In the second article of our AI series, we give an inside look into how we use AI here at Mesaic, and our unique vision for AI value creation in the conversational commerce space.
Now that we understand the evolution of AI and why humans are so keen to develop it, the second article of our AI series will narrow the focus on what AI means to us at Mesaic: a tool for maximum customer satisfaction.
At Mesaic, we are committed to revolutionizing the customer service model in order to maximize customer engagement, satisfaction, and loyalty. Our vision for AI focuses on balancing human and machine intelligence for maximum business success. It augments traditional commerce models by leveraging mobile technology and AI, thus permitting us to build innovative and agile solutions, from back-end to front-end.
Our solutions automate and optimise where appropriate, letting humans focus on the complex issues that most need their attention. To that end, we prioritise the synthesis of business process automation with personalization. Our vision also empowers employees by automating repetitive tasks, while delivering personalized experiences that build long-term customer loyalty. We design our approach to be as near ‘future-proof’ as possible so that our clients always know they are supported by technology that keeps pace with the market, trends in technological development, and their individual needs.
Ok, now we are ready to dive in. This article moves chronologically through a selection of use-case examples along the customer journey timeline that demonstrate how our AI vision creates value for our clients and their end-users, with an emphasis on two main themes: automation and personalization.
The visual below depicts how this process unfolds on our platform and the three main stages: engagement, checkout, and fulfillment.
The first set of use cases pertain to the initial stage of an end-user’s journey, i.e. the beginning of the conversation thread between a client (also referred to as a ‘tenant’ hereafter) and the end-user. A majority of the AI applications that specifically enhance end-user experience focus on this phase.
From the very beginning of a conversation, our platform engages the potential customer efficiently and personally, in order to understand his or her needs as quickly as possible. It is now common practice for a chatbot to conduct the initial stages of customer communications due to their operational benefits, e.g. decreased cost and increased efficiency. However, using chatbots also risks reducing the quality of customer engagement. To reduce this risk, we at Mesaic, train our bots to both compensate for their lack of a ‘human touch’ and learn how to personalize their interactions with individual customers. All of this effort serves to help us determine when customers want to interact with a bot, and when they don’t.
The foundation of our AI-powered bots is extensive natural language processing (NLP) training for intent recognition. We supplement this training with logic that detects when the complexity of a conversation surpasses the intelligence of the bot. In such cases, the bot triggers the need for human intervention and a ‘handoff’ from the bot to a customer service agent ensues. Using AI to conduct the initial, basic information-gathering steps of a customer interaction saves human time and attention, thereby allowing customer service representatives to focus on the complex issues that AI cannot solve on its own.
As mentioned in a previous blog article, chatbots can also be useful for personalization of service. For example, we enable our bot to adapt its ‘personality’ to that of the customer, or to the best response personality given the customer’s tone. So if the customer demonstrates anger, it would be best for the chatbot to adapt its tone to be more conciliatory or empathetic in order to resolve the customer’s anger. Over time, the technology would learn which response tones work best for different contexts, customers, and conversations, in order to always make end-users feel they are understood and valued.
Additionally, we integrate computer vision and live speech analysis into the platform so that we can provide fast and comprehensive intent recognition, and more personal customer service. With computer vision as a feature, the product could match content from the internet with a customer’s interests for the purpose of customer engagement outside the transaction setting. We have mentioned the importance of this in a previous article, highlighting that businesses need to anticipate and meet the future needs of the customer. Such engagement reminds end-users that the company (our client) values them not only for their business, but also as people with hobbies, interests, and passions. This personal touch in an age of mass digitalisation can make the difference between meeting customer loyalty goals, and not.
AI-powered speech analysis means enabling voice messaging on the platform. Voice messages have become incredibly popular; in fact, an average of 200 million voice messages are exchanged every day on WhatsApp alone. Voice messages save time relative to texting and give messages a personal touch that only human voice can convey. Additionally, voice messages leave much less room for tonal ambiguity than text messages. Thus, they take a lot of the guesswork out of determining the current emotional state of the customer. For example, it can often be difficult to determine if a customer is annoyed or simply prefers to send concise messages; knowing which can make a big difference in maintaining that customer’s loyalty. Enabling tenants, service partners, and customers to use voice messages will reap the benefits of speaking to a human representative at a call center without the hassle of actually calling a customer service helpline.
