AI

7 Things To Consider Before Getting A Chatbot For Your Business

Among all of the Artificial Intelligence (AI) type tech, such as bots, robots, and chatbots, a large amount of attention is being given to chatbots. Every business feels it needs one, but many don’t have a clear business reason or vision to fully utilise it. Often this desire for a chatbot is simply based on a fear of looking less tech-savvy than the competition. Despite this, chatbots done right can deliver huge value to a business.

Beyond the hype, there are good reasons for the growth seen in chatbots. Users are experiencing app-fatigue and don’t want to install yet another app for temporary or quick transactions, such as renting a car. Chat is also a more human interaction model, and when correctly implemented, is much easier than learning another applications’ menu structure. In addition, it allows your application to be compatible with voice services such as Cortana.

The business benefits of chat are also potentially huge, freeing staff from handling mundane queries, requests and emails and simultaneously giving the user a much faster response.

Chatbot Ready?

Thinking about implementing a chatbot for your business? Wondering how to proceed without making your customers angry and causing your business reputation damage? Remember that a chatbot is a new customer channel with many potential pitfalls. Here’s a list of 7 things to consider before implementing a chatbot for your business.

1 – It’s chat, not command-line

In the 1980s, Apple and Microsoft unleashed the Graphical User Interface (GUI) to the world. This was a huge milestone as it largely replaced command line as the application interface and opened computers to less technical users. Despite this decades-old advance, some people are now putting out what are essentially command line interfaces and calling them chatbots. This is the opposite of technological advancement. To check if your bot is actually just a command interface, ask if the users need explicit knowledge of keywords to use it. If yes, then welcome to the 1970s. A related problem is a trend of trapping the user in a script. A trapped user is not a happy user. If the user engages with your chatbot, they deserve a prompted, conversational experience and not a game of “guess the keyword”.

2 – It’s not “real” AI… yet

Although deep learning is radically changing the chatbot landscape with big milestones almost every other month, the technology still has severe limitations. A cleverly scripted demo can make it seem like a chatbot understands and possibly even jokes with the user. This is dishonest marketing and doesn’t represent where the technology really is. The reality is that cutting edge chatbots typically fall into two categories: firstly, there are deep learning chatbots trained on large conversation databases. These can yield a surprisingly human chat experience, but behind the scenes, the chatbot doesn’t really understand the content of the message and therefore, can’t act on it. These bots are largely just nerd toys giving a glimmer of what can be without delivering any real value.

The second type of bots is those that are trained to understand language mapped against a specific list of system-intents and are the ones relevant to a business chatbot. These are limited to only understand the intents explicitly added by the programmers. Ultimately, the user is occasionally going to ask things that your bot simply can’t action. For this reason, it’s best to be cautious with use-cases where the users are emotional… so maybe hold off on your complaints-line chatbot.

It’s also best that the bot makes it clear for the user that they aren’t talking to an actual person. People tend to be more forgiving when they understand they are talking to a machine instead of a seemingly unhelpful individual.

3- Don’t leave the user guessing

So, every app will have limits to its functionality. With a traditional application, the user can figure out these limits simply by looking at the menu and clicking around. A bot is different, and a poorly designed bot can seem far more limited than it actually is. The bot should describe its functionality on first use and whenever the opportunity comes up. Contextual help should be presented not only when the user requests it but whenever the user seems to be stuck or frustrated.

The reason so many bots fail in this regard is that the developers and stakeholders involved in a chatbot development project know the functional limitations of the bot, and are, therefore, blind to the lack of help and guidance. Make sure your bot is always telling the user what they can do.

4 – Some engagements don’t make sense over chat

When I need to complete a complex form in a browser, I can visually skim the entire form for context and rely on my browser to auto-complete many of the standard fields. I can also skip around the form to complete it as I desire. I hate completing forms but doing this over a web page is about as good as it gets. However, completing a large form entirely over chat shouldn’t be imposed on any undeserving customer as it’s an excruciating exercise. Typically, the poor user will need to complete one field at a time and navigating the form is difficult. There are ways to mitigate this, BUT the reality is that some types of user interactions should step outside of the chat model. The same is true for navigating and presenting large sets of data.

