Hello everyone,
It’s been a couple weeks since we launched our first major bot network onto a website. Currently, we are running 30 bots that work in unison to deliver three major services. I wanted to share what I have learned in the development of the bots, and what I have learned in the first two weeks of the system being live.
First and foremost, large bot networks are not for everyone and that is okay. Most users are only looking for a few bots max to run specific albeit simple functions. And guess what? Still a great use case for TARS bots. But what happens when you approach a large business or complicated government entity with the sales pitch of bots? Well, things are going to get a little more complicated because these entities have a lot more information and a lot more services that can benefit from bots. This means bot builders have to be open to large complex networks to meet those needs.
Like anything the first real step is to identify the needs for entity. If they have a well-maintained site, they should be able to pull analytics to offer some great starting points. Where is the traffic? How is traffic flowing from area to area? Is it intuitive, or are there hang-ups? In most cases we will find that even the best designed websites have hang-ups. After all, you are asking the users to filter through all the info on the screen and make the right decisions on where to go next. Easier to accomplish on a landing page, but a full site is another matter. You can also determine needs looking through customer feedback and listening to employees that work “in the trenches” as customer support or sales. They know best what people are frustrated with because they deal with them on a daily basis. Most administrators/managers do not, so be wary of their understanding of what customers want or what is frustrating them.
Now if their website is good, let it be your guide in creating bots if you intend on navigating users through information/services offered by the company/website. Your bot can easily transform into a replacement or reinforcement of the nav bar found on most website, except the bot “talks with people” who are navigating, meaning you can help people find exactly what their looking for with greater ease. I should have mentioned this earlier, but bots are not for everyone (and if you have pulled customer feedback from bots, you know what I’m talking about). You have users that solely rely on a search bar to find information (which is a feature I am tinkering with to include in bots as well to capture these users). You also have your nav bar users that will rely on breadcrumbs and bar for most of their searching (however, a poorly managed nav bar can run these users off really quick or towards the nav bot). Finally you have the battle-hardened old-school users that will trudge down every page until they find what they want (mostly these people choose this option because they don’t understand search or nav bars. They usually complain about how much they dislike computers). These users are really looking for a simple way to be led through to the information they need; they want the “on-the-rails” approach, but it has to be simple and easy to understand. A bot built the right way will appeal to these types of users the most as it takes the headache out of the processes and keeps it simple.
Okay, so you have this crazy daunting task of building a bot system for a big company; you’ve collected all the content you need; and you have a good baseline with their existing website. What’s next? Build, build, build! Much like taking the first steps to writing a paper (or book for that matter) is putting pen to paper or in this case gambits on the dashboard. Don’t be shy, don’t worry about how clunky it may feel at first, and avoid writing much if any conversation from the bot to start (I know, blasphemy!); we will get back to that later. Just start building out the structure of your bot. The general navigation should be rather simple using the website as a guide, but remember, it is only a guide. You will stumble upon better ways to offer options, name areas, and navigate sections. Continue this process until you have a working skeleton for your bot. Note: keep plenty of space between your gambit clusters or “conversation flows” as you will inevitable find areas that you will need to add additional gambits too.
Now, you may find yourself running out of “space” (e.g. the gambit load is just too much for the dashboard which currently happens around 180 to 200 gambits depending on a couple factors, but there is an architectural upgrade on the way, so this statement will be outdates shortly). In this scenario, you can easily create another bot and link the bots together, depending on your deployment method(s) of the bots, your users might not even notice the shift from one bot to the next. This technique is also really helpful for handling a bot customer service system that I will explain in a little bit.
Okay, so your structure is up and you are ready to start adding content. First thing to consider is your audience. Who are you talking to? If you write your conversation to appeal to 18 year olds and your primary demographic is 25 -45 year olds, you will miss your audience and lose a lot of users in the process. Here is where the customer service reps for the company are invaluable. They know the audience, and talked to them everyday. If you can even listen to a handful of their interactions with customers, you will be way better orientated to the feel your conversation should have. If this seems too much for you, hire a professional writer. They will help you significantly with this part of the process. The writing itself should stay consistent across every interaction. Keep is concise. Keep it simple. Give your users the most condensed info they need to still make the correct choices while interacting with the bot. For example, say you are introducing the user to the main menu with 10 choices they can make to proceed to different areas. You could say, " Welcome to the main menu! Choose where you would like to go. For inquires into what we do click the contact us button. For shoe options we provide click the services button. Etc." versus “Welcome to the main menu! What would you like to do?” and add enough information to each button choice that the user can identify where/what each button will do for them. In this case we could break down the button options into “High-top Shoes, Covered Sandals, Running Shoes, etc.” Remember that speech bubbles from the bot take up time from the user before the can make their next choice, ensure you are spending that time only with critical information.
Give your users options to move around your bot structure as much as possible. In my bots, I always include a “Back” button (when available) to allow users to move in and out of areas with more autonomy. I restrict this once they have chosen a path for a service and are actively moving to the second step in the process (they had their chance to back out when they first entered the specific service, moving to the next step is an acknowledgement that they want to be their). If a user finishes a service don’t automatically finish the conversation (unless you really want that report for lead/gen or other means) instead offer the customer an option to step back to a couple different main menu options and/or a contact us option in case they had issues in the process. Again the sense of autonomy goes a long way with users.
Now if a business has a lot of customer reps handling different issues/products/services/etc., you can build a filtering system in a TARS bot to send users to the correct reps that can handle their issues. This is done by linking bots to bots. If you build a menu that filters customers based on subject they choose, once they are filter into a specific area, you can send them to another bot tasked with the sole purpose of collecting said users info, questions, etc. and (using the configure option on the dashboard) insert the email or group email that the customer service reps monitor that are experts on said subject. This sends the report to the right people, right away, cutting out a lot of wasted time forwarding and manually sorting emails. You can also send ever report to the same email, but change the subject line to reflect where the customer can from, so in theory a group using one email with different specialties would just look for the appropriate subject line that relates to them.
Okay, I’m out of steam for now. I’ll revisit this topic and continue to add to this. It was essentially a free write, so excuse any glaring errors or wanderings in subjects for the time being. I’ll clean it up later 
