Sabine Ocker, Comtech Services
January 15, 2020

Since at least the end of 2017, we have been hearing that the chatbot revolution is coming. But there have been only rumors and hints of upcoming implementations. We hear of some chatbots being implemented, but almost exclusively in the marketing content domains with limited applicability to customer support or technical product documentation. Apologies to T.S. Eliot, but the chatbot bang has not happened.

Last month at the CIDM Round Table, we asked members to talk to us about their chatbot implementations. We had an excellent turnout and an interesting discussion, but as it turns out, no member organization in the conversation has a chatbot; they all came to learn about someone else’s. As one member said, “…in DITA you have the semantic structure and the intelligence already,” so if you have metadata and the ability to scope to the right granularity of content, such as just a procedure, then your chatbot will be in a better position to deliver a precise answer. The operative part of this statement seems to be the word “if.”

I started thinking about why this is the case. Given the leg up which DITA gives organizations to provide an environment to mine their semantic-task-based-action-oriented content for answers to help their users, why aren’t we in the chatbot revolution?

There may be many reasons, but I identified at least two.

The first is because of content silos. There is no such thing as a product documentation chatbot, a marketing chatbot, or a support chatbot. There is just a corporate chatbot. So, if there is only a company chatbot, then the effort to develop it, implement it, populate it with content, and ultimately fine-tune it must be an Enterprise-wide effort. Furthermore, any chatbot must engage with users before, during, and after they have made a purchase of a product.

Similar to other enterprise endeavors such as taxonomy, creating a chatbot requires involvement, input, and ownership from key stakeholders from all content domains which represent any stage of the user journey. Of course, user journeys vary from company to company and can encompass many more stages than those listed below, but most have a variation of at least these four:

  1. Discovery
  2. Evaluation and selection
  3. Delivery/Installation
  4. Ongoing Usage

The content which users may need to answer their questions in each stage:

  1. Product briefs, customer reviews, or product documentation
  2. Product specifications, sales brochures, product catalog, product documentation
  3. Product Documentation, Training courses, and content
  4. Knowledge Base articles, product release notes, support, and troubleshooting content

This means the enterprise chatbot will need content inputs from the Marketing, Technical Product Documentation, Customer Support, Sales, and Learning and Training content domains at the very least to be effective across the entire customer journey.

A complex enterprise chatbot ecosystem of content sourced by, owned by, and updated by these different organizations is challenging to implement and maintain, especially if that ecosystem has a built-in requirement of constant tweaking and fine-tuning based on feedback on how users are engaging with it. The challenges presented by these complexities are why most companies don’t have an enterprise chatbot. A company may have a marketing chatbot or a support chatbot, but since the content inputs do not cover all the stages of the customer journey, that conversational interface will ultimately be of limited usefulness to customers in the long term.

The other reason why we haven’t heard the chatbot bang relates to discovering user intents. As one CIDM member put it, “the accurate identification of intents and the entities associated with those intents is mission-critical [to] feed conversational interfaces.” There are multiple unique user intents for each stage of the customer journey. Therefore, implementing a successful enterprise chatbot necessitates the discovery and analysis of user needs generally and user intentions specifically across the four or six or ten points in the customer’s journey.

How do we uncover our users’ intentions?

A common mechanism is to mine search data to discover what users are searching for. The challenge with this approach is that conversational interfaces lend themselves to revealing what a person is trying to do. Evaluating user search terms in the form of noun strings is not as clear. This means it is much easier to fine-tune user intents once you know what they are than it is to determine what they are in the first place. If you multiply that difficulty in identifying user intents across five departments, you can understand why many organizations have awareness of only a sub-set of the total.

A CIDM member who is a technical product documentation manager put it this way: “We just don’t have a good sense of how our customers use our content — whether they search for more [in terms of granularity], or search for less — so in some ways that is what is missing, we need a way to gather that detailed usage information.” Another member said: “Our whole way of looking at it is to create content based on user research and then throw it over the fence. A chatbot necessitates data mining and other content tweaks after the fact, like tracking the conversation, mining the conversation, and we just don’t have a way to do that.”

The good news is once companies understand that chatbot implementation is an enterprise endeavor they will utilize the same existing processes and mechanisms already in place to ensure the new chatbot includes all the customer journey stages and user intents.

Defining and utilizing customer journey data is becoming an important part of many innovative corporate content strategies. If you are interested in learning more about customer journeys, user intents, and chatbots, and to discover how each might provide your users with a unified, consistent user experience, consider attending the new CIDM  Journeys conference.