A beginner’s guide to chatbots

One of the biggest software growth areas in recent years has involved bots – automated software applications programmed to conduct repetitive tasks to save people doing them.

Over the decades, bots have come in numerous forms, from the web crawlers powering Google and Bing to bots capable of navigating torturous Guitar Hero chord progressions.

The best-known example of bots in action is the chatbots found on many websites and mobile apps.

Like an interactive FAQ or Help section, they aim to resolve recurring issues and answer frequent queries, yet chatbots have become distrusted rather than respected among the public.

In this beginner’s guide to chatbots, we explain why they’ve become so ubiquitous, and how to handle a bot that isn’t achieving the desired results…

Full of chat

Chatbots are often deployed as a first line of customer service and support in companies with limited customer service staff (or firms who simply want to save money).

These software programs follow a pre-programmed flowchart from initial enquiry to final resolution, which may involve handing over to a real person.

Often anthropomorphised with human names and photos of attractive people, their assistance may be limited to displaying basic data like opening times, product descriptions or returns policies.

Some are restricted to clickable buttons on pre-selected topics the bot is equipped to handle, though many chatbots use large language models for more natural communication.

LLM chatbots accept open-ended input strings – “I wish to register a complaint” would be as valid as “I’m complaining about your delivery service”.

Unlike humans, chatbots ignore filler words. In the above examples, the word “complain” would probably be the only term deemed relevant.

A pre-programmed algorithm would instruct the chatbot to respond to any variant of the word ‘complain’ with a set response.

“I’m sorry to hear you have a complaint,” it might be instructed to reply. “Please tell me more about the reason you’re unhappy”.

The next input string will also search for key words – late, refund, damaged, and so forth. Its algorithm should recognise similes, such as ‘delayed’ instead of ‘late’.

It may be programmed to offer a set apology or compensation, though it’s more likely to refer the customer onto a real agent.

Some bots have machine learning intelligence, enabling them to learn from past interactions and incorporate these results into future conversations, though ML chatbots are expensive and complex.

Specialist companies will build LLM chatbots for client firms, helping to determine what customer issues the bot should tackle and how its answers ought to be presented.

ChatGPT is an AI chatbot with a far broader scope but the same LLM underpinnings.

I like quick bots and I cannot lie

A key benefit of chatbots ought to be speed, yet many are programmed to respond as slowly as a real person might, to replicate the sense of talking to a human.

One exception is the First Direct banking chatbot, which bombards you with replies and information almost instantly.

Chatbots can handle numerous separate conversations simultaneously, freeing up staff resources by answering mundane or common questions.

However, it requires careful programming to ensure the bot doesn’t provide irrelevant responses or misunderstand an enquiry, which will rile up many consumers.

Bots should summon human assistance if someone is upset or confused, rather than continuing to plough the wrong furrow.

(There’s a jarring handover as a pretend human hands over to a real one, though it’s a necessary step if the bot isn’t successfully resolving a customer’s issue.)

Chatbots aren’t emotive, so harsh language or sarcasm will be wasted on them.

If you’re not making progress, state you want to speak to a person, and the bot should hand over to the next available agent.

They’ll view the bot’s conversation before speaking to you, so remain respectful at all times.

Chatbots have undoubted merit in an age where messaging apps are more popular than social media, yet many companies place too much trust in them.

There’s usually another way to contact a business – by email, telephone, text message or even in writing – if your enquiry is non-urgent and their website doesn’t provide the answer.