Natural Language Processing

9 Oct 2018

Posted by Simon Richings


The UK customer services team of a leading global financial institution was inundated with 60,000 emails daily. Unsurprisingly it was taking up to two weeks for the team to manually categorise the emails for the right department to respond. This meant it was taking up to a month to get back to customer queries. Which naturally antagonised consumers further.

Emails were categorised into three sections: complaints, feedback, and queries. AnalogFolk was challenged to reduce the amount of time taken between receiving an email, categorising it and forwarding it to the correct department to reply.


The answer? Automation.

With machine learning, we taught computers a sample of scenarios and it then worked out what the general rule and concepts were from that sample, and then applied those rules and concepts to future scenarios.

For this global financial institution, we had a series of support emails that needed to be sorted into three queues. The solution was AI and natural language processing - reading the emails for intent, context, and entities, amongst other things - allowing people to crack on with the more complex tasks that require heightened emotional intelligence.

After all, automation is not about removing humans, it’s about putting people first.


We set about undergoing a four-step process:

Stage 1: Emotional analysis

Firstly, we used emotion algorithms to screen a database of emails. Everything from rage to loathing to grief. Ouch. This highlighted the levels and intensity of emotion so we could gauge the level of complaint.

Stage 2: Identify themes

We then identified recurring themes that were driving customer contact. We highlighted the scale of each theme, levels of intensity and how they matched with the existing three categories. We soon found out that there were in fact 12 common themes, not just the three previously identified.

Stage 3: Tech discovery

Next, discovery time. Using natural language processing (NLP), we were able to discern the intent of an email according to the business rules of the company and automatically triage emails. We built a prototype that tested the hypothesis and ensure the proposal of using NLP to automate the email process was possible.

Stage 4: Recommendation

The prototype validated the use of AI for the automation of the email sorting, and is now being used by the customer care centre. AnalogFolk also provided a recommendation and road map on how the initial pilot could be scaled to start automated replies to customer queries.


This financial institution is leading the way to a new and exciting era in which computers are more human and banks connect directly with their customers.

The initial pilot reduced the time taken to categorise the emails by 500%.

Our recommendation and road map is now being implemented by the internal R&D team with our team providing technical consultancy.

The working model of using AnalogFolk to develop and validate a hypothesis through rapid prototyping before handing it over to internal development teams was a success story for the client innovation team. Allowing them to make their ideas a reality, at a much faster rate than previously possible. 

Simon Richings
Posted by Simon Richings

Simon is responsible for the running of AnalogFolk’s creative output. He thinks part strategically, part laterally and part like a normal human – a combination that’s turned out to be particularly useful when it comes to creating compelling, award-winning digital work. He’s won Gold at Cannes, the One Show, the Clios, the London International Awards and more, but is also very proud of the NMA Effectiveness Grand Prix on his shelf. Formerly the Creative Director at Tribal DDB London, Simon has developed innovative ideas for Volkswagen, Hasbro, Philips, Guinness, Budweiser and the FT. Despite training as a designer, he’s a writer at heart and continually bangs on about storytelling and moving people. When he’s not scribbling on Post-it notes or pointing at other people’s screens while saying something insightful, he’s probably X-Boxing, exploring museums with his daughter or running 2.6 miles at a pace that’s more or less average for someone of his age and build.

View profile
Read next

Keep your finger on the pulse

Join our mailing list to be the first to hear about cool opportunities, hot events and more.

Password Reset

Please let us know your email address and we'll send a new password straight to your inbox.