Front Page

What are the capabilities available to tackle the issues and opportunities in customer service? It’s no secret that Machine Learning (where computers are given real-world examples of data to learn from, and they can then apply what they’ve learnt to new situations – just like humans do) has hit the mainstream, and is finding real business application across a range of industries and problems. One branch of Machine Learning called Natural Language Processing, in particular, has advanced spectacularly in recent years as deep learning neural network techniques have trickled down from image recognition, availability of computing power and data has exploded, meaning that language models trained on huge data sets like the whole of Wikipedia are available to be used commercially for natural language tasks. 

 This advance in Natural Language Processing capabilities brings the possibility of solving some of the challenges of Customer service, firstly by thinking about the way advisors interact with systems. The complexity of systems an advisor deals with screens, forms, processes, compliance – compounded by the number of systems and channels impacts training time, performance, quality, and satisfaction (both advisor and customer). What if some of this complexity could be taken away? What if systems, process, and compliance were invisible to the advisor? 

 This is where Conversational AI steps in, re-imagining the interface the customer service advisor uses as a conversational interface. Instead of screens and forms, dropdowns and fields, the advisor interacts with their systems conversationally in the same way they are dealing with the customer. They use a natural language interface to ask for the outcomes the customer is looking for and the conversational interface guides them through what is required to fulfil the need (to gather the information needed, to make the customer aware of the points required for compliance, to offer complementary services, to recommend next steps). The conversational interface is unifying – it doesn’t matter which “back end” is required to fulfil the action, the advisor sees only a single contiguous conversation, whilst on the other end of the interaction, the customer feels listened to. 

 A Natural Language interface is inherently intuitive, it’s the human way of communicating, learnt from birth. This means that training an advisor population to use a natural language interface is a very low effort activity. Processes or journeys that required specialist training and perhaps a call or contact hand off in the operation can be deployed to everyone through a natural language interface, reducing the need for handoffs and simplifying the capacity management planning for the operation. 

 Once a natural language interface to services, processes and journeys has been developed it remains evergreen, not subject to the trends and emerging best practices in UX. Admittedly there is a conversational design capability required to ensure the conversations flow well and are well written, but do this once and a great conversational design will remain great. As the vocabulary of the business expands (products and services are added, or changed) this is a matter of training the language model that sits at the heart of the natural language interface, rather than recoding. New services will require new development effort to build the fulfilment, but the natural language engine just needs to be retrained. 

 It is no great leap to think that an effective Natural language Interface being used by advisors for personal service could also be presented directly to customers through self-service channels in exactly the same way (there might be some “skills” that you only present to advisors). Build these interactions once and present them in every channel that is appropriate for customer service interaction. A Natural Language interface is inherently multi-channel, whether that channel is text (Webchat, Website, SMS, Async Messaging) or voice (IVR and speech to text transcription). 

 Presenting a skillful, transactional conversational interface to customers in self-service channels moves on from the chatbot model of Question and Answer or links to knowledge articles. The customer is typically looking to meet a need, rather than ask a question. Meeting the need directly without redirecting to a form on the website (or to a person) should be the goal of the interaction – move customers from asking “How do i…?” to saying “I want to…..”. Getting users to ask for the outcome they want to achieve, rather than the steps required to achieve it changes the dynamic of the interaction.  

 It is also very possible to operate a Natural Language Interface in an augmented personal service mode. Whereby the interface “listens” to the customer (either real-time transcribing voice to text, or text interaction) and predicts or recommends actions and automatically proposes the conversational interface skill and recognises key information passed without the advisor needing to transcribe. Direct access to customer interactions in this way also open the potential to detect sentiment and emotion to help the advisor and to provide insight into the way customers experience the service they are receiving

To find out more about how Utterworks think about the opportunity for Conversational AI in customer service take a look at our Fluent Converse product below

Fluent Converse

Have great conversations with customers. We believe that with Conversational AI you can simplify your interactions with customers, reduce process time, improve data quality, and increase customer and advisor satisfaction. It’s no secret that customer service is hard. Customer care systems are typically complex, particularly for enterprises in regulated industries with significant customer numbers and…

Continue Reading Fluent Converse

By that, we really mean great customer service is hard. Customer relationships can be expensive to establish and maintain, but are critical, especially for a commodity product or service where customer service is the only differentiator or cost to serve is a big component of price.   

Complexity grows with the number of channels offered to a customer and an organisation needs to balance the desire to encourage customers into lower-cost channels with a customer’s individual channel preference. Self-service really works for some customers, but others just want to talk to someone. 

