Improving the quality and accuracy of chatbots
How to successfully leverage partnerships to the benefit of the client?
Invenica and Volume (with the use of their product QBox) have each been able to advance the digital transformation journey of one of our shared customers (the leading global telecoms operator referred to here). This customers wants to automate 60% of support requests by 2025; with the addition of QBox, this is going to be possible.
Building on Invenica’s work, QBox has been helping optimise the NLP training data (across multiple providers) by:
-
Extracting valuable performance data
-
Offering ongoing monitoring into key metrics of customers’ automations (enabling them to dramatically increase the number of support requests that can be automated)
-
Facilitating enhanced customer support
-
Realising significant operational savings
QBox is a piece of conversational AI middleware. Once a roadmap for the transformation of the customer-support function has been drawn up, QBox helps test, understand and fix NLP training data. It also provides ongoing monitoring support after the automation and AI customer agents have been introduced.
Chatbots, voicebots, chat engines and digital humans all require NLP. Once the conversational AI has been devised and integrated by Invenica, QBox offers incredibly deep and never-seen-before analytics that enhance a customer’s investment into conversational AI. By including QBox or the right customer, Invenica can offer a full-service solution, from inception to optimisation.
Invenica - QBox partnership
We have partnered with Volume to use their product QBox to help our clients optimise their conversational AI platforms
We have recently partnered with Volume, who have developed and brought to market a new and exciting conversational AI middleware tool: QBox.
This is a performance tool that improves the accuracy of a company’s conversational AI real estate. It tests your chatbot with your chosen provider and helps you optimise your training data. Our partnership promises to offer a full-service solution, to significantly enhance a company’s investment into conversational AI with the potential to realise very significant cost savings as a result of operational efficiencies.
QBox enables you to improve enterprise chatbots’ accuracy and performance in a matter of minutes, giving you 100% confidence to retrain and deploy. It analyses and benchmarks chatbot training data by visualising and understanding where it does and doesn’t perform, and why. This insight enables chatbot developers and NLP data modellers to make informed decisions about how they develop the performance and scale of their chatbots. Watch their video here!
QBox overcomes many of the challenges associated with managing your chatbot ecosystem and scaling your knowledge pool without impacting performance as your chatbot real estate grows.
QBox is NLP provider-agnostic: across all the major providers, it can improve the three global KPIs of clarity, correctness, and confidence of your models.
Our partnership introduces the potential to enhance companies’ investment in conversational AI across all sectors to make significant operational savings. The conversational AI market is growing significantly – and with strong partnerships like ours with Volume, we are very well placed to introduce the infrastructure and technology to help companies reach and exceed their targets, thereby potentially saving significant operational capital.
“Conversational AI is an amazing way to interact with a system or an application in a manner that most humans are used to. Making that conversation flow organic, proficient and accurate is a challenge that QBox helps solve.”
Gareth Mee, CEO, Invenica
When it comes to getting in touch with companies, customers really don’t want to be left hanging on the line. They want a rapid response to their queries - and they’re happy to take their business elsewhere when they don’t get this.
According to research from Hubspot, nine out of ten customers rate an ‘immediate’ response as important when they have a customer service question. This has led to a situation where two-thirds of customers have become ‘serial switchers’, changing brands due to a poor experience – something that’s costing businesses more than $75bn a year.
It’s in this context a Tier 1 bank in the Americas turned to Invenica for help. The firm’s call centre handles millions of customer calls every year. However, until recently, they relied solely on a traditional dual-tone multi-frequency (DTMF) interactive voice response (IVR), a legacy touch-tone system that was having a negative impact on the bank’s customer experience. The system’s menu tree was often difficult to use, causing customers to get lost and even abandon their calls. The bank needed a new system that could improve the customer experience and drive customer retention.
