The HelloDone Platform

Before ingesting inbound queries into our NLP modules for analysis, the platform runs a number of key routing and filtering tasks. These include spam filters (to reduce unnecessary compute), language identification and identity & permissions status of the user. We match channel profiles to retailer profiles for personalised requests - ensuring a secure verification procedure where a customer is using the channel for the first time.

Once the inbound messages are filtered and routed, the platform applies the HelloDone NLU capabilities using a combination of two key systems, HD Luminate™ and HD Clarity™. These are two of the AI systems we have built specifically to resolve retail post-purchase queries. These might either be generic “What is your returns policy?” or personal “I want to change my delivery to Monday”. 
‍
The capabilities of this system far outweigh the sub-optimal CX of standard bots. This is about deep learning enabling a huge variety of ways for customers to get to a resolution, built with retail customers at the core. This is not about rule-based provision of response phrases  - there is no “bot-builder” here.

The datasets used for our deep learning models continually evolve based on real world inputs. This means the platform gets better and better at resolving the kind of queries customers have and the utterances used to describe them.

We have built a unique method to test our systems at scale. HD Torch uses AI-based test cohorts, built automatically and running autonomously as a way to test the capabilities of the entire HelloDone stack. Effectively we are using AI to test AI. This means we can point thousands of testers at our platform every hour, all realistically configured, including typos, slang and grammar traits, to check that we are providing the best possible CX to the real-world customers of our retail clients.

This is where the customer intent has been recognised and the separate pieces of data are pieced together as part of a full end-to-end two-way conversation flow to resolve customer needs. At this stage we know the customer, the product, the carrier...in fact we know all of the data points required to resolve customer issues as easily as possible. Where the customer has more complex queries or has already set up a customer service case, then we can hand them over to a human agent in seconds, all in the same channel.

Once we have built the most appropriate response for the customer as part of the ongoing conversation, this is then rendered according to the channel in question. HD Stitch™ is a service which we have built to transcode, centralise and create product image collages where appropriate, built on AI image processing technology.

All clients have 24/7 access to a bespoke data dashboard containing views of all the key metrics in real time (and across configurable time periods). Requested metrics include but are not limited to KPIs such as Resolution rate, Wismo to Order volume, Recognised vs Unrecognised Utterances, Top 10 conversation resolutions, New Customer Acquisition Rate, Click Through Rate, Agent Handover Ratio, Conversations per order, Messages per conversation, Geographic Heatmaps.