Many in the AI industry have a high regard for virtual assistants and really don’t like calling them bots at all. Standard bots are not designed for complex tasks–they can’t understand natural language nor can they perform sophisticated processes like understanding customer intent. Putting them in the same category as virtual assistants, which can process transactions and communicate like humans, seems to be a disservice to the technology itself.
In this two part series, we’ll be taking a closer look at AI chat and how it has evolved over the generations. Specifically, we’ll be talking about chatbots designed to serve businesses—in other words, virtual assistants. (We’ll be calling them “chatbots” for the sake of brevity.) Armed with this knowledge, hopefully you can avoid a “do-over” when implementing AI chat for your business.
3 factors to keep in mind
- How much time will it take to roll it out? (What’s the time-to-market?)
- How much will it cost upfront?
- How much control is there once it is up and running?
1st Generation AI chat is good for simpler tasks, and it’s costly
The first generation of AI chat has been around for almost a decade. While it’s capable of performing moderately complex tasks, it often ends doing lightweight and simplistic tasks. Why? Because costs and time-to-build are both extremely high. The reason they’re so high is that on the backend, first gen chatbots are built on very basic rules-based code. This means that they use a preset list of “if-then” statements to provide answers to questions or perform tasks.
As you might imagine, writing coding for every single possible scenario into rules is a large and complex endeavor. It requires significant man-hours on both the client and developer sides to map out typical situations and write them all down. When it’s all built the number of supported use cases will then be limited by rules. Most companies don’t want to spend the massive amount of time and money it would take to make the system function for every possible scenario. So the system ends up focusing on a limited number of simpler tasks. Even with this limited scale, rollout time averages several months.
Making changes with 1st gen AI chat is a big deal
Here’s the likely functionality you’ll get with a first gen chatbot, which is based on how much time and money you spend: It will present user-specific responses, and it will clarify user intent with menus in response to simple questions. It’s a good chatbot for helping customers deal with Tier 1 issues, like password resets (assuming you spend time writing the right rules). However, unless you invest significant money, don’t expect to see significant operational savings or big leaps in customer satisfaction.
2nd Gen AI chat is slightly more sophisticated
With so much work required to massage data, second gen chatbots also share similar downsides with the first gen. They’re both labor intensive and costly. Rules and labels limit the number of supported use cases. And the maintenance cost is high because new data is constantly coming in that needs to be labeled. There’s also no real-time control of system responses.
The customer experience with a second gen chatbot is a little better than first gen. That’s because—thanks to all the labeled data—the system can respond to more customer scenarios than first gen chatbots.
1st and 2nd gen AI chat are good for certain use cases
If you’re looking to rollout a chatbot in a couple months that serves customers on a very limited basis, a first or second generation AI chatbot is something you might consider. They’re best for large corporations with big budgets. As long as the chatbot is only needed to perform Tier 1 tasks–and few changes are anticipated–first and second gen chatbots can be quite serviceable.
Stay tuned for Part 2 of this series, in which we’ll examine third gen AI chat. If you want to learn more about how Rulai has designed its chatbot to help businesses just like yours.
Or, if you want to read ahead, you can get our view of Generation 2 in our Solution Brief, which you can get by clicking the image below.