Behind the scenes: how we implemented conversational AI

As a live chat and bot provider, we often consult customers on conversational AI. To do so effectively, we believe it’s important to practice what we preach.

So, we’re sharing a behind the scenes use case. Here’s a look inside our own application of conversational AI here at Parker Software HQ.

Setting the scene

WhosOn has been the UK frontrunner in chat technology since 2003. More recently, however, we’ve introduced chatbots into the product – both available to customers as well as deployed internally for use on our own group websites.

Traditionally, chatbots help customers navigating a website. Though this undoubtedly assists human live chat operators – easing the load of FAQ and routine queries – it does so indirectly.

Operators and bots work in parallel on a help desk, rather than intersecting. And it was this that we believed needed a radical review.

As long-time advocates and creators of smart chat software, we wanted to push bot boundaries. We’re ever striving to find new and innovative ways to incorporate conversational AI, and this branched relationship between bots and operators presented a prime place to do so.

Starting at home, in our own help desk.

Identifying internal opportunity

For us, a common time sink is finding out whether customers contacting us for technical assistance have an up-to-date support and maintenance contract in place.

(We offer on-premises installations of our products as well as perpetual licences, so many long-standing customers are not subscribed for a current support package.)

If the customer has support in place, they are eligible for live help from a technician on the spot. If not, we will instead raise a ticket and investigate within a set time period.

So, to check support status, our operators must first take a software serial key from the customer who has reached out. Then, they search our internal database using this key to ascertain the customer’s level of support.

This is a constant and time-eating process that happens each and every time a customer seeks support from one of our websites. It delays the chat, delays our ability to advise customers appropriately, and delays operators with manual admin.

Here, then, is where we saw a ripe opportunity to optimise with conversational AI.

Introducing hybrid chat

To reduce this persistent help desk bottleneck, we began work on a hybrid chat project. That is, a bot capable of supporting our visitors and our operators alike.

First, we defined a custom string that teaches the bot to recognise the format of our software serial keys.

Then, we built a custom integration into our internal database. Once the bot was trained and plugged into our database, we set it to monitor all internal chat sessions on the alert for serial keys.

Whenever a key is detected within a chat session, the bot queries our database, retrieves the relevant data and returns it as a pre-formatted ‘whisper’ message to the operator.

So, the agent gets a message visible only to them, that contains:

  • – The company the serial key is registered to
  • – The name and email of the serial key owner
  • – The key registration date
  • – Details of the customer’s support agreement and its expiry

This happens without delay, within a second. Conversational flow suffers zero disruption.

The operator can respond right away, using the information fed from the bot, to give the appropriate support.

The customer, meanwhile, gets a rapid reply – oblivious to the bot’s ‘whisper’ message on the operator’s side of the chat.

A closer look

A universally useful resource

Importantly, the bot can still perform its traditional role of assisting website visitors with conversational support.

But it also does the same for our company live chat operators.

Our hybrid bot is now there to support every conversational participant – offering pre-programmed responses to visitors as well as fetching and sending needed data from our systems to operators.

The bot works in every direction to speed and streamline chat sessions, for a help desk blending human support with AI speed.

The business impact of hybrid chat

Prior to developing the bot’s hybrid capabilities, we measured the time it took for an operator to respond to a typical message requiring a serial key check. On average, it took 37 seconds for an operator to:

  • – Copy a serial key from a chat session
  • – Switch apps to the database
  • – Run a manual search
  • – Locate the correct record
  • – Open it and find (then read) support contract details
  • – Switch back to WhosOn
  • – Send the appropriate response

And, out of the 9,493 chat requests we receive per annum, some 60% are from customers requesting support. (Necessitating a serial key search.)

This means that, over the course of a year, a bot performing this one service alone saves us over 58 hours.

This not to mention the time saved from more traditional customer FAQ answering.

So successful is hybrid chat, in fact, that we have since made the service available to our customers.

Now, customers can train their own bot to monitor chats and define their own custom strings – be it a post code, a ticket number, or any other data that the operator must interrogate – and have the bot ‘tag-team’ relevant information back into the chat.

A new chat frontier for conversational AI

Hybrid chat offers the best of both worlds. The warmth and scope of human live chat agents, accelerated by the pace and proficiency of chatbot support.

All with conversational AI working from every angle, to support every party.

We call this the new frontier.

Useful links