Why slot-filling chatbots will never meet human expectations

slot-filing-chatbots

If you believe the hype, building chatbots is easy.

After all, there are a lot of them out there. Facebook, for instance, claimed 100,000 unique bots for Messenger back in April 2017.  But despite their number, many of them are failing to impress, leaving users underwhelmed and frustrated. To understand why chatbots have not lived up to their potential it’s important to consider the difference between slot-filling and delivering conversational intelligence.

Bots may work fine if customers are prepared to follow a linear, prescribed way of engaging with them. Answering the questions and providing information in exactly the way that the bot developer intended. But unfortunately, as humans we tend not to do this—we communicate using natural language. We branch off at tangents, we circle back, we miss out crucial facts and figures, we ask for clarifications.  It’s this type of interaction that engages audiences, and delivers the most value to your business and your customer.

If you don’t know much about bot conversations, or indeed building any conversational UI, let me explain a little about how the majority of “AI” development tools used to build chatbots work.

In order to carry out a command, the bot first needs to understand the intent, such as I want a pizza. From there your typical bot will ask what type, how many and so on. We call this slot filling. It’s a fairly basic conversation and quite laborious for the user.

If the user deviates from the planned script by asking the time—a perfectly logical question if you’re trying to plan your evening ahead—most bots can’t cope. They don’t understand the entire conversation enough to answer the question and then bring the user back on track to the original intent. Nor are they able to realise when the original intent has changed. For instance if a mobile phone customer started out querying a bill, but then decided they actually needed to upgrade their contract.

In contrast, Natural Language Interaction (NLI) allows the user to ask in just the same way they might over the phone.  I want three margarita pizzas to collect in 20 minutes. It’s up to the bot to work out the details. At the same time it can cope with interruptions about delivery or credit card acceptance and the go back to the order process. This style of interaction delivers a much smoother and user friendly experience.  But it is one of the things that most bot development tools don’t provide.

Building a system that understands complex dialogues is difficult. Even interpreting the simple responses from a user such as “What’s available?” The bot needs to be able to understand that it’s with reference to What type of pizza and when the user says it again, that it is now with reference to the collect or deliver part of the conversation.

Teneo from Artificial Solutions is different because it allows dialogue to be managed out of the box through an easy to use graphical UI that supports a collaborative development environment. It is this dialogue management feature that allows for the user to deviate from the initial conversation and then return.

In addition, Teneo Language Resources and machine learning provides the humanlike understanding of complex sentences. The unique hybrid way in which Teneo enables the development of dialogue management removes the need for the vast amounts of training data and leads to significantly shorter development timescales.

But frustrating customers with slot-filling conversations shouldn’t be a business’s only concern when looking at chatbot development.  A lack of data ownership is also an issue, potentially your largest one. Sat in the middle between your app and your customer, the technology provider knows everything your users say.

A customer may have ordered 3 deep crust margarita pizzas to be collected at 7.30pm, but that is all you know. Other information, such as asking for toppings you don’t offer are lost. Or at least, they are to you. The chances are high that the owner of the chatbot technology knows exactly what your customer asked.

It might be for legitimate reasons such as developing the technology, but it could equally be used to sell on to third parties or for internal systems serving adverts. Either way, these companies are making money out of your customer data that could be used to improve and increase your bottom line.

With Teneo, you own the data. We even provide the tools to help you mine it for vital insight and turn it into actionable business data. If strawberries and cream pizza became the latest craze, you’d be one of the first to know. Not read about it a few days later in your newsfeed.

If you’re betting your business’ future on AI then you require enterprise strength software built for enterprise use with features you’d expect as standard such as roll-back, multi-language support and scalability. It’s surprising how often these features aren’t available along with support guarantees and product roadmaps, even in the paid for software. But of course many of them are still being paid in a round-about way with the data generated by your customers.

However, more importantly, it’s a point solution. You can’t take the application and port it easily to other channels. You can’t expand your investment to other areas of your business. It doesn’t provide a platform to use natural language to further your digital transformation throughout the company.

Finally, slot filling chatbots don’t deliver a true conversational interaction, one that delivers contextual understanding that is consistent across different channels, in any language. Features that are essential for a great customer experience and to increase brand engagement, because you never know what a human will ask next…

“Oh! Can I get extra pineapple with that?”


The Future of Customer Engagement and the Role of the Intelligent Digital Assistant

Leave a Reply

Your email address will not be published.

top