Chatbots, are they the panacea for Customer Experience?

Published in

Becoming Human: Artificial Intelligence Magazine

15 min readAug 6, 2019

Chatbots have the potential to dramatically improve customer service, reduce the cost to serve, and provide vast amounts of detail about customer interactions. They are one of the hottest new technologies in improving the customer experience. But as with all technology, a bad implementation will result in a bad experience. In this article, I’ll discuss the keys to successfully using chatbots to improve, not destroy, your customers’ experience.

What is a Chatbot?

A chatbot is a computer programme that mimics a human in an instant messaging conversation. In plain English that means you seem to be chatting with a person, but it’s actually a computer. There’s nothing sinister about this, a well implemented chatbot is upfront about the fact that it’s a computer not a person. If you want to do something quick and don’t feel like talking to a human then a chatbot is a great way to get it done.

At home we use bottled gas to heat our water, big 45kg bottles that last two months but always seem to run out when you’re in the shower and the weather is freezing, so I’m usually in a hurry to order a new one. Being able to send the gas company chatbot an instant message on my phone to say “Help! Send me a new bottle now” is brilliant. Even better is making sure my wife and I don’t double up and both order a bottle, I can ask it “when did we last order a bottle?” to find out if she’s beaten me to it. Our previous gas provider had a nice website where I could order a bottle, but that meant going upstairs to the computer and remembering our account number.

Another great example is Air New Zealand’s chatbot called Oscar. I wanted to know how many bags I could take on a flight, it wasn’t one of the routes I normally take, it was a small plane, heading to a small airport and I’d purchased the ticket ages ago. I could have called their contact centre, the people there are lovely but I didn’t have ten minutes to go through the IVR, wait in the queue, and then explain what I wanted. I could have searched their website, but I wasn’t near a computer and I didn’t feel like trying to navigate it on my mobile. So, I sent Oscar a message, “how many bags can I take to Kerikeri?”. “He” asked a couple more questions about booking numbers, and then gave me the answer, 20 seconds, job done.

There’s a vast amount of technology behind a chatbot that makes it work, such as natural language recognition, machine learning, and artificial intelligence, but as with all technology the right question to ask isn’t “how does it work”, but “what can we use it for?”.

Should you use Chatbots?

Let’s be clear, chatbots are great for some things, but they are only as good as the existing customer service you provide. This is a key point, if your organisation provides a poor customer experience today, chatbots won’t fix it, they will probably make it worse. This seems to be contrary to all the hype about chatbots being the silver bullet to great customer experience, but consider this scenario:

A customer calls your contact centre asking for an update on their order, after a long wait in the queue, your agent asks for their order number and then gives them the status that it will be delivered tomorrow. So far so good, a chatbot would have eliminated the long wait and improved the experience. Then the client asks if delivery can be delayed so it is delivered at the same time as their two other orders. Now the wheels fall off. Your agent doesn’t know about the other two orders, they were from a different department, and your systems track order numbers, not customer names.

The customer is going to have poor customer experience because your systems aren’t customer centric, your focus isn’t on the customer it’s about tracking orders. Putting a chatbot in front of that isn’t going to fix it, if anything it will expose it.

“You have to start with the customer experience and work backwards to the technology” — Steve Jobs

Before you even consider using a chatbot you need to ensure you are ready for your customers to be interacting with you autonomously, by that I mean you won’t have the “human band-aid” factor protecting your clients from your systems and processes. Contact centre agents often have dual screens and multiple applications in front of them, they manually collate information from numerous sources to provide what appears to a seamless experience to a customer. If you want a chatbot to replicate this you will need to start integrating your systems.

Human, App or Bot?

There’s a lot of hype about bots replacing apps and contact centre agents, if you buy into the hype we’ll all be interacting just with bots within a few years. Here’s a few snippets of what the industry is predicting:

Bots replacing Apps

For the past decade, the drive has been to create mobile apps, “if you’re in business and you don’t have a mobile app then you won’t be in business long” or so the saying goes. This has led to a proliferation of apps, some great, some not so great. Recently there has been a significant drop-off in the uptake of apps, some predict this is the beginning of the end of apps, and the start of the rise of bots (chatbots, not terminators):

Some predict that bots are the new apps. Consumers are overwhelmed with apps, and they prefer to converse on the messenger apps they’re already on. If you think about it this is a natural progression, considering that messaging is the root of all mobile communication. 50% of U.S. mobile users haven’t downloaded an app in the last year, but billions of people are already using messaging apps. Platforms like Facebook Messenger, Kik, and WhatsApp have a combined user-base of more than one billion. Even just Kik alone has more than 300 million users who have exchanged more than 350 million messages with bots. — Blake Morgan, Forbes, March 2017:

(Blake’s article is really good, I’d recommend reading it if you are interested in chatbots.)