Let this also guide your business case for your bot. Avoid using chat when other traditional channels make more sense.

5 – Get the tone right

Your chatbot is an extension of your brand. It is the robotic receptionist welcoming the customer into your digital office. Therefore, it should have a speaking tone that’s in harmony with the aesthetics of your brand. Nobody wants a dad-joke quipping bot helping with funeral cover… unless the user really, really needs some cheering up. A bot should also know its users and potentially modify its tone and language for specific users or specific products. A huge note of caution here: a bot that’s trying too hard to use “hip” language for a demographic can come across as embarrassingly awkward or, at worst, culturally insensitive.

6 – Chat-channel specific pitfalls

Chat channels are 3rd party apps and mechanisms that your bot can use to talk to the users. There are many to choose from: Facebook messenger, Skype, Email, SMS, and voice to name a couple of the popular ones.

Let’s start off by addressing the obvious omission from the list, WhatsApp. At this writing, there is no official WhatsApp bot API. It is possible to hack a bot into WhatsApp, but this goes against their terms of use, and are, therefore, typically blocked. The good news is that WhatsApp has officially hinted at changing this and have started releasing bots for select enterprise customers.

With all these other channel options, why not just launch on all available channels? Unfortunately, this probably isn’t the best strategy. Different channels support different feature sets e.g. images, rich-cards, form-elements etc. This will require that either your bot has differing behaviours developed per channel, which is very costly, OR the bot conforms to the lowest common denominator, which is small text messages. Small text messages are boring and limiting – sorry Twitter, but it’s true. Bot channels also have differing security models which may not be compliant with your business security and user authorisation. Speaking about compliance, how does your business feel about its sensitive data going through a 3rd party app? For some businesses, that’s a deal breaker, although I’d argue they all seem fine with you using a 3rd party web browser.

So where to start? Start with one that works for you and your customers … and see the staged roll-outs information below.

7 – Going big bang

With any software project, doing a single large launch into production is risky and with chatbots, it’s guaranteed to fail. For starters, the semantics (e.g. jargon) used inside your organisation will differ to that of your customers when talking about the same thing. You say tomatoes, they say ketchup. A bot that performs flawlessly with your internal testing team could fail horribly the moment real users start applying their own wording and flow to the requests. The reality is that chatbot technology will require multiple iterations of testing and re-tuning across a diverse set of users. Through each iteration, the requests from the users should be monitored and used as learnings into the next training iteration. The sentiment of conversations should also be monitored. A benefit of chatbots is that user’s failed requests often point clearly to what new features need to be prioritised in next releases.

All the above complexity is multiplied if the bot is intended to work with multiple languages so proceed very carefully, ideally with a single language at a time.

So how do we roll-out a chatbot? The best strategy is to limit and slowly increase the scope of the customers who can access the bot through each iteration. Start off with only an internal test group, then staff, then pilot users and then use other mechanisms such as a single channel to limit access. Learn and tweak, tweak and learn through each increased set of users. Marketing a big launch of the bot should only happen after the bot is performing adequately with the pilot group of users.

Summing it all up

Chatbots are going to increasingly become an important tool influencing the way we engage with our customers, other businesses, and systems at our places of work. Bots can add real value and are the perfect mechanism for some customer engagements but don’t warrant throwing away your website or mobile app just yet. Over the last year, we’ve seen some very public failures, and this will likely increase as more ill-conceived chatbots hit our chat apps. Keeping it simple and treading carefully should keep you off that list. Then again… why waste time when you can be launching that super-sentient, Mr. guru-chatbot that made inappropriate jokes when the sales guy demonstrated it.

You may also like

Leave a reply

Your email address will not be published. Required fields are marked *

More in:AI