How AI, Machine Learning Engines & Voicebots Are Making Businesses More Efficient

May 2, 2019

The State of Digital Transactions in 2018

Not long ago, consumers left their homes and drove their cars to Main Street to go shopping, paying in cash for their clothing, groceries, electronics, and everything else under the sun. The advent of personal computers ushered in the era of online shopping, with the likes of Amazon, Alibaba Group, and others driving ecommerce to new heights. Today, smartphones have become pervasive and have contributed to the “always-on” commerce world as we know it. Consumers can shop and pay anywhere, anytime from a small, handheld device.

We have made great strides over the past 30 years, and we will see this trend in seamless transactions continue as technology becomes more advanced and the payments “experience” all but disappears. The automation of tasks that once took great labor forces to execute has distilled commerce into a seamless experience. However, there are still several challenges that must be overcome as well as opportunities to improve customer experience.

Mobile Eclipse of Desktop in Ecommerce & Changing Consumer Preferences

Mobile has been on the rise for the past several years, finally eclipsing online traffic from desktops in Q4 of 2015, and continuing to rise ever since. Black Friday

2017 saw online sales volume increase 24% year-over year, with a generous 42% of those orders originating from a smartphone.

The mobile-first mentality is in full-force, with smart merchants and retailers (and any company, really) rushing to adapt their websites, sales platforms, and digital content to be mobile-optimized. This makes sense when you consider Gen Z, a generation raised with smartphone in hand from day one, is poised to make up 40% of all US consumers in the next two years.

Brick-and-mortar retail is still on the rise, jumping about 3.8% in Q4 2017, but commerce is trending in the direction of online commerce, including mobile web and mobile apps. Mcommerce retail sales are projected to account for about 28% of all ecommerce retail sales in 2018 - a number that could grow to 45% over the next three years.

The growth of ecommerce - and mcommerce, in particular - has fueled innovation in the fintech sector, leading to advancements in the way consumers transact, invest, and bank. Technological advancements in artificial intelligence (AI) and machine learning have branched from their origins in manufacturing to the fintech space, enabling the automation of business processes that makes for smoother business operations - and consumer payments.

Conversational Commerce in a Changing Payments Landscape

Once used primarily as a means of communicating with friends and family, messaging apps have morphed into a commerce channel. It’s a valuable—and still largely untapped— channel for brands looking to connect with and sell to consumers and build loyalty. This large-scale opportunity seems to have come about overnight, with messaging apps surpassing the top 4 social networks for active users just three years ago. To take advantage of this opportunity, brands must invest in the technology that enables conversational commerce to take place. Conversational commerce that uses AI can streamline interactions with consumers and relieve some of the resource demand burden on businesses. Chatbots (and voice bots) can converse with customers automatically and intelligently. AI facilitates simplicity and customization in interactions with consumers. Both chat- and voice bots can make suggestions, interpret questions, engage in small talk, offer custom content, answer questions or address issues, and sell to people. Additionally, these tools can collect information from these conversations which can be used for additional personalization and to complete transactions in a compliant manner.

The future of payments is enabling consumers to pay when, where, and how they want with seamless flow. Tech companies like Uber have been pushing embedded payments into the mainstream, creating an expectation from consumers that they can simply select what they want while payment transactions happen behind the scenes and without additional effort on their part. Conversational

The Fraud Innovation Problem in the Card-Not-Present (CNP) Space

While seamless, embedded payments are certainly the goal, there are several challenges that lay ahead of this achievement. For one, embedded payments demand that security be baked in as well. Payments should only be easy for the consumer—not for fraudsters or non-authorized entities.

As mobile payments continue their upward trajectory and we start getting into voiceactivated payments territory, businesses will need to adapt equally sophisticated fraud prevention tools to ensure the safety of each and every one of the transactions they process. What’s more, organizations must implement this technology without adding friction to the process.

EMV Strengthens Fraud Prevention at POS, Accelerates Card-Not-Present (CNP) Fraud Problem

As merchants, fintechs, financial institutions, and technologists continue to innovate and improve online payments, fraudsters are mirroring these efforts. According to Accenture, data security remains a top priority, as the financial services industry projecting $31.3 billion in global card losses in 2018. Fraud is a business—and business is good. Today’s bad actors have the funding, technology, and wherewithal to commit highly-advanced fraud against unsuspecting businesses and consumers.