Bots let you use natural language to get tasks done. This is one of the reasons many people use Siri or Cortana to check the weather forecast, set a reminder or send an email: it’s just faster. As bots increase in their capabilities, we’ll start to use apps less. Right now, you probably flip between a few different apps to book a weekend away. It’s the same if you’re search for something you want to buy locally: you might go to a website, search for a product, check stock and then get directions in Google maps to show you how to get there. — Jim Martin, pcadvisor.co.uk, March 2017

Bots replacing Humans

Chatbots are beneficial for both parties: developing chatbots is cheaper than training and hiring human customer service agents for the company, and customers often prefer a brisk mobile interaction over talking with someone in person or with the call center. Consider this statistic from Gartner, that artificial intelligence will amount for 85% of customer relationshipsby 2020.

Chatbots are amazingly efficient for both users and brands — they cut costs for companies by winding down the expensive interactions with traditional call center agents or customer service representatives, and they also make it easier and more convenient for customers to communicate with brands, streamlining interactions. Most customers would rather get a quick recommendation and purchase an item on their mobile phone instead of sifting through online reviews and trying to buy products in person or by calling a representative. — Blake Morgan, Forbes, March 2017

Wait, Not so Fast

However not everyone is predicting that chatbots will be the answer for all, with a recent study finding that most people still prefer a real person:

PSFK’s recent study, The Future of Retail 2017, says that 83 percent of consumers want a live person on the phone to handle customer service issues and, specifically relevant to chat, most users abandon chatbots after two messages. — Taylor Jones, Customer Think, Feb 2017

So where does that leave us? Chatbots will become far more prevalent, but they wont be the answer to everything, we will still need a mixture of customer contact channels.

Activity Based Customer Experience

The future of customer experience isn’t a single channel, it’s a blend of multiple interaction models based on the activity the customer is performing. So the answer isn’t bots, apps or contact centres, it’s a combination of all three used to create a seamless experience. This isn’t a new idea, many views support the integrated channel approach that combines chatbots with existing channels:

While 2017 is being dubbed by many the “year of the chatbot,” if your customer care organization is looking to choose between man and machine, it’s asking the wrong question. Companies who are successful in chatbot implementations will keep seamless live agent service backing-up chatbot experiences to solve more complex customer care issues. Lose the personal touch and you may lose the customer too. — Taylor Jones, Customer Think, Feb 2017

The ideal customer service offering combines the power of bots and humans. The bottom line: When a customer reaches a point of frustration, an automated response just won’t do. There will inevitably be some situations that bots simply aren’t trained for. Businesses will have to choose the right use cases for automation and build in the right handovers, or escape hatches, to let customers talk to human operators when it’s sensible to do so. — Royston Tay, Zendesk, October 2016

Integrated Experience

Here’s how an integrated activity based customer experience might work:

I need to change my flight from Friday to Saturday, so I open the airline’s app on my phone, it has my credentials so I’m already logged in and I can see all my upcoming flights. An app makes sense for this as it serves many more functions than just customer service, it has virtual boarding passes allowing me to scan a QR code at the gate, it has maps to the airport, and shows which seats I can choose. A bot can’t provide that richness of information. I pick Friday’s flight and click the chat button to open a conversation with a chatbot.

I ask the chatbot “Please move my flight from Friday to the same time on Saturday” (funny how manners seem necessary even when talking to a machine). “Thanks Mike, you are now booked on flight XYZ836 at 1pm Saturday”, fantastic, the chatbot did exactly what I wanted, instantly.

But I also have a more complex requirement, “I need to take some extra bags and I want to use my wife’s air miles to pay for them”. “Okay Mike, I’ll get one of our people to help you with that. They will call you in under 2 minutes”. The chatbot realised this was beyond its capability and has put me in the contact centre queue for a callback.

Less than 2 minutes later my phone rings and I’m talking to a human, who has read the transcript of the chat session and has my wife’s air miles balance in front of them. We get the baggage sorted and end the call. Now the app on my phone has all the updated details including the extra bags and slightly reduced air miles balance. One more thing, a quick chat with the bot “Any chance of a window seat?”, “Sure Mike, you are now in seat 12A, enjoy your flight”.

The key to this example is the seamless handover between the app, chatbot, and human. It would be a very different story if I had to tell the chatbot which flight I wanted to change, or if I had to tell the contact centre agent who I was, what flight I was on, and what I wanted to do with my extra baggage. Multiple channels integrated as one can make for a fantastic customer experience, but if you fail to integrate them you will create frustration and drive customers to only use the richest channel, which is typically the contact centre.

When to use a Chatbot?

