How to Avoid a Failed Chatbot

IntraSee started implementing chatbot technology five years ago, before Oracle had a product in this space. They did work in usability and user experience, recognizing that websites were becoming increasingly difficult to use in the world of SaaS. Users were frustrated as they tried to access dozens of sites, and the old way of navigating enterprise information was not going to scale. IntraSee discerned that the future was in talking to our machines.

From their time and experience in the space, IntraSee has identified the common themes surrounding issues with chatbots and developed fixes or preventative methods to address those common problems.

After reading this post, you’ll have a game plan for maximizing your chances for success with chatbots.

The Trough of Disillusionment

This summer, Gartner released the 2022 Hype Cycle, showing chatbots in Phase 3, or the Trough of Disillusionment.

How did we get here?

Five or six years ago, we made huge advances in Artificial Intelligence. Everyone fell in love with the possibilities of AI.

This marked the beginning of the Hype Cycle. Everyone wanted to get a piece of the market. In the rush to get in on AI, companies developed low-quality bots. Furthermore, price-undercutting kept effective chatbot technology from reaching consumers.

Customers weren’t able to get an accurate view of good bots versus poor-quality ones, and many went with the cheapest as a result. However, the bots didn’t work well. Help desks were receiving just as many phone calls as they had before investing in AI. They were paying for a service that didn’t help their businesses.

Sales people were claiming AI is magic—that it learns on its own. “You just plug it in and let it go.” This was false.

This atmosphere confused the market and soured opinions about chatbot technology, landing chatbots in the Trough of Disillusionment.

The Promise Fell Short

In a recent survey by Candidate.ly, 57% of teams were using some kind of chatbot for recruiting. 71% in the study who implemented Chatbot technology thought their bots were average or below in effectiveness and ROI. If the implementer felt this, the user probably considered it much worse.

On a brighter note, the study found that, “…if technologies like chatbots were ‘well-designed and implemented,’ most customers would be happy to use them.”

Why is Bot Satisfaction so Low?

The top 4 reasons for low satisfaction with bots are listed below:

• No-Effort Narratives – False promises of an easy button that result in a false state of confidence
• No Breadth – Bots that only answer a small portion of the questions users need help with, leading to user frustration
• Lack of Personalization – Generic answers that feel like IVR, making users feel like the bot doesn’t know them and can’t meet their specific needs
• Bad Accuracy and Fake AI – Failure to understand or provide answers that are related to the question—especially bots that misunderstand organization-specific lingo

Conversational UI Market

While all bots tend to get lumped into the same category, there are nuances and capabilities that divide them into four distinct camps. IntraSee has categorized these as follows:

Stage 1 – FAQ Bot – Simple bot that manages simple, routine questions from consumers and internal users

Stage 2 – Vertical Chatbots – Chatbots are built into existing On-Prem or SaaS applications intended specifically to communicate about existing application functionality

Stage 3 – Enterprise Digital Assistant Platforms – Powerful platforms that need development. These platforms can achieve the greatest accuracy levels, but they must be built by the customer

Stage 4 – THE Conversational Enterprise – An out-of-the-box model that can be built upon if desired, but comes mostly built.

An example of fitting a chatbot into one of these categories or subcategories is PICASO. PICASO uses a Stage 3 platform (ODA) but delivers functionality in Stage 2. It fits a 2.5. SalesForce fits a Stage 2.

If you’re interested in implementing a chatbot, make sure you know which stage you’re looking for and only compare the apples within that specific bin.

Keys to Success

 There are certain keys to success that IntraSee has experienced in the chatbot field. For best practices, follow these steps:

Brand Your Bot

Your machine should have a name, visual identify, and a personality. It should be yours, not something run by ten other clients. This will help with adoption. Repeat the name over and over, in all your communication.

Below is a marketing sheet from IntraSee client, Loyola University that brands their bot, LUie, and helps customers to know what can be achieved when they interface with LUie.

Accuracy is Job #1

Accuracy must be measured. You have to know how your bot is performing. Measure data points, run test cases against it, eliminate NLP confusion. This is something you have to keep your eyes on constantly.

Here’s an example of an IntraSee client that was tracking accuracy and language accuracy. This client is in 60 countries with over 100,000 employees. Every country can have differences in the way the bot answers questions.

Seed & Cultivate

AI is not magic. Help your bot to evolve. It’s not a one-time project—it’s ongoing growth and increase in effectiveness.

Another IntraSee Client took this approach of viewing AI projects as evolution. They empowered their bots to learn from experience—deploying, using, learning, and evolving to meet customer needs. They’ve increased the visibility and breadth of their bot and reacted to user feedback to create a powerful, highly communicative conversational interface.

In the last 12 months, they’ve provided 120,000 answers through automated UI. To run a Helpdesk, the starting price per call is $19.

If you multiplied this client’s 120,000 answers by $19, you get an ROI of $2,280,000. Plus, it’s better for the customer because it’s available 24 hours per day in 100 different languages.

Plug into Data vs. Copying & Crawling

Crawlers don’t work. Crawlers are used by search engines, return bad results, and users don’t trust them. If you plug in data, you can answer questions precisely. Here’s another example of LUie from Loyola University:

Handle All the Requests

Become the concierge for the customer. Allow the customer to bring all of their questions to one chatbot. Below is an example from an IntraSee customer that breaks down questions by category and quantity. There is no clear winner in question category. There are a lot of tiny slices, called the long tail by IntraSee. Mind the long tail by being prepared to answer everything that a user might ask. Check in regularly to see which questions cannot be answered so that you can build out answers to those questions.

Create Magical Experiences

Look to add the unexpected by crafting experiences that make life easy. Keep the user’s objectives in mind. Don’t link to information. Don’t simply explain an answer. Actually assist the user in their quest.

Here’s an example of a magical experience with a credit card company that took the user about 60 seconds:

Meet IntraSee’s Digital Assistant (IDA)

Based on the time and expertise IntraSee has gained in the Chatbot market, they created their own Chatbot. IDA is the IntraSee Digital Assistant.

IDA runs on the Oracle Digital Assistant platform in the Oracle Cloud. IDA is unlike PICASO in that it is not PeopleSoft specific. It is meant to be enterprise-wide.

IDA can be used for higher education, prospective employees, retirees without logins, and unlimited use cases. Here are some key takeaways about IDA:

• Enterprise-grade digital assistant
• Uses machine learning AI
• Built on Oracle’s AI and Cloud
• Scalable to thousands of questions in over 100 languages
• Authenticated and non-authenticated chats
• Multi-channel capable (Web, Teams, SMS, etc.)
• Pre-build catalog of questions/skills/integrations
• Integration adapters such as PeopleSoft, HCM Cloud, Microsoft, ITSM, LMS
• Add questions, answers, and topics
• Conversational satisfaction surveys
• Role-specific & personalized answers
• Alerts, nudging, and suggestions
• Automated deployment and testing (no expensive staffing/consulting!)
• AI that gets smarter over time

VIEW FULL RECORDING

Visit Gideon Taylor at gideontaylor.com