Home

AI Chatbot

My role was to redesign Contact us page and help the Customer Experience team to improve customer service. I’ve partnered with researchers, developers and managers to enrich customer service. This was achieved by giving the customers' ability to solve their problems without the need to contact AO. We incorporated a chatbot on the page which lead to a more personalised and interactive experience for the users. This change helped the company save the cost of customer support.

AO, formerly known as Appliances Online, is an online-only retailer operating in the UK, Germany and the Netherlands. It specialises in household appliances and consumer electronics.

To comply with a non-disclosure agreement, I have omitted any confidential information. The information provided in this case study is my own and doesn’t necessarily reflect the views of AO.
A text-based chatbot on Contact us page

Background

This started as a project for a Hackathon that the company runs every year. Software developer Simon Cragg came up with the idea to create a voice assistant that enables our users to ask product related questions or for help and advice. During the Hackathon, we created a working prototype, and were given an award for this idea. The company was interested in investing in the future development of this idea. The business saw this as a potential to generate more sales and improve customer service by promoting self-help functionality on the website and save the cost of customer support.

Simon demoing AO Assistant
Second place at The AO Hack

The challenge

Customer’s demand to get help in different ways have increased. We’ve observed that quite often customers were asking questions on Contact us page like “Can I speak to someone on live chat?”. Customer Experience team wanted to accelerate operations with minimum effort and reduce cost by saving human resources.

AI-powered customer service

Initial business requirement was to save the cost of customer support by removing the phone number from the contact page. The company assumed that by removing the phone number on the page, it’d reduce phone calls to customer support. It turns out that the sale agents convert almost 40% of calls into sales, which means that by removing the phone number we don’t make the situation any better but in fact worse. I gathered evidence from sales agents that they convert phone calls to sales; this helped me to convince stakeholders not to remove the phone number from the page.

Abigail testing AO Assistant, it helps her to contact a manufacturer

Research

Input features

During formative usability testing we provided two types of inputs – speech and text. We wanted to understand what input channels our customers preferred when it came to interacting with the software. Some language understanding frameworks such as LUIS by Microsoft allow us to build voice or text assistants, so we needed to know which one we should focus on. Early lab sessions showed that our customers prefer to type rather than talk to the assistant.

I conducted competitors analysis and examined the existing page using Heuristic evaluation method and performance analysis tools like Lighthouse. I also used a behaviour analytics tool (ContentSquare) to understand how the page performs and find usability issues.

Insights

To gather more insights I was actively listening to customers at the call centre for a month. I’ve learned that usually customers don’t want to call for support and they don’t like being put on hold and waiting while agent helps them. People prefer to resolve issues themselves if self-help tools provided.

I requested search queries from search field on the page to understand what type of questions people ask and their needs. I also asked the Customer Experience team to provide me with a list of queries that customer support collects from phone and email conversations. Search queries and questions that people ask during calls highlighted vital topics that people are interested more. They need to know more regarding delivery, they need to know how to save money when purchasing from AO, and they need to contact a manufacturer, e.g. if their appliance is faulty.

User storyboard
Performance analysis

An average number of times people viewed Contact us page during a visit was 1.15, which means that people saw this page as the last step in their journey. Also, the average time spent on the page was much higher comparing to other pages.

Performance analysis showed that Contact us page receives over 40,000 visits a month and Customer centre gets around 12,000 calls just from that page. Majority of these calls agents generate into sales though, however many of them could be resolved automatically via live support software, e.g. chatbot.

Emma and Katie creating feature hypotheses

Ideation

I involved the stakeholders in the early stages of the project, as well as developers and copywriters through workshops and regular playbacks. I’ve run discovery, ideation and design studio workshops with the team. During the ideation session, we had a brilliant idea to notify users on the page about upcoming events that might affect deliveries.

Proto-personas

Collaboratively with the team, we drew a few personas based on our findings and customer segmentation insights provided by the Marketing team. This gave us a shared understanding of users' needs, requirements and the obstacles they faced.

Emphasise

After creating personas, we made assumptions about what users were trying to do based on behaviour analytics data. The data collected by Customer Experience team helped us understand what customers were saying. Collected insights helped us to gain an empathic understanding of the people we are designing for and the problems we are trying to solve. We learnt that people want to be informed, they want to do things on their terms and they want to save themselves some money.

Key features

After getting an understanding of what we were going to build, who the users are, what are their needs, and what we wanted to achieve, we prioritised a list of feature hypotheses. We prioritised and filtered ideas by voting and using the feature matrix method. By the end of the workshop, we had a list of useful features that we could test.

Collaborative design

I invited everyone to a workshop where we had a chance to sketch out their ideas and share them with others. Then at the end of this session, we discussed the sketches and voted for the best ideas. This exercise allowed me to gather draft wireframes that I later refined and used to produce testable prototypes.

Observations

We’ve learnt that informing users that they’re interacting with a chatbot; not a real person, prompts them to phrase their queries in a more direct command-like manner, which in turn results in reduced error rate.

Teamwork matters when it comes to design
Ideation workshop outcome

The outcome

We dropped the mailing address from the Contact us page because only around 12 people sent letters by post per month to customer support. This form of communication was very slow and not cost effective.

We split the project into smaller lightweight releases, this allowed us to avoid big mistakes, learn and iterate more often. Since launching the MVP with FAQ and top ten popular questions, the number of calls to customer support has decreased by 1.5%, saving the company extra staffing costs.

We observed that customers were very happy to interact with these new self-help features. There is still much more work to be done before releasing the chatbot, for example, FAQ details page.

The chatbot on the Contact us page will let the user find answers to their questions more efficiently, troubleshoot order-related issues and reduces the need to call customer support.

For confidentiality reasons, I have omitted the actual metrics.
Next project

Core Notes

A helpdesk software used by customer support agents and managers

Read more