We’ve all seen them, used them, complained about them – chatbots with infinite loops which you can’t escape from, chatbots which don’t seem to understand the simplest of requests, chatbots so limited in what they do, you still end up needing to call customer services every time.
For people who don’t work in the field, I know that these bad conversational AI experiences tend to make people doubt the technology is up to the job and think it’s only good for really simple asks. I also know first-hand from working with the tech that this simply isn’t the case.
Is it the technology?
The main limiting factor related to this technology in the last few years has been on voice CAI bots rather than text-based bots. Speech to Text (STT for short) can still struggle at times with poor quality audio, background noise or strong regional accents, although huge progress has been made on the latter, in particular, and the improved audio quality of smart speakers and other connected devices (vs. a telephone line) means this is becoming ever more accurate.
However, the quality of speech recognition was never a factor for classic web chatbots, which start off with text-based input anyway. So why aren’t they better? Another common misconception is that it is just down to the poor choice of conversational AI solution. Obviously, this can be a factor – not all conversational AI platforms are as capable and easy to implement and maintain as one another. Similarly, developing your own CAI solution seems to me a special blend of wheel reinvention and masochism, when there are some really good products on the market. But it’s far from the whole story. I’ve seen bad experiences created even with the very best platforms and good ones created with more limited tooling.
So why isn’t the average chatbot vastly better?
- Limited scope/lack of ambition: the scope of the chatbot is limited to a few FAQs – while the chatbot may work perfectly, its limited nature mean users will quickly realise it can’t do much for them and they’ll give up on it.
- Doesn’t handle the most common user asks: this may be a consequence of the first issue, but you could also add plenty of functionality into a chatbot in the wrong areas. The key is spending time upfront making sure you understand the high-volume use cases and doing some testing and tuning as part of the initial implementation once real users start interacting with the system.
- Failure to handle unexpected input: any chatbot available to the general public is going to run aground somewhere i.e. it will be asked something it doesn’t know – but the trick is in how this is handled
- Failure to handle interruptions: in the middle of a process (e.g. ordering something) a user may think of something else and go off at a tangent – again the trick is in handling this. The very best platform I’ve seen for this is Teneo where you can jump out to another topic, have that conversation, and then be offered a chance to resume the thing you interrupted.
- Failure to remember input you’ve already given: as a user, if I’ve mentioned a reference number, a phone number, an email address in anything I’ve already said to a bot and then later get asked for that information that’s frustrating. Only picking up one thing at a time is very limiting and not reflective of how people talk to one another.
For example, if I say, “I want to book a service for my car KN11 CGY” and the bot replies “what’s your registration number?” as it’s only taken the fact you want to book a service from that first statement, that’s frustrating. This issue is more easily dealt with by some CAI products than others, as is remembering that input when the user moves onto a different topic of conversation.
- Poor testing: will harm quality in any kind of digital transformation and doubly true in conversational AI. This is because the ‘happy path’ might only work for a subset of users who express themselves in a particular way and doesn’t handle how significant volumes of users articulate what they are looking for or trying to do.
So if your existing web chatbot or IVR (interactive voice response) suffer from some or all of these issues, or you’ve been afraid to try the technology because of ones you’ve seen, it’s time to seek some professional help. Here at Valcon’s CAI practice, we’re on a mission to deliver great conversational AI. Why not chat to us about a proof of concept or a pilot and see what great CAI could do for your business? I think you’ll be pleasantly surprised.
Anne Jenkins is a senior manager and CAI expert at Valcon. To reach out to Anne please email [email protected] or call +44 (0) 7397 149 619 today.