You can’t have missed the buzz around chatbots in recent years. These human-mimicking pieces of software have come a long way, springing into existence as clumsy efforts to pass the Turing test and being honed by developing NLP technology and machine learning into what we have today: practical tools for handling certain predictable tasks.
But is that buzz actually warranted? The way some talk about chatbots, you’d think that they’re on the cusp of driving widespread human obsolescence. Install a chatbot, so they say, and all your generic support problems will be solved, leaving you to recline on a sun-dappled beach somewhere with a refreshing boozy tipple.
In this piece, I’m going to answer the question of whether chatbots are truly capable of solving all those problems — and if you’re at all familiar with Betteridge’s law of headlines, you’ll have a solid idea of which direction I lean towards. Let’s begin.
How chatbots have become mainstream
When it comes to digital UX, no industry has done more to push things ahead than SaaS. Apple has done more for UI, of course (pioneering so many of the touch-friendly input methods that we rely on for mobile use), but it was the move from one-off software purchases to monthly or yearly subscriptions — perhaps driven somewhat by a determination to combat digital piracy — that made it a top priority to elevate UX standards.
Top Articles on How Businesses are using Bots:
1. Chatbot for Healthcare: Chatbots Can Be Money-Savers for Hospitals and Clinics
2. Higher Education Chatbot: Chatbots Are the Future of Higher Education
3. How Chatbot Performance Metrics Differ by Industry
4. Chatbot Conference 2019 in NYC
And as SaaS honed their systems, the e-commerce retailers that used their systems took note of the selling power of UX, soon adopting the new methods. If SaaS hadn’t become so prominent, I’m not sure that chatbots would have come so far in such a short time, because people would have been reluctant to rely on them.
In the SaaS world, one of the key operational demands is convenient for in-depth support. That’s the inevitable consequence of making software a service — it ceases to revolve around isolated transactions between which a seller can simply ignore its previous buyers.
Scaled up, though, this level of support becomes incredibly challenging. You need a large support team, the technological setup to handle numerous cases, and the consistency to maintain a strong standard at all times.
No matter what resources you can bring to bear, it’s a problem to meet that demand. That’s how chatbots entered the mainstream. By making note of all the most common queries and programming set responses, service providers could rapidly automate many of their support tickets, leaving their teams to handle more demanding queries.
Subsequently, e-commerce retailers found ways to push things ahead through folding their rote follow-up procedures into the chatbot model. If you’re using a generic template for marketing spiel, then you can add that template to a chatbot and have it take over that dull task. It won’t pass for an expert salesperson, but it’s cheap, easy, and fast.
Why they’re not as easy as they’re made out to be
If you’ve ever happened upon a site that strongly advocates for chatbots, or even creates them, you’ll know that they’re very commonly spoken about as though they’re incredibly quick and simple to implement. This is far from the truth.
This is mainly because chatbots have very little dynamic functionality — you don’t select an AI personality type and pass it to your customers to engage with their problems and figure out solutions. You deploy a static call-and-response tool with a limited ability to interpret context.
When you arrive on a site with a chatbot and it pops up with “Hello! How can I help you?”, that, of course, isn’t a construction assembled from a huge range of viable fragments. It’s simply one of a list of greetings, and those greetings need to be written and placed within context. The same goes for every response or action. In some cases, you can use a pre-built chatbot library and have success without making manual changes, but you’re still relying on written content.
As such, there’s also no room for improvisation. At best, you can simulate improvisation through anticipating unlikely circumstances and providing suitable settings, but getting that granular is going to add significant time to the already-daunting chatbot setup process.
It suffices to say that using a chatbot is not a comprehensive solution. If it’s easy to set up, then it’s extremely simple. If it’s even slightly complex, it’s an arduous thing to configure. Does that mean that chatbots are useless? Not at all — a simple chatbot can be extremely valuable when used appropriately. It’s merely a tiny part of a much larger and more complex system.
How chatbots can factor in support tasks
Let’s take a look at a couple of sample tasks to see how a chatbot might (or might not) provide effective support. Firstly, we’ll consider something that’s a perfect fit for a chatbot: dealing with an already-placed order.
When a customer places an order through an eCommerce store, they’ll be presented with some kind of unique identifier for that order in the form of an order confirmation number (or string). If they then return to the store with some kind of query or action they want to issue regarding their order, they might want to speak to a member of the support staff — very often there will be a user panel that allows changes, but let’s assume that the customer isn’t aware of this.
In that kind of scenario, having a sitewide chatbot can be tremendously useful, because it can pop up as soon as the customer reaches the site, asking them if there’s something it can help with. As it happens, there is: they’d like to know the status of their order. When they enter a relevant query, the chatbot can pick up on the significant terms (“order” and “status”, for instance) and advance to the stage of requesting the order number. From the reply, it can pick out the string that meets the order number format, and extract it to retrieve the status.
Now let’s look at a less useful scenario: handling a customer grievance.
When a customer is unhappy with the service they’ve received, they’ll often find it difficult to articulate their problem, not least because they don’t really care to. Feeling victimized, they’ll expect the company to know what the problem is already, and will be further irritated by being asked to painstakingly detail the source of their frustration.
Imagine that customer happening upon a cheery chatbot. Not only will they feel insulted at being asked to deal with a chatbot, but they’ll be further annoyed by being placed into a position of needing to jump through linguistic hoops until they get the required response. The best-case scenario is that the chatbot rapidly brings in a real person to handle the case from there.
The takeaway from this is that a chatbot can be useful in context and for the right kind of task. It’s certainly great for sprucing up a site and leaving a punchier impression. Yes, it may sometimes be irksome to hear the notification sound of a chatbot pop-up, but it’s often better to stand out than play it safe (definitely the case in the competitive e-commerce world).
What’s more, once you have the infrastructure figured out, it becomes a quick fix for any site you want to improve — particularly useful if you want to diversify your portfolio, because as long as you buy through a marketplace that exclusively uses a suitable CMS (like this one for Shopify stores), you’ll know that you can roll out your chatbot in minutes and make your workday even more efficient.
With all that said, we’re about ready to answer the titular question, so stay with me for a little longer as we wrap things up. Here we go…
A chatbot is not a magic bullet
Can chatbots solve all your mundane support problems? No. Not now, likely not in the foreseeable future, and potentially not ever. Why? Because the type of AI that often gets spoken of alongside chatbots is extremely shallow and doesn’t warrant its inclusion as a type of AI.
This isn’t to say that machine learning doesn’t have vast applications, because it does. In time, I’ve no doubt we’ll see machine learning solutions capable of rapidly analyzing extensive customer support records and picking out answers to common queries.
This is essentially what data-driven healthcare diagnosis systems are getting good at turning differential diagnosis into a fully-automated quiz that will eventually identify the problem. But diagnosis is just one part of the healthcare world, and no set of stock responses will be able to handle patients who are mistaken, or confused, or elect to lie.
Regardless of the purpose, once a certain level of complexity is reached, human support will be required — support with a capacity for semantic nuance that NLP is still incredibly far from achieving, and the ability to be creative.
If you want to implement a chatbot on your site, that’s a perfectly fine idea. Use it alongside your standard contact options, a live chat service, and a knowledge base. Make it a part of a comprehensive support system and it will thrive in that role. AI is the perfect augmentation: just don’t view it as a panacea, because it’s not even close.