Let’s be honest: Chatbots kind of suck (for now)

Let’s be honest. Chatbots, in their current state, aren’t that great.

They’re basically nothing more than glorified IVR systems. A viable chatbot solution without shortcomings doesn’t exist today, and the chatbots that exist are far from self-sufficient. The various solutions are either too simple to add any real value or too difficult to maintain. Without the development of an efficient, low-maintenance and easily integrable solution, chatbots will have limited impact on the customer service industry. Until then, chatbots won’t learn to crawl, let alone walk. And forget about running!

Rules-based chatbots suck

Most of the chatbots consumers interact with today are considered rules-based chatbots. These chatbots depend on a database of finite information that is coded and curated by humans. The chatbot looks for specific keywords in an interaction, scans through data, and spits back the most appropriate response based on the keywords they recognized.

An example would be a banking chatbot. You ask for your balance, it retrieves it from a database and sends it to you. This isn’t new ground-breaking technology, though. IVRs have been routing calls based on a simple prompt for years. The experience is great and efficient — as long as the answer is readily available. If a customer has a unique request that the chatbot cannot understand, it will respond with an error message or advise the customer to call support.

The weakness of rules-based chatbots is that they’re limited by the information to which they have access. They can’t determine if the response they’re serving is out of date or inaccurate. The certainty of the data depends entirely on humans, so developing a well-built database is essential.

The curation of a complex database of prompts and responses from which a chatbot can work is a perpetual task. The best available solution is contracting a third party to do the dirty work, but that’s never as good as using your own team, and it’s expensive. Companies also have to consider that the database will need to be refreshed and updated over time. As of now, there’s not an efficient solution that will continually “teach” the bot new information without some manual interference.

Curated chatbots also suck

There are chatbot solutions available today that compensate for the frustration of an error message by passing the inquiry on to a customer service agent. When the chatbot becomes “stuck” or unable to understand the request, it informs the customer that the interaction will be responded to by an agent shortly. Thus it functions more like a ticketing system (think Zendesk) and less like a conversation in real time.

That completely defeats the instant gratification one would expect from a chatbot interaction. It’s not efficient for the customer or the agent, and the customer experience becomes incredibly impersonal and robotic. It feels essentially no different than waiting on hold after being transferred.

The ideal chatbot solution would be so seamless that a customer wouldn’t be able to determine whether they’re talking to a human or a chatbot. The handoff would be uninterrupted and the interaction easily passed to an agent. All prior details of the conversation would be immediately available, allowing the agent to pick up right where the chatbot left off. A smooth transfer is essential to maintain a personalized experience for the customer. The ultimate goal is to smudge the line between human and chatbot just enough that customers can’t determine which they’re speaking to — nor will they care.

It’s not out of the realm of possibility for someone to create an “out of the box” chatbot solution that can be installed quickly, integrates well into existing workflows, and is affordable. Once chatbots are uncomplicated and easily maintained, we’ll begin to see them live up to the hype of completely changing the way we interact with businesses.

However, we have to keep in mind that their abilities are dependent on us. If they’re going to be a practical resource in the real world, we can’t keep limiting their functionality because we’re OK with solutions that are only “good enough.”