Where customers want personal service, this function needs to be scaled (with people) to meet customer demand. You can’t have customers waiting for extended periods in telephony or webchat queues if you want to deliver a great experience, retain customer loyalty, and have them advocate for your brand. In complex operations, this presents a capacity planning challenge that needs to consider (amongst other things), channel options, average handling time, rate of resolution on first contact, breadth of services offered, skills of the customer service advisor population, routing and handoffs, training time, and seasonal fluctuations in demand.  

 Complexity, cost to serve, and customer satisfaction are intimately linked. Both a customer’s and a customer service advisor’s time are precious – a swift and successful conclusion to a conversation is in everyone’s interest, especially if it includes an appropriate cross or upsell. That this conversation with an advisor could be happening on multiple channels including voice, webchat, messaging or email adds to the complexity. 

 The complexity of the tools and technology landscape for advisors is worth specific consideration. An advisor working in an environment where they are required to “learn” the applications, interfaces, screens, forms, and fields required to meet a customer’s need will spend a significant portion of any interaction focusing on navigation and recall of the steps required to complete a process or journey. This will often require the support of a separate conversation guide and some sectors will have important compliance rules and procedures layered on top.  

 Organisations best endeavors to operationalise regulation and compliance often results in a number of tools that an advisor is responsible for combining in one conversation –  knowing the process, the system, and the regulations takes a toll on the advisor and generates a fear of getting it wrong in QA.  Often this results in a trade-off an advisor might make between data quality and speed (do they search for exactly the right category for this complaint in the 3-level hierarchy multi-value pick list, or pick the first one that looks reasonably close?) 

 The role of advisor should be focused on the human elements of customer service: empathy, emotion, understanding, patience, humor.  In many cases, advisors are simply acting as Human APIs to the systems that do the thing the customer needs. “The systems” have been designed in a way that needs to be trained and remembered and requires the advisor to lead the customer through a rigid process that often struggles to handle the subtleties of a real conversation – human conversations aren’t always linear – and advisors are often in the position of trying to complete a process or journey before being able to handle the new piece of information or request as it occurs to the customer.  

 It’s no great leap to suggest that the more advisors need to be trained in, and have their interactions governed by “the system” and its complexities, the less human and therefore less satisfying the work will be. This in turn leads to greater rates of churn in the operation and a corresponding detriment to productivity as new advisors are trained in the system. 

 The systems and operational landscape is often complicated by a legacy of multiple applications being required to serve the needs of the customer in an organisation offering multiple complimentary services. Each of these applications needs to be trained and learnt, and the overheads on the operation mean that not all advisors are trained in all services and a need for skills-based routing and handoffs is often the result. This can be magnified by transformation projects looking to add new services, or re platform applications or to simply upgrade to maintain support. 

 The technical implementation of the system also has a role to play in the efficiency of an interaction. Common architectural practice means that many systems, in the absence of any awareness of the specific customer’s need, pre-fetch lots of information just in case it is required. This makes the system very sensitive to performance fluctuation and demand spikes and often leads to the classic “I’m just waiting for my screen to load”. Throwing more capacity and money at the system isn’t necessarily going to solve this problem.

How conversational solutions can change your key metrics and the way you look at them.

The only consistent truth about customers is that they want what they need, quickly and efficiently, at a time that suits them and that they want to do that in a way that suits them and at their convenience . Traditional measures seek to manage the performance of a contact centre to be most efficient for the company and are based on measuring the deflection and then handling of contacts when not deflected. This is all fine with a traditional approach, but what if you could change the rules and give the customer what they want, when they want it in any channel, at a price that is much lower than current approaches. Then what if this approach could be used not just for customers but across your business.

At Utterworks we like to think of every contact as the start of a conversation and not the start of a process, this isn’t a rules-based approach but an intelligent conversation. Once a customer goes to the trouble of contacting your company, they want that interaction to give the outcome they want as quickly as possible, in a timescale that suites them, to their full satisfaction and also in a channel that works for them too. This interaction is an opportunity for companies to shine and they can do this without compromising quality, in fact they can improve quality and reduce cost to service by using the latest in Artificial intelligence and Machine learning. This article highlights five traditional key metrics and how the Fluent platform can improve the interaction.

1.         Average First Response Time (AFRT)

Average First Response Time is the time taken by customer support to send its very first response to a customer. Usually, the shorter the AFRT the better, and at Utterworks we believe that for all customers there should no queuing or waiting i.e. within seconds for voice, SMS, WhatsApp, messenger, social media etc.  Where the AI is able to, it should deal with all inquiries as quickly as the customer wants, this includes allowing customers to take as long as they need to respond. This takes away the importance of this metric as all first-time responses are immediate.

