told me a parcel was out for delivery to my home and invited me to type MENU. I was not expecting a parcel, so I asked what it was. Six seconds later, the bot replied that it had "received an invalid response" and asked me to "select one of the options provided above". There were no options above. I tried again and received the same answer. At 11:12, the courier, RAM Hand-to-Hand Couriers, told me the delivery had failed because the contact was not available, although I had replied in its own chat that morning. Two minutes later the identical message arrived again. I name the company because the exchange is real, not because RAM is unusual. In my experience, it is typical, and it raises a question every board should be able to answer: has anyone at the top ever been a customer of their own channel?
CONTEXT AND BACKGROUND
For many South African companies, WhatsApp has become the new front counter. Meta's Business Agent, unveiled in June 2026, can answer questions, book appointments and complete sales in a customer's own language, and more than a million businesses use Business Agents on WhatsApp and Messenger. The business decides when a human takes over.
The customer's experience has not kept pace with the deployment. University of Minnesota researchers note that between 53% and 77% of respondents across several surveys report a bad or frustrating chatbot experience, and that some customers have resorted to repeating phrases such as "chicken nuggets" until a system hands them to a person. In a poll of 1,001 American adults by Parloa, a company that sells AI customer service agents, 43.9% of those who had tried to get around a chatbot had shouted "human" or "person". The most common complaint, ranked first by 25.9%, was a bot that does not understand.
INSIGHT AND ANALYSIS
The RAM exchange is instructive because nothing in it was broken. The bot did what it was built to do: it accepted typed replies and discarded any that did not match its menu. The answer existed: the failure notice that arrived later named the parcel's sender. The system had not been designed to give it to the person asking.
[Insert RAM-chat-screenshot-2.png here, then delete this line]
The failure notice arrived twice, two minutes apart. Personal details redacted.
A design like this survives because nobody with authority experiences it. I would wager that no member of RAM's board has received this message on a personal phone, and I suspect few boards have received their own. The 2026 South African Customer Experience report found that only 21% of businesses surveyed have a formal policy on disclosing customer-facing AI, while 85% of consumers want to know when they are dealing with a machine. About one in four consumers turns first to online search for help, yet only 7% of businesses recognise that behaviour.
There is a second cost that rarely reaches the boardroom. An unexpected parcel message that tells you to type a keyword looks like a scam. In August 2026, Singapore's police reported at least 251 cases of fake courier messages that asked victims to reply "Y" or "1", a step that made a phishing link clickable. Malaysia's postal industry body has warned consumers not to pay for parcels they did not order. My refusal to play along was the correct security reflex. The bot treated it as an error.
The strongest counter-argument is that this was an old-style menu bot, not artificial intelligence, and that generative agents will understand a question like mine. That is fair, and it shifts the risk rather than removing it. An agent that understands free text can also answer confidently and wrongly. Either way, someone must decide where the machine stops and a person begins, and that decision belongs to leadership.
IMPLICATIONS
The price of getting this wrong is mostly invisible. In a Qualtrics survey of more than 20,000 consumers in 14 countries, nearly 1 in 5 who used AI for customer service reported no benefit, almost four times the failure rate of AI use in general, and 34% said they reduce spending after a negative experience. Customers who leave quietly never appear in a complaints register.
For boards, the remedy is personal before it is technical. Directors and executives should be customers of their own automated channels, on their own phones and without warning the team. Status codes such as "contact not available" should be audited against what happened, because boards read those records as fact. Section 54 of the Consumer Protection Act entitles consumers to timely performance of services and timely notice of any unavoidable delay, and a channel that cannot hear the customer struggles to meet that standard.
CLOSING TAKEAWAY
The courier's slogan is "From A-to-B, seamlessly". The gap between that promise and my morning is not a technology problem. It is a distance problem: the people who approve a channel are rarely the people who receive it. Before asking about models, vendors or savings, every organisation should ask one question. When did our chief executive last use this as a customer? If nobody can answer, the board is governing a dashboard rather than an experience, and its customers already know the difference.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence's impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan's work at https://www.aiforbusiness.net
Comments