Fraud is accelerating at a higher rate than ever, targeting businesses that are not up to par with compliance and security regulations and best practices. The ability to dramatically lower instances of fraud at the point-of-sale (POS) has also aided in the online fraudster gold rush. The EMV (which stands for Europay, MasterCard, Visa) rollout in Europe and North America set the standard for security and global acceptance. Backed by chip-and-pin technology, it all-but-eliminates fraud at the POS by requiring the purchaser to enter a personal identification number (PIN) as opposed to a signature on a receipt. Paired with the chip technology, which cannot be cloned the way a magnetic stripe can, it drastically reduces fraud for face-to-face transactions. Recent reports note that counterfeit fraud instances have declined by about 70%.

While excellent news for brick-and-mortar merchants, CNP merchants fare worse. The fraud attack index rose drastically immediately following the 2015 liability shift for merchants, increasing from $2.7 in Q4 2015 to $4.98 in Q4 2016. It’s a conundrum for merchants, who strive for omnichannel excellence without sacrificing security. While many fraud tools can aid in combating online fraud, they are often clunky and require a lot of human intervention. Merchants with already-stretched-thin resources may find it difficult to maintain the pace.

This struggle presents an opportunity for machine learning, which can drastically cut down on the number of resources merchants need to effectively fight fraud. The key is for merchants to educate themselves on the current state of fraud, and to invest in emerging technologies that can not only streamline payments, but keep them safe.

Machine Learning in Payments: The Advent of Fraud Prevention Automation

A Brief Overview of Machine Learning in Payments

Machine learning has been the foundation of advanced analytics for years, but it is more recently that its applications have branched out to more and more industries. As it has become more broadly accessible at a lower cost, industries that run the gamut have adopted its applications to automate processes across a number of functions. McKinsey Global Institute projects that 47% of the US workforce will be automated by 2030.

Fintechs in the payments realm have made great strides with machine learning applications in the sector, particularly as they relate to payment card transaction monitoring and authorizations. The proliferation of Big Data has made the need for superhuman numbercrunching capabilities essential.

Between the vast amount of data and the increasing sophistication and magnitude of fraud attacks, machine learning has the potential to put legitimate businesses back in control. Today’s machine learning capabilities have opened the door to the following:

Modelling evolution & reduced cycle time: whereas batch learning causes models to become out of date, machine learning enables modelling to evolve, turning learning into a continuous cycle that improves over time. As transactions are scored over time, machine learning can adapt scoring models, rather than needing to start from scratch when anomalies are identified.
Real-time decisioning with Big Data: machine learning facilitates distributed data processing without trading data for latency, or vice versa. This enables real-time decisions as in-memory advances and risk scoring can be done at ultra-low latency.
Less manpower, increased effectiveness: flagging, scoring, and analysis tasks once completed by humans can now be automated, detected, and completed more accurately and more quickly than ever before. The ability to detect subtle patterns and anomalies without bias surpasses the ability of humans, allowing for error-free processing

Machine Learning for Payments Fraud Prevention

The characteristics outlined above make machine learning a critical technology for payments fraud prevention. Specifically, the large number of unique instances of fraud is vast and beyond human capabilities. This long-tail distribution along with quickly changing patterns make machine learning ideal for combating the problem. Another important consideration is the impact of fraud on customer experience. As fraudsters improve their ability to mimic legitimate customer behavior, good customers will pay via intrusive security measures, overreaching fraud controls, and poor user experience. Machine learning, however, has the ability to intelligently learn subtle fraud signals and differentiate between true fraud and a legitimate customer who may have unwittingly tripped a pre-programmed fraud flag.

Fraud is expensive along the entire lifecycle of a transaction. Overly-sensitive fraud tools a) cost money to implement, and b) drive away good customers that abandon purchases that are too complex. Manual reviews and false positives alone make up close to half (40%) of merchants’ total cost of fraud prevention. As emerging payments channels gain popularity and social and conversational commerce gain momentum, these costs can be expected to rise. LexisNexis posts that fraud prevention via remote channels is up to 7 times more difficult to prevent than in-person fraud.

Reduced manual review time: Fast iterating machine models can reduce manual review queues, decreasing time to completion
Improve precision in manual reviews: Machine learning can detect subtle patterns and evolve fraud scoring more quickly and accurately, making it the perfect augmentation to human decision-making
Be channel-agnostic: As noted earlier, commerce is happening on an increased number of channels—even those that weren’t created with commerce in mind. Machine learning can be applied to an omnichannel fraud prevention strategy and integrated with enterprise systems for streamlined process automation.
Reduce false positives with behavior analysis: Behavioral modeling and pattern recognition can reduce friction by only flagging truly fraudulent or suspicious transactions for review. Those that meet the threshold for a “safe transaction” may bypass more advanced security mechanisms, enabling frictionless transactions.

The key is finding the right machine learning solution that can easily integrate with your existing systems to automate burdensome and costly processes.