2.         Average handle time (AHT)

AHT measures the time spent by a support agent on handling a transaction with a customer. AHT takes into account talk time, hold time and any after-call tasks the agent has to perform.  This measure is fine for managing the performance of a call centre but doesn’t necessary help the customer, if the outcome isn’t reached to their full satisfaction for whatever reason.  At Utterworks we believe that the platform should manage the conversation with the customer at the speed and time that they want, so if the customer want to start a conversation, then pick it back up the following day because they are busy, then the platform should handle that, even if the customer wishes to change the channel. If you work in this way then you always respond to what the customer wants and no longer are driven by your own handling time, it is no longer a driver of performance and cost.

3.         First contact resolution rate (FCRR)

FCRR measures the proportion of tickets that are resolved on the first interaction. Customer satisfaction is closely linked to FCRR: the quicker the solution, the happier clients tend to be. We believe that most organisations do not track this rate fully, they track from first call, or maybe first text. We believe that this should start when a customer first visits the website to find an answer or make an enquiry, we believe this is the first opportunity to start a conversation with your customer, before queries become issues, and outcomes can be reached at first contact. We also believe that allowing customers to interact in the way that they prefer, gives more opportunity to have a conversation and proactively sell, resolve, or improve your relationship and grow your business.

4.         Net Promoter Score (NPS)

The NPS is an index that assesses the willingness of customers to recommend a brand or service to others. We believe that customers that are able to have conversations with businesses are happier, they are not frustrated by processes, organisational rules or support constraints. The Utterworks platform is built to ensure that context is understood, that information that is already known doesn’t need to be asked again and the customer can drive the direction of the conversation if they wish to. 

5. Customer Lifetime Value (LTV or CLV)

LTV is an estimate of the total amount a customer is likely to spend with a business, over the course of their lifetime. It is calculated by multiplying the average value of the purchase, by the number of times a customer may buy each year, and the average number of years they might remain a customer, over the cost to serve.  We believe that providing a high quality service builds deeper and longer relationships with customers, keeping the cost to serve very low and margins high.

Successful customer outcomes start from the very first visit to a website or first interaction with your business. By understanding the outcomes the customer is trying to achieve and satisfy that through the use of Artificial Intelligence our platform can transform your business. Moving businesses to become truly digital without the need to replace all of your existing platforms is truly transformation.

RPA is a pretty hot topic in the world of task automation, but the term is often misunderstood – mostly thanks to the vendor’s marketing teams. The image of a workforce of capable robots is an appealing one, but the reality is that the term robotic refers to the type of work being automated (the repetitive, mundane work that has an employee feeling like they are going through the motions robotically) not the automation itself. Most RPA solutions are much more like an Excel macro than a highly intelligent and dynamic workforce ready to jump on any task you give it.

Don’t get us wrong, RPA has a place in some organisations for sure, particularly where there is short term demand for automation to fill process holes, or to speed you to market before end to end automation is in place. We have used it ourselves to scale a product with a new advertising campaign when the manual invoicing and payment processes were a bottleneck to meeting the predicted demand.

Where RPA really doesn’t yet have an answer is in the area of judgement, particularly where the inputs are unstructured text (for example email). That’s where new NLP technologies come in. Enhancing RPA capabilities with the ability to interpret unstructured text and make a judgement (determine intent, extract important entities with context) really opens up the possibilities for process automation that RPA by itself doesn’t address.

Messaging is a great channel for customer service.

Often, the best kind of customer service is silent running, a service that delivers exactly what the customer expects when they expect it – with no ambiguities or interruptions. Clearly, however, that’s not always possible, products and services can be complex to deliver, may involve external parties, may be vulnerable to poor data quality, or to things a customer may inadvertently do. When you need to deliver customer service in circumstances where the product or service is not delivering what’s expected, the way you deliver that customer service is really important. Adding friction or frustration at the moment a customer is not getting what they need from the product or service they are paying for is not a recipe for a long-lived relationship. This is where channels become important.

Pushing customers to channels that require them to stop what they are doing and wait for you to respond to their needs in real-time feels less and less like a contemporary experience. Messaging and social are now ubiquitous and for many their preferred channels for interaction – you need to meet your customers there. The demographics are also shifting with more and more segments of the population using smartphones and messaging apps.

In the UK WhatsApp is the leading messaging platform, followed by Facebook Messanger and iMessage (Apple Business Messaging) with SMS still being used widely, though the volumes have been trending downward for the last couple of years.

Messaging as a customer service channel has a number of advantages, first and foremost the asynchronous nature allows the customer to have a customer service conversation on their terms, at their pace and at times convenient to them. Secondly, messaging channels can be made very secure and can be used to establish identity effectively and then maintain a meaningful long-running (and contextual) dialog with a customer. And thirdly, these channels are great opportunities for automation.

At Utterworks, we love messaging – we built these channels into our Fluent Converse platform from the very start and we think when you start to exploit the benefits you’ll love messaging and most importantly, your customers will too