A final example of AI applications in the pre-sale phase of the customer journey demonstrates a specific application of a highly generalizable feature. Imagine the client is a car dealership and that winter is approaching. The Mesaic Conversational Operating Platform for this tenant could be programmed to routinely check the weather forecast and search the product catalogue for products or services that are commonly needed in the given weather conditions, e.g. winter tires that can handle driving on ice and snow-covered roads. Then, the platform could query the client’s customer relationship management (CRM) for a list of customers who are most likely to need the service, and then send them a message asking if they would be interested in placing an order for it. The platform would accomplish all of this automatically, naturally with the option for external (human) approval. This is a very specific example but the fundamental logic could be customised to support virtually any product recommendation or upsell function, thus ensuring that customers get exactly what they need, when they need it.
The next set of use cases pertain to the section of the customer journey after a customer has confirmed a service request and before it is fulfilled.
Immediately after a customer places an order, the tenant must allocate a service partner to the order, one that ideally maximises the probability of high customer satisfaction. This decision is a difficult optimisation problem, as it depends on several factors including the geographic location of the customer, urgency of the order, and the availability of each servicer, among several others. In mathematics, such problems are referred to as linear programming tasks and are notoriously tricky to solve. Nonetheless, machine learning (ML) algorithms equipped with the proper input data can solve this multivariate optimisation problem quickly and with high accuracy.
AI can also be used to supplement the service partner’s expertise. As the partner completes an order, he or she can input status updates and details relevant to the order. Before final completion of the order, the system can automatically ensure that the partner has not missed any upsell opportunities or loose ends that could result in subsequent customer dissatisfaction. Note: the AI is not there to replace the servicer’s role, simply to augment it.
After the servicer successfully completes the order, the conversation thread enters the post-sale fulfillment phase. At this point, the customer receives a prompt to rate his or her experience with the tenant and service partner on a scale of 1 to 5, with a value of 1 corresponding to low satisfaction and 5 corresponding to high satisfaction. However, ratings are inherently subjective and exhibit significant variance across customers. A low rating could result from a large number of different factors, but the scale has no way of indicating which variables contributed most significantly to the customer’s dissatisfaction.
One way to collect more comprehensive feedback might include issuing surveys or requesting written comments from customers after a transaction. However, these methods do not guarantee higher quality feedback, as they require significantly more time to complete and customers may not feel motivated to complete them unless they are either very satisfied or very dissatisfied with their experience.
Thankfully, Mesaic is developing another way to strengthen the quality of ratings: combining them with other interaction and performance statistics collected automatically by the platform. In this way, AI adds value by performing additional analyses of the conversation, e.g. the average response time, sentiment level fluctuations, average conversation length, etc., combining these metrics with the base customer rating, and feeding everything back to the client’s CRM. This feedback loop leads to the improvement of future interactions with this customer, but also those of all other customers. The platform automatically learns from and synthesizes every piece of information it receives so that it never misses an opportunity to improve user experience. So, if we know that some customers assigned to Partner A experience relatively long message delays, demonstrate lower than normal sentiment scores and also lower ratings, we can deduce that the low ratings stem from negative ground truths of the interaction, and not the limitations of the rating scale. The system can then detect patterns of ratings and interaction statistics that differentiate clusters of customers who are generally more likely to give good ratings regardless of the interaction statistics, those who are generally more likely to give bad ratings, and those in the middle.
In any case, this feature allows clients to draw better insights from rating data, quickly diagnose potential issues based on the presence or absence of certain trends in the data, and learn from successful cases. Importantly, the platform archives all of this information and uses the accumulation of knowledge to continuously fine-tune the matching algorithm. As a result, the platform learns how to best optimize pairings of service partners to customers in order to maximise customer satisfaction and support a seamless customer journey for the client.
So far, we have covered only the beginning of AI-powered value creation at Mesaic. There is much more in store, so if what we have discussed in this article has made you curious, get in touch to find out what Mesaic can do for you and your customers.
The next and final article in our AI series will switch gears from the value AI can generate for business to the ethical concerns raised by the technology - because for us at Mesaic, no conversation about tech is complete without an evaluation of its ethical impact.