From switchboard to AI assistant: a short history of answering the customer

Alexander Graham Bell's United States patent 174,465 was filed on 14 February 1876 and granted on 7 March 1876. Its title was not about speech at all: it was called an improvement in telegraphy.

A telephone on its own connected two fixed points. What made it useful to a business was the exchange. The first commercial telephone exchange opened in New Haven, Connecticut, on 28 January 1878, with 21 subscribers. Its switchboard could connect up to 64 customers, but only two conversations at once.

That limit is the first lesson in this history. The new channel worked; the capacity to handle what came through it did not keep up.

Exchanges needed operators. In September 1878 Emma Nutt began work at a telephone company in Boston and is widely recorded as the first woman telephone operator. The role she represents, a person whose whole job is to connect callers to the right place, is the ancestor of every front desk and reception team since.

Operators made the system feel personal, and they were also its constraint. Every call waited for a human to plug it in.

Almon B. Strowger's patent for an automatic telephone exchange, filed in 1889, was granted on 10 March 1891. Automatic switching let a caller reach a number without an operator in the middle.

Britain moved more slowly. The Post Office took over the National Telephone Company in 1912, and manual exchanges lasted for decades. Telephone historians record the last manual exchange in England closing at Abingdon in June 1975, and the last in the UK at Portree on the Isle of Skye in October 1976.

Automation removed the operator from connecting the call. It did nothing about whether anyone picked up at the other end.

From the 1950s, answering machines began to appear in British offices, in a market licensed by the Post Office; in the United States, consumer models spread in the 1970s. For the first time a business could capture a call when nobody was free.

Capturing is not answering. A message on a tape is only as good as the person who plays it back and calls the customer. This is the gap that still costs businesses enquiries today.

Ray Tomlinson created a networked email system at BBN in 1971 and introduced the @ sign in addresses. The HTML form, which made the website contact form possible, was standardised in RFC 1866 in November 1995. LivePerson told the US Securities and Exchange Commission that it began offering its live chat service in November 1998.

Each channel made it easier for a customer to get in touch. Each also added another inbox. By the 2000s a business might receive enquiries by phone, email, form and chat, with no single place showing which had been answered.

Joseph Weizenbaum described ELIZA, a program that held text conversations by pattern matching, in Communications of the ACM in January 1966. It is usually called the first chatbot. It understood nothing, and Weizenbaum was struck by how readily people treated it as if it did.

OpenAI introduced ChatGPT on 30 November 2022. Large language models made it practical for software to read an enquiry, understand what is being asked, and draft a sensible reply in seconds.

Every step made arrival faster: the exchange, automatic switching, the answering machine, email, the form, live chat. Few made ownership clearer. The recurring failure is not that a message cannot reach a business; it is that once it arrives, nobody is sure who is dealing with it.

AI can now draft the reply. That makes the old questions more important, not less: who owns this enquiry, who approves what is sent, and how do we know it was followed up? A useful AI system answers those questions by design, keeps a record of every action, and leaves a person in charge where a mistake would matter.