The Complete Guide To Building Support Chatbots in 2021
Technology | April 30, 2021
The most talked-about support capability of this year is going to be Chatbots. Considering the popularity of chatbots in the consumer world, customers see it as a default option in any support site. A successful chatbot program in a Hi-Tech or B2B environment requires a lot of training, customization, and integrations, making it an expensive experiment considering the maturity of the solutions (Currently they are no more than glorified search engines).
Based on our experience at Hashout, we suggest that you start your chatbot journey by integrating an inexpensive, off-the-shelf chatbot solution with your KCS knowledge base to deliver a functioning bot.
While it is easier said than done, you could go from "What is a chatbot?" to "My chatbot is working!" in a matter of weeks, with just two minor changes to your article template.
- Add a new field to your article template to update Chat variations (Utterance) of a specific issue. By default, it should have the article title. This will be searched by a chatbot to locate relevant articles. When a new variation is identified chatbot adds it here.
Add the Chat resolution (Response) field to the article template; this is used to author a crisp chatbot version of the article resolution. Below are some best practices to consider when authoring chat responses,
- Keep the answer to less than 100 words. Use bullets or break lengthy solutions into multiple short messages. This will improve readability.
- Wherever possible, provide a link to the detailed process instead of the actual process.
- Use GIFs in place of multiple screenshots.
- Embed videos to illustrate complex processes/troubleshooting steps.
To begin with, you can also consider auto-filling this field by combining the Title, Summary, and Link of the article. Later revisit the top used article to write custom chat solutions. Tip: Use 80:20 to identify top content and have a chat solution written only for them during launch.
Of course based on the success and adoption, one could decide to invest in a comprehensive solution, team, and integrations.
Artificial Intelligence is the buzzword of the century and one of the avatars it has taken in the hands of consumers and businesses is Chatbot.
While the technology to make chatbots independently intelligent is still at its grassroots level, the concept itself has already brought several chatbot solutions to the table. These solutions require extensive training, maintenance, and feedback and the level of effort (LOE) varies depending on the complexity of the questions it needs to answer. Many support organizations experience this bitter reality during their assessment of the solution or within months of implementing a solution. That said, one can safely conclude that the success of a chatbot has a non-linear correlation to the effort invested in training. It is non-linear as the LOE could reduce post the initial setup.
As per the 2018 Technology Adoption And Spending: Support Services by TSIA only 15% of their member organizations have adopted Chatbot and it is the least adopted technology out of the 23 categories.
To push adoption, chatbot solution vendors like IBM Watson, Dialogueflow, LogMeIn ...etc have come up with innovative approaches and tools to enable Ease of training and Maximize ROI.
Based on the approaches taken, most solutions in the market can be broadly classified into the following three categories:
- Leverage existing content: Focus on indexing existing content where there is an ongoing investment like Support Knowledge Base. These offer the least customization capability and require very little effort in training and maintenance. They have no or minimal conversational capability.
- Highly customizable: Focus on facilitating a solid platform where any simple or complex use cases can be solved with advanced integration capabilities. Implementing such platforms requires a sophisticated team of Chatbot Architects and Software Developers.
- Centered around live chat: Focus on providing chatbot as an extension to the Live chat capability and highly rely on live chat responses to train itself.
There are solutions like Bold360 that offer key elements of all of the above yet have limitations across all of the capabilities. On the other hand solutions like IBM Watson provide chatbot API only, however, it offers solid integration with SalesForce LiveAgent.
A traditional knowledge base consists of one or more long-form templates, each with its own content and meta fields. A typical problem-solution article has the following fields,
- Meta: Product, Feature, Version...etc.
A problem-solution article could be as small as 100 words and be as long as a few thousand words with screenshots, tables, and videos. The success of these articles depends on two key factors,
The power of Google to index each and every word of the article and suggest a list of articles relevant to customer keywords. Google also considers metrics like CTR, keyword density, site ranking...etc to rank results. Customer readiness to adapt to the technical depth, level of detail, and style in order to find the information they want. With chatbot it all changes.
There no Google, your bots NLP, and search engine do the heavy lifting. Chatbot gets just one chance to get it right, unlike the Google search page which can provide several results ranked by relevance. Customers are willing to share more; The avg length of a search query in Google is less than 2 words however the length of the issue description in a chatbot is more than 3 words. Customers expect a crisp and direct resolution, preferably without navigating out of the bot window. In summary, your chatbot content strategy should be search engine agnostic, recommend resolution with high confidence and capture sufficient details around the issue that can map to available entities.
Companies often have a portfolio of products that span across a variety of support audiences like, End-users, Internal Professional services, and partners. Determining the right combination of product, issue types and audience your chatbot will service is key to the success of your program.
First, define your target product and the audience you want to transition into chatbot by thoroughly analyzing the following data points:
- Case details with Product, Issue type, Features/component..etc
- Support cost per case by product
- Pageviews support the site by product
- Product install base..etc
The outcome of your analysis should help identify the capacity, features, integrations, and placement of the bot.
Second, list specific issues or groups of issues that the chatbot will be programmed/trained to support during launch with high-resolution confidence. You can get that list by analyzing:
Case meta - Product, Category, Issue type, Disposition or Closure codes...etc Case subject and description Knowledge Base article attached (or Linked) - Organizations practicing Knowledge-Centered Service (KCS®) have article linking or attach as a standard practice in the case flow. These articles have primary and meta fields that contain rich information about the issue, environment, cause, and resolution.
Dispositions often provide shallow or no insight for this purpose, as they are kept very broad. A sample analysis of the case subject will provide better insights as long as they are in customer words. Avoid this if it is not the customer's own words. Article attach data is an excellent data point too, however, it should be used only when the Attach Quality or Link Accuracy is good.
With insights from steps one and two, you are all set to invite chatbot vendors for a demo. After every demo assess how the solution helps you support the identified audience, product, and issue group while determining if the cost is contained by the volume of cases you will offload from assisted support.
Before you start your pursuit for the perfect chatbot solution, keep in mind the following:
- Conduct a thorough assessment of your organizational strength
- Need and capabilities
- Pick an intersection from the Venn Diagram where you want your ideal solution to be and then invite vendors for demos.
Wish you all the very best in your chatbot journey!
HashoutTech is a Salesforce, Adobe, and LMI Partner with expertise in handling end-to-end implementation and integration for Service Cloud, Community Cloud, Chatbots, and Adobe Experience Manager.