Sam had started a bot development capability for his digital agency.
As part of the business development process he had visited many clients to explain the benefits of bots and by doing so noticed something interesting.
Regardless of all the different scenarios he had explained (and many clients were impressed and interested by what he had to say) all of them were interested in chatbots for customer service use case.
The customer service use case was something they could understand intuitively:
The cost of customer service agents was high.
Much of their time was spent answering simple, repetitive questions.
Bots could answer these types of questions more effectively in many cases than the agents could.
The cost of bots was such that if they replace some customer service agents with bots they could very quickly recoup the cost of the bot from the savings in labour costs.
Not only that, but the bot got better over time and opened the door to many other innovations.
Sam therefore decided to specialize his bot business on customer service bots as they were in high demand and from discussions with his customers and with other agencies he had established the economics were good. Once he had nailed this niche he would move to adjacent markets.
He knew that even though some of the tech around natural language processing (NLP) was sophisticated, the tools out there were so good now that almost anyone could implement a good NLP solution. These days it was a configuration job rather than a data science job. This meant that the skills required to implement the solution were well within the capabilities of even his most junior developers.
He was also aware that great customer service solutions did not try to do too much. The technology was not good enough to engage in human-like conversations with customers. The NLP engines were very good at understanding the first question that the customer asked but if the conversation got more complicated than that or the bot failed to understand the customer the first time round, it was important to have a human intervene immediately.
The fact that the main focus on the NLP was the first question or interaction also meant that the task was much simpler from a technical point of view. It’s true that some companies were going down the route of trying to build a truly conversational experience but so far this approach had lead to escalating complexity and failures. Sam had no interest in taking this path.
He had observed that there were many emerging offerings for customer service bots on the market so the perhaps market might be competitive. At the same time every business needed this type of solution in some form so the opportunity was huge! In fact Sam saw the market as currently massively under-served and believed it would be this way for a few years at least.
There were many companies that offered proprietary, off the shelf solutions for customer service. Sam considered being a reseller of this type of solution but he didn’t think that these proprietary solutions were the best solutions longer term. He wouldn’t use them for his own business so he didn’t believe he should be offering this type of solution to his clients either.
He didn’t like the thought of being locked-in to a proprietary system for a few reasons:
He didn’t like the idea of tightly coupling the natural language processing engine (NLP) with the rest of the software. The best provider of the natural language engine might be a different provider than the best provider of analytics services, or the best provider of the connectors to the chat platforms.
Even if their natural language solution was the best right now, there was no guarantee that they would be the best provider in future. It should be possible to switch NLP engines in future.
Different NLP engines might be good at different things so you might want to use more than one. For example, one NLP engine might be good for IT questions, another might be good for general FAQS.
NLP was not the only capability that was needed for a great customer service bot. Relying on text and NLP alone was definitely not the way to create an amazing customer experience.
Text interfaces are very limited and therefore need to be augmented with graphical interfaces. Having to rely on the vendor to provide these interfaces in future would not be optimal.
Human in the loop functionality (allowing the bot to be escalated to human agents if it fails to understand something that is said to it) is critical for customer service. Human in the loop functionality would become more sophisticated in future. It would include customized interfaces for the agents that include canned answers or customized responses. It should be possible for anyone, including an in-house bot developer, to customize the human in the loop for their own purposes without relying on the vendor.
The content should be easily managed by the content team who should be able to develop use tools for A/B testing and other market related analysis.
It needed to be easily possible for internal developers to integrate the service with internal systems without relying on the vendor.
It’s possible that other services could eventually be offered over the customer service channel.
A customer inquiring about a room at a hotel may be offered a discount and the opportunity to book immediately in the same channel for example. In house or third party developers need to be able to code these additional services in the channel without having to go to the original vendor.
He wanted his solution he offered to be simple however. Being future proof and extensible was important, but at the same time it should be very easy to get an initial solution up and running.
If he used Botpress he could sort out the extensibility issue. Botpress could effectively act as a middleware to all the best chatbot tools on the market and well as providing many standard components out of the box. These components could be switched for third-party components or customized as required in future.
To offer his customer the best solution he would select a set of tools that he believed offered the best value solution for the customer.
To start with the architecture would be very simple. He would select the best NLP engine from the main NLP vendors (Google, Facebook, Microsoft, IBM, Rasa) and then use the standard components from Botpress to connect to the required messaging platform and to provide supporting capabilities such as analytics, human in the loop and role based security.
Once this basic solution was up and running with the customer he could think about improving it if necessary. This could do this by adding NLP engines, upgrading the analytics package (to a third-party provider if necessary) or adding customizations to the human in the loop feature.
Of course decisions to further customize or to add new components or services to the system would be driven by analyzing the customer interactions with the system. Changes could be made quickly as required by the appropriate developers.
Sam decided to begin with implementing the solution for his own business first. This allowed him to experiment with some of the tools available and choose what he thought was the best offerings for the initial setup.
Not only did Sam immediately start to win many customer service projects, he found that once a successful implementation of customer services was achieved, his clients quickly wanted other services added to the channel.
Since customer service is something that every company in the world needs more or less there was no shortage of clients.
It’s only once he started implementing the solution this way did he appreciate how much his clients valued the fact that he was able to help them choose the best tools for their business rather than trying to lock them into a single, non-extensible, bloatware solution.
Sam and his team have become very good at figuring out how to use customer service bots to very quickly make high value, high impact improvements to a company’s customer service function.