The solution was an AI conversational language platform and customer service virtual agent (VA) able to help customers find answers to their enquiries as quickly as possible – both in free speech and text. The challenge with producing and deploying such a system was that it needed to handle significant levels of complexity. The bank had set high benchmarks for performance in semantic accuracy and task completion. It also required the VA to be capable of serving customers speaking English and local dialect in a mixed-language model – something no contact centre had ever implemented before.
What we did
To develop the solution, Invenica used the technical expertise of its DisruptCX (DCX) practice, which specialises in supplying emerging technology to contact centres. Following a review of the bank’s existing systems, DCX swiftly identified a conversation language platform and VA as the solution best suited to meet the project’s needs.
From there, DCX selected a platform and designed and deployed a VA able to retain context around multiple customer queries in a single conversation. This was supported by the use of machine learning to continuously learn new patterns and improve predictability – allowing the bank to serve customers who are unable to express what they want as clearly as others.
DCX then developed and tested algorithms for dealing with the uncertainty of voice input. So, for example, where the confidence in the intent of the voice input is below 90%, the engine provides up to five inputs that it believes could be accurate. These are sorted by likelihood, with mathematical checks used to confirm which input is the most likely to be correct. Invenica added to this with an application that helped the platform to learn and improve the bank’s self-service conversion rates.
Invenica leveraged its previous R&D project experience – along with its expertise in the development of microservices – to implement the capability for continuous improvement, allowing the platform to be extended in novel ways. In developing a modern microservice ecosystem of functional components, Invenica made it easier for the platform to respond to the bank’s changing requirements while minimising disruption to the customer experience.
What we delivered
With so much complexity and uncertainty involved, the Bank needed a solution based on a scientific and research-driven approach. That’s why Invenica and DCX delivered an AI-powered conversational language platform and VA that could not only handle customer enquiries more efficiently but also learn from each one. Every conversation the platform now has with customers improves future outcomes as it encounters new scenarios and ever-more complex interactions.
The platform meets all of the bank’s KPIs for intent accuracy and task completion rates. It also provides multi-language support for customers speaking English and local dialects. Ultimately, this provides the bank with the best possible platform to improve its customer experience, drive retention rates and cut operational costs by taking the pressure off its existing systems and contact centre agents.
Good news travels, but bad news travels quicker – and further. Three-quarters of people tell at least one person about a poor customer service experience. And, now, with the reach and immediacy of social media, how firms respond to and manage issues plays a huge role in building brand image and perception.
In a recent survey, 56% of customers said the most important thing in customer service is that their issue is resolved satisfactorily. It’s no surprise then that the same survey found that eight out of ten customers spend more with companies that give them a good customer experience.
This means that operators who haven’t invested in the tools to improve their service experience (whether customer- or agent-side) become vulnerable to customer attrition. There’s also the likelihood of falling behind competitors when it comes to overall innovation. On the other side of the story, Telcos who already have the technology to deliver a smooth customer experience are freed up to focus on improvements in agility and performance – all of which add to their brand image.
It was in this context that the customer service function of a leading global telecoms operator was struggling to deliver the right level of value to customers. As well as causing a high operating cost of costumer care, this situation resulted in a low customer satisfaction scores (NPS, CSAT, FRT, AHT, etc.) and poor reviews on sites like Trustpilot – which, in turn, influenced the company’s brand perception.
Offering a wide variety of products and services, the operator required a complex suite of tools to support its customer care teams. It also struggled to make the most of its data. The sheer amount collected and exchanged had created huge sets of information that no one had been able to curate or analyse.
On top of this, the lack of performance data made it difficult to identify the key areas for improvement. It was hard to tell which product support areas were underperforming – and what tools they needed to improve. And this made it near impossible to take a targeted approach to the problem.
What we did
For Invenica, the first step to solving the problem was to identify the core challenges across the relevant teams – including customer care teams onshore and offshore and those dedicated to specific markets. This was achieved by surveying the relevant stakeholders: customer agents, sales representatives, managers, product owners and customers. The survey was important for gathering insights, determining actual performance and setting targets for a healthy customer satisfaction index.
