Chatbots are being hyped as the next big thing in marketing. They should, marketers insist, be an integral part not just of customer service, but of your entire content marketing strategy.
In fact, far beyond marketing, they have use cases in every department of your business, every customer touchpoint, and even B2B interfaces of your industry or company internal processes. However, just like any new technology, chatbots are somewhat overhyped and the C-suite has no clear picture of how useful they will be in the end.
Deploying a chatbot requires a lot of preparation not unlike the work involved in prepping a department for machine learning or artificial intelligence.
For a general overview of machine learning deployment, see: Machine Learning in the Organization: First Steps
But let’s get started.
The Low Hanging Fruit
The easiest applications for chatbots are without doubt high-structured, repetitive information such as that contained in standard customer service dialogues. Whether it is information about your opening hours, making simple reservations, finding the location of offices or getting a step-by-step guide through bureaucratic processes, this is information any chatbot can give in a reliable way.
“When will my order arrive”, “can I make an appointment on Thursday at 2 pm”, “is this item available in red”, “what is the balance on my credit card account” should all be easily programmable responses. Experts estimate that over 70% of calls pertaining to information contained in FAQs or customer service manuals can be easily and quickly transferred to a chatbot API. Agents are taking their time to answer questions, while customers with much more pressing requests are made to wait: here the chatbot has a clear use case and offers immediate benefits.
In other words, chatbots should be used to remove large volume questions with standard answers from the queue.
Pitfalls for Basic Chatbots
NLP or natural language processing is the key to even basic chatbots. Different people use different expressions for the same thing. Whether you want “double cheese”, “extra cheese”, “more cheese” should all be understood without having to confirm. There is nothing more annoying than having the chatbot ask you “did you mean…?”
The key here is to build very simple decision trees modeled on existing workflows, avoiding brand-specific lingo the customer may not be familiar with and making use of existing customer data. An airline chatbot is a good example.
If I have booked 20 flights from A to B in the same time slot and class over the last 5 months, the chatbot should be smart enough to look up my past reservation data and thus not offer me choices I have previously rejected, while being flexible enough to make a slightly different reservation this time.
If I have consistently ordered the vegetarian meal, this information should be available to the chatbot also.
Personalization is the Key
This is not personalization per se, because hundreds of customers will have made the same type of reservation. Customization of a flight is not extremely complicated. Other than the date, time, class, window or aisle and perhaps the choice of meal, there is not a lot that goes into a flight reservation.
The same is true for any chatbot handling retail orders – and of course the AI behind voice assistants like Alexa.
But here is where we get a little more nuanced. In the coming years, we will see more elaborate dialogs. Currently, chatbots deliver simple answers to simple questions but are unable to build on those fragmentary interactions to conduct human-like dialogs.
Multi-stage conversations are not yet handled satisfactorily by chatbots and can do more harm than good to your brand’s reputation. An annoyed user is not a happy user.
In the example of the airline reservation, if the customer is a frequent flyer and no seat is available on the flight they are requesting, the chatbot should immediately hand off to a human operator. If a customer consistently ordered the same type of washing powder for months or years, do not bombard them with alternative choices.
It Takes Time and Common Sense
Depend on your industry, deploying chatbots takes time. Estimate at least a year to get to over 70% of requests handled by the bot! In the initial phase, make sure bot interactions are monitored and failures identified early. There should always be a human operator at hand to jump in.
Retailers, airlines, even hospitals, have started to adopt chatbots in that way. The key for customer satisfaction here is the remember that the caller should be in a relaxed state and friendly towards the chatbot service.
Detection systems for natural language processing are now being developed that can tell whether a caller is angry, agitated, confused or in distress — in which case a human interlocutor is preferable than a chatbot.
Finally, there will be situations where are chatbot is not the right technology. The complaints hotline should not be the first place to launch a chatbot, as you can reasonably expect a level of stress or agitation.
Repeat purchases of standard goods are easier handled by a chatbot than complicated orders. In banking, new generations may trust computer-controlled systems more than older generations, but always keep in mind who your customer is and how familiar with the technology they are.
Think Voice For the Long Term
While chatbots on social media platforms or your website are easier to implement because of the prevalence of text parsing, the future belongs to voice.
3-5 years from now voice will be the key enabler of chatting, and your business should be ready for it. In the meantime, take baby steps and pay attention not to annoy your existing customer base.
Writing Your First Chatbot Script
Chatbots are the next big thing, even though current implementation is still sketchy. In some industries – such as airline customer service – they have taken off quickly, others have struggled to come up with the right content, brand voice, and implementation
The best place to start planning your marketing chatbot is yourself. Have you used a chatbot? Do you like using chatbots? What problems have you found using chatbots.
I am sure that you will agree a chatbot has to meet the following criteria to appeal to users:
- Responses have to be fast
- Responses have to be accurate
- Responses have to be natural
Remember the acronym: FAN – fast, accurate, natural.
Fast, because nobody likes to wait, especially on mobile. Make sure you have the right solution in place. Responses don’t have to be timed exactly, that would make the conversation feel unnatural. Anything between 1- 6 seconds is a good response time. The chatbot solution you choose probably has a feature called random or dynamic reply timing; use it.
Accurate because, well, what use is a chatbot that gives wrong answers. That means examining your data for accuracy and relevancy.
And finally, natural, because we feel more comfortable talking to real human beings, so the technology has to at least make an effort to sound like a real person. Chatbots are not Google search, they are not meant to deliver stilted answers, but to engage the customer into a real conversation. Or at least something that feels like a conversation.
Bearing in mind the above, the script you write for your chatbot should be based on actual conversations. For that, you need data – recordings of actual interaction with customers. Even without expensive data analysis, you will be able to extract some key elements that can guide you when writing the chatbot script.
Some experts have recommended using slang expressions and abbreviations, even emojis in replies. While this may give a more “authentic” feeling to the chatbot, it may not sit well with your brand voice. When I am making a reservation with a 5-star hotel, I don’t want the chatbot to call me “mate” or answer “yeah man!” Chatbot scripts have to match the brand and application they represent.
Make It Sound Human – For Everyone
If you do not have access to customer details at the time of interaction, remember that you are writing a script for all ages. The chatbot must feel equally professional and natural to a senior citizen as to a teenager. Sophisticated chatbot programs also allow you to program several choices for each response depending on the client persona (e.g. age of the user or type of service requested)
Even though human conversations can guide you when writing the chatbot responses, do not rely on one sample alone. It may be an outlier.
That said, even if you don’t use slang or emojis in your answers, your chatbot program should be able to understand them to a certain degree, especially if your customer profile includes younger generations. Natural language processing is a key feature of most chatbot solutions, but not all of them offer enough flexibility to accommodate the type of customers you are dealing with.
Chatbot scripts must be written from the perspective of the user. Always remember who you are talking to, and in which situation. A chatbot for ordering food or concert tickets is a different animal from a chatbot on a customer complaint website.
And finally, stay away from the canned responses pre-programming with many cloud-based chatbot solutions. Don’t even look at them. At this stage, AI is not general enough to offer solutions across the board, so you will have to program the chatbot with data befitting your organization, your customer profiles, and your use case.