Stats

When you’re an inventive company, you want your employees to be spending their time on fresh new ideas, not repetitive email queries. Rock Solid Knowledge used a customisable artificial intelligence (AI) driven solution to enable Garden Games customer service agents to spend their time on more creative tasks. 

When I saw our customer service team with notes on the desktop ready to copy and paste answers to the most common questions asked, I thought that if we could get a computer to be able to read and identify those common questions then the computer could answer. It sounded simple in my simple brain, after all I had read that AI would change the world and it was already here. The reality is different. There were plenty of 'experts' who could tell me what AI would eventually do, but I could not find real experts who could make it happen today. That was until a mutual connection suggested I talk to Andy.

John Cardy, co-founder and director, Garden Games

Background

Garden Games is an ecommerce company who sell and manufacture outdoor garden games and sports equipment via their two online shops: Big Game Hunters and Sports Ball Shop.

They use KHAOS Control, a business management solution, to manage all products, orders, and customer related data.

Garden Games employees interact with customers using the Zendesk help desk system, historically spending a lot of time answering simple or repetitive emails.

Building an AI helper bot

Garden Games felt that a machine powered by AI could answer these frequent requests. Rock Solid Knowledge coined this solution a “helper bot”, whose task would be to understand and automatically reply to help desk tickets.

The helper bot would work out the crux of the email, finding more information as needed, automatically returning a tailored, user-friendly reply. This allows employees to focus more on complex and creative tasks instead of manual email replies.

Prototype to reality

After an initial prototyping phase, Rock Solid Knowledge are happy to report that we have successfully built and delivered this helper bot. But how does it work?

Technical flow

First, we use Zendesk triggers to send all customer emails to our helper API, which carefully processes the email as follows:

  • Remove unwanted content, such as forwarded text
  • Split into sentences using an OpenNLP library
  • Send each sentence to LUIS for further processing

LUIS is where the AI magic happens, so let’s learn more about it.

Microsoft LUIS

LUIS is Microsoft’s cognitive service that uses machine learning to understand the intent of a sentence. LUIS can figure out what a customer wants to know, such as their order status or a price.

LUIS also picks out the useful parts of the sentence, called entities. Entities include order numbers, locations, and product names.

LUIS decides the sentiment (positivity or negativity) of each sentence. The helper bot can then adjust the ticket priority.

We discard any non-important intents such as disclaimers and sign-offs. But what if the customer wants to know where their order is, or another answer stored in KHAOS Control?

KHAOS Control

Since we know what the customer is asking about, we can automatically retrieve this information from KHAOS Control. For example, a customer’s order status, or whether a product is in stock. Previously, agents had to manually check this information.

We developed a sophisticated and secure approach using Microsoft’s Azure Service Bus, ensuring that we didn’t need to add any risk to the customer infrastructure.

Headless Umbraco

The bot now has a full set of intents, entities, an average sentiment, and any extra product or order information. This is all the data it needs to automatically craft a suitable response.

Rock Solid Knowledge are an Umbraco Gold Partner and knew how powerful the Umbraco content management system (CMS) would be in enabling agents to help us shape the bot replies.

We built a headless Umbraco site to use as a bot management hub, enabling customer service agents to:

  • Easily update global email replies
  • Manage threshold settings, such as % confidence required to reply to an intent
  • Determine the start of a forwarded email
  • Customise individual email replies according to the intent, shop and channel
  • Assign replies to specific specialist Zendesk agents
  • Set the priority of intents
  • Decide whether intents can be combined in one reply
  • Set intent replies to private or public

Zendesk ticket tweaking

With a fully tailored reply, the helper bot replies to the customer ticket via Zendesk itself. Depending on the Umbraco settings, replies can be public or private notes on the ticket. The bot can also automatically assign the ticket to helpdesk agents depending on the intent, channel, and shop. For example, it can assign purchase orders from Sports Ball Shop on eBay to a specific Zendesk agent if that is the business logic required.

Additionally, if the email is a complaint or very negative, the bot can set the Zendesk ticket priority to Urgent. This all works in harmony with any existing Zendesk triggers that the agent has already setup, and according to the negativity thresholds set in Umbraco.

Success

We’re proud to say that our AI helper bot has been live for Garden Games since January 2020, with many successful customer replies. We’ve fine-tuned and optimised it as we’ve learned more about the business domain, and we’re in a place where we have a fully-fledged product that can be customised and rolled out to other customers in the eCommerce space.

Since this launch, we’ve also built another AI-based Zendesk app, AimHappy Lite, due to launch Q1 2021. AimHappy Lite is a free app that will detect the sentiment of Zendesk tickets and set priority accordingly.

Looking for an AI solution for your business?

Looking for an AI solution for your business? Perhaps you’re a KHAOS Control or Zendesk user who wants a similar solution to this helper bot? Rock Solid Knowledge have a track record in excellent delivery, so we’d love to hear more about your requirements.

Andy knew what they could do and what they would have to develop, and his honesty convinced me that he could get the job done. Now we have a system that does what we asked, it reads enquiries, answers the ones it can and leaves the people here free to work on the more complex issues or on developing new outdoor toys and sporting play products. Andy and Emma turned out to be real experts at getting things done and the system they developed is like having an extra person here, but one who never takes a day off. It is just as we hoped it would be.

John Cardy, co-founder and director, Garden Games

Related articles

Powering up a customer service helper bot with AI

When you’re an inventive company, you want your employees to be spending their time on fresh new ideas, not repetitive queries. Learn how Rock Solid Knowledge built a customer service helper bot for Garden Games, powered by Umbraco, KHAOS Control and Microsoft LUIS.

Posted: Mon, 11 Jan 2021

Read Powering up a customer service helper bot with AI

Machine Learning Club

You might have heard of "machine learning". Maybe it was through an advert for an online course or someone complaining about the dreaded “Algorithm” that rules most of online life. We’ve been running a weekly Machine Learning Club to skill up our developers in this space

Posted: Wed, 16 Dec 2020

Read Machine Learning Club

Creation of an Alexa skill using an internal HR API

With the automation of HR management here at Rock Solid Knowledge we wanted to create a new way to remember who was meant to be in the office. So with a recent office purchase of an Amazon Echo, I decided to take on the development of an interactive program that could be invoked via Alexa...

Posted: Thu, 23 Nov 2017

Read Creation of an Alexa skill using an internal HR API