Chatbots: between the hype and the first practical use
In 2016 everyone is talking about chatbots. Here is where they actually work and where they are a marketing promise.
This year the chatbot topic has become almost unavoidable. Facebook opened Messenger to bots in the spring, large companies are announcing conversational interface strategies, consultants are promising customer service automation. The noise is significant.
I approach this with caution - not because the technology is useless, but because the gap between what marketing promises and what actually works today is very large.
What chatbots do well right now
Looking honestly, chatbots at today's level do one thing well: routing and initial classification of requests.
A user writes a message. The bot determines the topic - technical question, complaint, document request - and directs it to the right place: an operator with the right profile, the relevant FAQ section, the right form. This is unimpressive as a demo, but genuinely useful as an operational layer.
Another working scenario is structured dialogues with a limited number of options. Booking an appointment, tracking an order status, collecting simple data according to a known form. If the dialogue tree is bounded and well-described, the bot is reliable.
Where the promises diverge from reality
Free-form conversation in natural language on complex topics is what demos show beautifully but what breaks quickly in production. Users phrase questions unpredictably, context is lost, the system mishandles ambiguity - and the conversation hits a dead end.
A separate problem: companies look at a chatbot as a way to remove humans from the process and cut costs. In short timeframes this rarely works: the bot needs training, maintenance, and updates when processes change. It often costs more than it seemed.
The most honest assessment: a chatbot is not a replacement for an agent. It is a tool that handles simple repetitive tasks and frees agents for complex cases.
The organisational trap
I have seen several projects where a chatbot was conceived as a "smart assistant", and what resulted was another communication channel that nobody maintains.
A chatbot is a product. It requires an owner who monitors which requests arrive, where the bot makes mistakes, how user behaviour changes. Without this it degrades.
How to assess whether you need a chatbot
A few questions that help check whether an investment makes sense:
- What share of incoming support or service requests are repetitive standard queries with a predictable answer?
- Do we have data on what users ask - a history of requests, conversation scripts?
- Who will be responsible for the bot as a product - who will improve and update it?
- What will the handoff to a live agent look like, and how friction-free is it?
If the first question gives the answer "many" and the rest have answers - a pilot makes sense. If not, it is worth confirming first that the problem actually exists and that a bot specifically solves it.