NLP (Natural Language Processing) Chatbot

sofa ASSISTANT

This project details the UX Design for an online sofa assistant.

Problem: Customising a couch is difficult in a standard online store.

Role: Product Designer

Incl. Strategy, UX/UI Design, Conversational Design & Writing, Graphic/Visual Design.

Overall Conversation Diagram Chart

Here an online assistant helps a user find a sofa. It details what is called the happy path (the easiest route for a customer to purchase a product) and it also illustrates sub-flows and a catch all.

Conversation Diagram Chart

The Conversation Diagram Chart EXTRACTS

Main Flow Extract

Secondary Flow

Catch-All

Main Conversation Flows

This is part of the main flow taken from the spreadsheets. It illustrates how buttons are used to direct the user towards the purchase of a couch uniquely accessorised to suit their needs and taste.

Here is a snapshot of one of the conversation flows from the Secondary Flow Product Details.

Conversation Spreadsheets

Download the Conversation Spreadsheet to view the Full Conversation Flows detailed above.

Download

EXTRACTS FROM THE BOTMOCK PROTOTYPE

The Main Flow or Happy Path.

The Secondary Flows for the Chatbot.

View the Overall Prototype Schematic

Here I built the user interface prototype with the Botmock layout/prototyping tool. It contains a number of flows, including the 'happy' path/main flow, sub-flows and a catch all. All the paths are detailed in the Conversation Spreadsheet above.

Botmock Layout

Prototype of Customer Journey's

Buying a couch

This is a full user journey for buying a couch.

USABILITY TESTING/WIZARD OF OZ TESTING

Research

The Wizard of Oz is a well known way of user testing AI prototypes/new builds. Above is a diagram of the test set-up used for the sofa bot with a CAI (conversational artificial intelligence) interface and a modelling system backend. Here is a link to a paper discussing the process.

Building a bot in SAP CAI

This bot uses conversational AI to interact with the user. It offers an easy flow for the user to follow similar to the prototype. The structure of the bot in SAP CAI requires intents and these are based around customer questions, the broader the base of customer questions in the bot the more likely the bot will be trained to handle user inputs accurately. The sample customer questions can be gathered during the Wizard of Oz testing phase described above.

Conversational Style of CAI

Greeting

Here are some examples of possible greetings in CAI.

Fallback

Here is an example of the fallback or catch all used by CAI.

Humour

Here are some examples of the funny side of the bot!

OVERALL LAYOUT OF CONVERSATIONAL ARTIFICIAL intelligencE

Intents

Designing/building customer intents involves using the information gathered from the Wizard of Oz test and focus group along with predicted sentences/queries customers might use.

Build

Intents

This is the overall schematic of the bot and how the intents interconnect.

Connect

Intents

Connect

It is possible to integrate CAI with social media tools and digital assistant software to get a really great user experience.

Logs

Monitoring

Connect

It is possible to see the types of intents/questions the user is asking the bot. This is an ideal way to pick up on the trends and customer needs. It's also a way to train the bot to deliver the correct answers quickly.

Monitoring

You can monitor how the bot is performing in a more visual way and create your offerings to meet customer demand.

Analysis

Monitoring

Looking at specific customers flows in detail can highlight particular choices around accessories. This can influence inventory control and predict future stock requirements.

Conclusion

Key Outcomes & Results

The prototype has been designed demonstrating the Artificial Reality aspect of the tool & gives the user the ability to customise their sofa. The next stage in this project would be to implement/build an actual AR tool to allow the user view their choice of sofa in their own home. Then integrate this AR tool into the CAI environment or interface this AR tool with the CAI environment.

What I learnt

Key Outcomes & Results

Designing a chatbot is slightly different to normal UX/UI design, it is important to work on the dialog flows and practice writing them in a way that is natural and replicates a normal conversation between two people, so substituting a bot for a person is an easy transition.
The animated gifs and other graphics are very important in the bot (for user engagement) so it worth having both strong conversation writing skills and graphic design skills available when developing a bot. The copy and visuals are very important to ensure the user understands the journey through the bot.
The bot can be used for many things like FAQ's and buying products but it can also be used in conjunction to the main task as a marketing tool, exposing the customer to more of the company brand.