There are many factors to consider when deciding if the activity is suited to a chatbot, some of the key factors are:

Customer emotion — is it a sensitive interaction that requires a human to be aware of the situation? You probably shouldn’t use a chatbot for emergency services calls, suicide hotlines, or customer complaint departments.

Customer demographics — are your customers comfortable with technology, would they prefer a chatbot or a human? City based millennials might prefer a bot, but rural baby-boomers might like the opportunity to have a chat with Beryl at the contact centre about the weather while they get an update on their insurance claim (apologies to everyone for the horrendous over-generalisation and stereotyping there).

Complexity of interaction — is it a simple ask and respond query or something more complex? Chatbots are good at simple requests, don’t expect them to handle complexity yet.

Information type — can the answer that your customer is seeking be easily provided in text, or does it need a picture, a video, or a live update? It’s easy to use a chatbot to order an Uber “Uber — get me a car now”, “Thanks Mike, you Uber is on its way, it should be there in 10 minutes”. But I don’t want to have to keep asking how far away it is, it’s much easier to look at the Uber app and see where the car is on the map.

How Good are Bots?

What can a chatbot understand, how smart are they? It all depends on how much you spend, how much you train it, and how much data you can give it access to.

Like all things, there’s a spectrum from low-cost basic bots that can only handle simple scripted interactions, through to AI based bots that are contextually aware. It’s important to determine where on that spectrum you need to be, using a cheap simple bot to handle complex interactions will quickly fail, likewise investing in a highly-advanced bot to provide very simple tasks isn’t a great investment.

Context

One of the key differentiators along that spectrum of bot intelligence is the ability for the bot to understand and remember context. Think about this example:

Q: “Hi bot, how high is Mt Everest?”

A: “Mt Everest is 29029 feet”

Q: “What’s that in meters?”

A contextually aware bot will understand that the second question is related to the first question, and “that” in the question is referring to 29029 feet, so it could transform the question into “What’s 29029 feet in meters?”. For a human this is second-nature, but for a chatbot it’s a very advanced ability. Try it with Siri, Google, Cortana or Amazon Echo and see if they can do it, these are some of the smartest bots in the market and they can struggle with context.

Context also relates to facts that the bot should know about the situation from past interactions or other systems, such as:

Q: “Siri tell my wife I’m going to be home 10 minutes after the gym”.

Siri needs to know who my wife is, what her number is, and what time my gym appointment is for. To do this the chatbot (Siri in this example) needs to be integrated with other systems, for Siri this usually is the case (she has access to calendars and contacts on your phone). In a business setting this means your chatbot needs access to your CRM system, orders system, accounting system, and service desk system.

Training

You wouldn’t put someone on your customer service desk without training them first. The same applies to bots, they need to be trained before they are released into your business, and that training isn’t a one off task, its on ongoing exercise to constantly refine and improve your bot.

Ian Collins of CrowdCare outlines a five step programme to training chatbots:

Step 1: Pre-training your bot

Review your existing chat logs (if you use chat) and call recordings and look through your typical support material that contact centre agents use. Look for the top reasons your customers are contacting your company.

Step 2: Analyse all your customer interactions

A bot solution needs to have a tool to provide real time analysis of interactions, this will tell you what your customers are trying to do with your bot. You may have thought they just wanted to order a pizza, but actually you are getting a lot of questions about the nutritional information of your pizzas. This helps focus your training in the right places. This is key.

Step 3: Training the bot

Think about how you match the bot platform to the information it will provide. For example a text only bot needs more work to provide complex information, whereas a chat platform that can include images and videos is better suited to rich answers. Collins recommends that you have a single bot system, not one for simple Q&As and a different one for conversational interactions.

Step 4: Measuring effectiveness

“Delivering the wrong answer is far worse than not delivering an answer at all”.

That’s a key fact that is often overlooked, if you unleash a bot on your clients and it starts giving wrong answers or wrong information you could destroy your customer experience and brand overnight. To avoid this you need to monitor every question and interaction and look for false positives, if these are rising above 3% then you have a major issue.

Step 5: Continuously improving

Just like a real person, bots need continual training. This might be around new services and products, it might be new ways to solve problems, or an increase in bot intelligence that enables it to do more than it was initially trained to do.

Summary

It’s not a question of “should I be using a bot?”, but “how do I best use a bot?”. They will become a common part of any customer interaction model and can provide a very efficient and cost effective way to service customers. The key is not to look at bots in isolation, but to integrate them into your wider customer experience, create a seamless flow between all your channels, so that your website, apps, bots, contact centres and face to face interactions all work as one.

Consider the interaction, its complexity, emotional factors, your customer demographics and the information you want to provide.
Integrate your systems before you expose them via a bot.
Create an integrated customer experience model.
Make sure you follow all five steps of the bot training programme, and
Most importantly, use your imagination and that of your customers to see just what could be possible.