Portfolio - Cazoo chatbot | Kirti's Portfolio

Redefining the chatbot capabilities with YEXT to provide relevant content to the customer for pre-sale queries.

Cazoo Chatbot

The problem

How do we resolve pre-sale customer enquiries through the chatbot and allow them to self-serve without the need for them to call CS, increasing the chatbot containment?

The Cazoo CS team were receiving a large amount of pre-sale customer queries the chatbot that could be self-served by customers online. Customers were requesting to transfer to an agent on 'I'm looking to buy a Cazoo car' dialogue, which had a lot of of content available on the site.

Research • Service design • Iterative UX design • Prototyping • Usability testing • UI design

Context

The challenge was that every transfer to an agent was costing Cazoo money and time. CS were extremely busy with the after-sales cases that needed to be dealt with on case by case basis. The containment rate for the pre-sale dialogue was currently only at 23.8% .

Containment rate (the % of interactions that don't reach a human agent)

23.8%

Total transfer requests for pre-order

60%

Transfer requests over time

43.34%

My role

It was necessary initially to establish:

Which dialogs are causing the most unsatisfied results
Dialogues that had transfers requested
User input messages in the chat before transfer for 'I'm looking to buy a Cazoo car'
Gaps in the current UX flow to initiate self-serve

Working collaboratively with the content designer, it was essential to analyse transcripts within I'm looking to buy a car to identify where content opportunities can be expanded. Also, investigate cause for the bot confusion spike.

The process

Top queries: Looking into the data we found that the queries that were causing low satisfaction and transfers were mostly around 'viewing a car', 'part exchange', 'finance', 'valuation' and 'features and specs', which mostly had help content available on the support site.

Lack of content: The chatbot transcripts revealed that there were a lot of content gaps that could help the bot serve relevant responses, but updating them manually could be a time-consuming process.

UX flow issues: The current UX flow was providing a response to most queries with a dialogue and a link to an article, but that one article wasn't necessarily answering the question. Also, the user captcha for e.g. on the finance flow wasn't providing any sub-categories or support content with the user wanting to speak to an agent.

The research findings

The problem discovery

To gauge how other e-commerce sites use the bot to serve responses and content, I looked at several websites to get an understanding of what would benefit us and our customers. These were Intercom, Squarespace, Zoom, Samsung, apple, and Tidio. With almost all of them, the common pattern was that help content and support links were being surfaced in bot responses to aid customers with the queries and self serve.

The findings above led me to look into:

Find ways to surface more help articles to the user so they have a better opportunity to self-serve
Find ways to serve effective bot responses on user input captcha at several stages in the flow such as for finance to increase containment
Amend the UX flow so the users have an opportunity to go back to select a different option so its not a linear process

Market analysis

Technical constraints

The current chatbot was from the Salesforce platform, which had limitations in its capabilities to serve content. Therefore working collaboratively with engineers, I looked into YEXT, which was being used for the site search as the possible solution to integrate with and maximise the bot's output.

Concepts - Usability testing

Hypothesis: If the chatbot can serve content directly from the support site integrated with Yext, we should be able to support most pre-delivery queries increasing containment.

Using a similar approach to other e-commerce sites, I designed a concept with a prototype for usability testing to validate the hypothesis. The positive feedback validated our thinking that customers were happy to self-serve by receiving help content by the bot before wanting to talk to an agent.

The solution discovery

Final Design output

The concept from the usability testing was simplified. This enable the team to launch the product with YEXT integration so that it can easily be measured for impact.

Improved UX flow allowing users to receive multiple recommended support articles.
Improved understanding of the user input captcha by the bot to provide an enhanced user experience.
Allowing the customers to go back to the main menu aiding in selection and self-serve.
Improved UI consistent with the brand and design system

The solution - YEXT integrated chatbot

This new solution was launched as an A/B test so the impact can be measured and compared with the current solution. The next iteration is now in discovery looking at how we can embed support within the chatbot so customers can access help content straight away, which should aid in reducing the CS contacts further.

The next iteration is now in discovery looking at how we can embed support within the chatbot so customers can access help content straight away, which should aid in reducing the CS contacts further.

Success metrics

(30 day period)

contact deflections (containment)

250

CS cost saving of

-£875

Containment increased by

10%

Kirti Devi Ram