Solution: The Opus Machine Learning Engine for Suspicious Activities Investigation

Background: Opus Machine Learning Engine

Opus developed an algorithmic and statistical scoring engine, leveraging our payments domain. This engine creates a machine learning vector-based model for feature identification, modelling, and classification. Opus’ machine learning engine is a user-centric multidimensional model, with the main dimensions including:

Algorithm Selection: Our proprietary methodology uses business case-driven algorithm selection, which leverages concepts from:
- Logistic Regression
- Support Vector Machine
- Naïve Bayes
- Neural Networks
Building Machine Learning Model : User Inputs: Users have the choice of which data features should be included:
- Configurable weightages
- Input values for other algorithmic parameters Feature Management: uses feature identification via Decision Trees/Random Forests and PCA. It also allows for the manual handling of features.
Feedback & Self-Learning: User feedback is incorporated to the input feature dataset and to the model.

The platform allows inputs from a business’ core systems, databases, and websites, including real-time transaction data, historical data, batch feeds, data marts & warehouses, portals and reports. These structured and unstructured inputs (aka learning influencers) undergo sanitation, aggregation, and manipulation as it is digested by the platform. The AI/machine learning platform then uses algorithms and self-learning to automate and streamline core business processes, including:

Scoring outcomes of events
Classification
Failure analysis
Intelligence support

Machine Learning Engine Application for Payments

Our machine learning platform can be used to automate decisions process for multiple payments business processes.

The key features of the engine include:

In-premise and On-cloud ( private and public) deployment
Multiple machine learning algorithms customized for payments use cases
#VisualFirst- an intuitive design to abstract the technology complexity of the model
Aligned to user-assisted learning and time series analysis

Arming Compliance Teams with Automated Identiﬁcation of Suspicious Transactions

One of the biggest use cases for our machine learning engine is to aid compliance teams in automating fraud prevention and streamlining the review process. Currently, many organizations are combing through millions of transactions to identify suspicious activity. In many cases, anomalies (a cross-border purchase, for example) get flagged as fraud by legacy systems, prompting the compliance team to review. With transactions in the millions, this becomes a costly, largely inaccurate exercise with a high number of false positives.

Opus’ machine learning engine intervenes at the point where transactions are flagged by current systems. The engine identifies only highly probable suspicious transactions, which are then passed along to the compliance team for review, based on influencer values. By using inherent historical data and feature identification, businesses can drastically reduce time and resources spent reviewing a high number of transactions, improve false positives, and handle first day attacks.

Voice Bot Technology to Adapt to Changing Consumer Preferences

Our Ailene Voice- and Chat Bots are backed by Amora, our machine learning and AI engine, to easily integrate with your core enterprise systems, data and customer engagement channels to transform your customer experience.

Given the shifting tides of commerce outlined in this paper, voicebots are quickly becoming the next frontier for merchants that want to remain profitable and in the good favor of consumers. Our voice bot technology provides AI-backed intelligent understanding, enabling Ailene to engage with customers in real-time, natural conversations. She is able to read customer moods and reply with appropriate responses, mimicking a human experience.

Using advanced machine learning models, Ailene can analyze conversation data to make intelligent business decisions in real-time, automating essential services like inquiries and transactions.She also easily integrates with popular messaging and voice frameworks like Google Assistant, Alexa, Facebook, and Slack.

Ailene leverages an NLP hub, powered by Amazon Alexa (voice) and Google Dialogflow (chat), to can engage in small talk, capture intent, and understand and translate financial vocabulary. This equates to better, more automated customer service workflow. The technology allows for easy exchange of information between Ailene and your human customer service team, allowing her to escalate situations to a human, should the need arise.

Most importantly, Ailene integrates with your databases, web-services, or any other legacy framework layers to produce desirable results for your customers, whether it’s a completed transaction (using two-factor authentication) or simply an answer to a question.

Conclusion

The current state of ecommerce is focused on mobile and omnichannel experiences. Consumer preferences are evolving quickly, and merchants must keep up to be successful. Mobile commerce is only the beginning, as we begin to see payments trickle into messaging apps, voice assistants, and other connected devices. This new era of payments requires an embedded-first mindset and technology to facilitate execution.

As this evolution takes place, companies must also prioritize security and compliance. Balancing optimal customer experience across channels with effective yet seamless security will be the focus. Advancements in machine learning and its application to payments will provide ways to automate, streamline, and secure payments without demanding additional human resources and the associated costs. Not only will AI and machine learning facilitate embedded payments by way of chat and voice bots, but they will remove friction from the process by automating fraud prevention tasks behind the scenes. It’s a tall order to balance customer preferences with security, but it will be mandatory for companies that wish to compete in 2018 and beyond.