Chatbot Development

5 Bot Development Mistakes and How to Avoid Them

Building a bot can be a difficult process, with plenty of trial-and-error. There are a few common mistakes that developers make along the way; fortunately, these can be easy to avoid.

By Paige Twillmann
June 7, 2019

Bot development isn’t easy, especially if you’re trying your hand at it for the first time. Even veteran bot developers can make mistakes—and these mistakes can often frustrate users and lead to a negative experience overall. We sat down with Paige Twillmann, Global Solution Lead of AI & Bots at LivePerson, to discuss a few common development mistakes and the best solutions for avoiding them.

1. Lack of transparency

You need to be very transparent to your customers when they’re talking with a bot. Customers can become uncomfortable when they think they've been talking to a human, only to realize down the road that they have been talking to a bot all along. In many cases, customers don’t mind talking to a bot, but you want to remove any element of perceived deception from the interaction. Last year, California Governor Jerry Brown signed a piece of “bot disclosure” legislation mandating that companies immediately notify customers whether or not the customers are speaking with a bot. The legislation is the first of its kind and goes into effect on July 1, 2019. It is likely that other states will soon follow California’s lead.

Solution: Always be up front with users that they are chatting with a bot. While it can be gratifying to have created a bot that is so human-like in speech that customers aren’t aware that they are not speaking with a live representative, this is an easy way to frustrate your users and have them end the interaction prematurely. Make sure that your bot introduces itself properly and establishes that it is, in fact, a bot.

2. No transition to a human representative

Bots can be a great way to streamline communication on your site, especially when it comes to customer service. However, humans will likely always have to be a part of customer interaction, and it’s important that your bot knows how to transition the conversation when necessary. For example, if a customer has been having an extended interaction with your bot, and the bot deems it necessary to transfer the customer to a human representative, the existing conversation script would need to carry over. If it didn’t, the customer would be forced to re-explain the issue to the human representative and start the entire process over again, resulting in frustration.

Solution: Leverage a seamless messaging interface. It’s important to make sure that you're leveraging a seamless messaging interface to continue those customer conversations. If the bot doesn't know the answer, the last thing we want to do is pass the interaction to a human agent who has little context, or, even worse, offer a phone number, redirecting the customer to a phone agent with even less context. Having your bot backed and supported by a warm handover to a human agent in a messaging-type environment, where the context is being shared, is one of the most efficient ways to optimize the customer experience.

3. Your bot is too repetitive

Customers can become frustrated with a bot very quickly when they find themselves having to repeat their question over and over again. The development issue here is that the bot hasn’t been programmed to ask qualifying questions, or its script is simply too limited to respond properly to the customer’s query. If the bot isn’t giving the customer the answers they are looking for, they will immediately seek a human representative. While there isn’t a problem with this, if it is happening on a regular basis, it is defeating the purpose of your bot and leaving the customer with a negative perception of your site or offering.

Solution: Build logic into your bot. Design your bot to ask clarifying questions without actually reiterating and re-asking the same exact thing that the customer just asked. Building logic into your bot means that it will ask these kinds of clarifying questions when the customer query is something it doesn't understand, rather than just saying, "I don't understand." It is a good kind of trick to help avoid repetition of the same question and the bot’s speech becoming an endless pattern of "I don't understand." "I don't understand." "I don't understand." Allowing yourself a great space to test and fail to adjust for those customer expectations is important. Make sure that the bot development is in line with the expectation of where the bot is living. From there, roll out soft launches to adjust and iterate on the bot once it's in its live phase so that you can make sure that it's meeting the expectations of what it should be able to answer.

4. Not knowing your audience

I think, for this point, it’s best to provide an example. The Cosmopolitan in Las Vegas recently launched a chatbot that acts as a virtual concierge for its hotel. The bot is a fairly lighthearted in tone, and gives the customer the information they seek while also maintaining an air of levity. This is perfect for Cosmopolitan’s brand voice, but can you imagine if a chatbot with this same tone was installed on a bank’s website? Users would be trying to ask important questions about their finances, and the bot would be replying with jokes and risqué factoids. It wouldn’t add up, and the customer would be left frustrated, as their experience with the bot did not match their expected experience with the brand. On the other hand, if a customer went to use the Cosmopolitan virtual concierge bot and it had a very serious tone, this would also be off-putting and not aligned with the expected experience.

Solution: Match your bot’s personality with your brand. This seems like a straightforward recommendation, but don’t be deceived by its simplicity—it’s integral to the success of your bot, and therefore must be a central consideration. Before considering developing a bot for your company or website, make sure you are intimately familiar with your brand’s voice and what kind of interactions customers would be expecting to have with your brand in the virtual space. If you’re a life insurance company, your customers are not going to want a bot with a tongue-in-cheek conversational script. A bot shouldn’t have one developer putting words into its mouth and writing its entire script; it’s important to welcome collaboration and not let your bot development process become too siloed. The bot’s personality may seem like a good fit to you, but by bringing other developers into the fold, you can get new perspectives and mold the bot into the best it can possibly be before you get to the point where you are launching it to a user base.

5. Giving your bot the wrong home

Customers will have certain expectations around the capabilities of your bot, depending on where it exists on your website. For instance, a bot that pops up to chat on the home page would be expected to have a high degree of general knowledge about the site and all offerings, while a bot that only pops up when the customer clicks "Pants" on a retail website would be expected to have a more specific knowledge base centered around pant styles and sizes. An easy way to frustrate customers is to not give your bot the proper information to exist where you have put it. As we saw in the “Not knowing your audience” example earlier in this blog post, customers become irritated when their expectations of an experience do not match up to the reality.

Solution: Use test phases to understand customer expectations. Let’s bring in another example here, to give more context to the issue. We recently created a Baggage Bot for a popular airline. It exists in the baggage section of the airline’s site, and customers are able to use it to ask basic questions about baggage. To understand what kind of baggage-related queries we could expect from customers in this section of the site, we did a soft launch, where the bot was only live for a few hours a day. In these soft launch hours, we were able to learn more about customer interaction with our bot than we learned in the previous three months of development. Take advantage of soft launches and don’t be afraid to put your bot out there for a few hours and then take it back offline to make tweaks. This applies to most bot development mistakes and can be a great way to recognize the weak points of your bot.

Above all else, remember that bot development isn’t a static process. Your bot is something that will require constant maintenance and upgrading on your part so that it can grow into the most efficient bot possible. Make sure to track your mistakes and learn from them—even the most disastrous development missteps can be fantastic learning opportunities, especially for novice developers. Keep striving, and keep building!