Invenica then designed a roadmap for the transformation of the company’s customer support function – featuring steps like the rationalisation of tools, plus the introduction of automation and AI customer agents. The next step was to carry out a rationalisation of the firm’s customer care tools. This process involved the creation of new customer journeys and the streamlining of data – making it easier for agents to respond to queries effectively and efficiently. Invenica designed and developed the missing components and assembled the customer care platform, taking care to integrate it with existing legacy systems to provide functions such as identity, billing and supply chain management.
What we delivered
The changes and additions made by Invenica removed some huge barriers to performance (such as slow information discovery, a lack of insight from customer history and defragmented sets of tools for different products and services). They also increased the capacity to resolve customer queries through the automation of support tasks. All in all, the process reduced handling times, increased the number of solved tickets and delivered a better overall quality of service.
Taking an agile approach, Invenica transformed the tools incrementally, introducing time and cost savings to go with the improved customer experience. It also integrated new tools with the right systems where required – allowing agents to seamlessly discover and access the data and information they need.
This transformation extended into a review and reform of operating models. It also featured the creation of adequate training and documentation materials to help users get to grips with new features, including:
• An intelligent search engine
• Integration with telephony
• Identity management
• Chat engines (text/voice)
The team implemented QBox to continuously monitor and improve the accuracy of chatbot, you can learn more about this here.
With the tools in place, Invenica set new KPIs to match user expectations. The teams are now constantly measuring the customer satisfaction index to make sure their results reflect the vision and targets set by their leadership.
It’s tempting to think that the sales process ends when a customer makes a purchase. However, it’s vital for firms to show the value they can provide after the sale is completed. After all, customers will have just been through a whole process to choose a supplier, so they’ll be keen to see what other products are on offer.
This is where recommendation engines come in. Using technologies like AI and machine learning, they analyse customer data to build accurate, individualised profiles of the people making purchases. Customers see a targeted selection of relevant products they might be interested in – making it easier to find what they need and buy it. Companies gain a wealth of insight that ultimately increases the likelihood of future sales.
Source:https://www.salesforce.com/solutions/industries/retail/resources/product-recommendation-engines/
With the customer data collected from these profiles, companies can build a picture of the content and solutions a specific customer requires – making it easier for sales advisors to recommend the products that meet those needs. Intelligent product recommendation also allows for natural, logical upsell and cross-sell opportunities. Clients, through their behaviour and history, demonstrate interest – and the product recommendation tool automatically pairs that behaviour with the right suggestions.
Small transactions become larger ones, with an increase in average basket size, and clients who might not have been on the path to making a purchase suddenly find themselves interested in doing so. As an example of how powerful this can be, an Accenture report highlighted how 91% of consumers are more likely to shop with brands who recognize, remember, and provide relevant offers and recommendations.
/What we delivered
To begin with, the Invenica team implemented and trained a machine learning model to predict the products and services that any given customer is most likely to purchase. It did this based on:
/// The customer’s profile (and related behaviour)
/// Previous purchases
/// History and behaviour of similar customers in the same market
This reduced the processing time and fed data into an unsupervised learning engine clustering algorithm that Invenica used as a recommendation seller.
/What we delivered
The outcome was an engine that provided sales and marketing teams with the ability to design campaigns and target a list of recommended customers for each product or service. It also supported customer-facing assets, specifically the upselling of products to website users. This allowed the business to use the technology across multiple channels (including ecommerce, in-store, and SMS), while keeping the core of the data intact – an important element that delivered enhanced learning from both approaches.
With the prototype working, Invenica launched it within an initial market by optimising the machine learning models and integrating them with real-time customer data. This included creating views to extract insight for campaigns, working closely with the commercial and technical teams to integrate the engine with existing channels, and exploring potential new channels like RCS and sponsored social content.