Vendors use artificial intelligence (AI) technology to build automation at the core of their IVA solution. An IVA is not one monolithic piece of code but multiple automation pieces that combine to fulfill the tasks required to process an end-user query.

At the start of the customer journey, AI is used to convert speech to text and sometimes also to translate a foreign language to English before further processing. Once the end-user input is available as text, the next step is to pull out the intents—the end-user’s purpose—from the query. This is accomplished with the help of AI-based natural language understanding (NLU) and natural language processing (NLP). Furthermore, AI can be used to predict the outcome of the conversation, to answer fact-related questions, and to perform sentiment analysis. Finally, in the return part of the process, AI is used to convert text to speech (if a voice channel is being used).

These multiple AI elements are built in a variety of ways by different vendors and may be augmented by linguistic and semantic models. A key trend is the emphasis on ease of use in deploying a solution, with low code or no code (LCNC), a popular approach for enabling a business’s domain experts to build an IVA solution. Increasingly sophisticated IVA building blocks are becoming available from the major public-cloud providers, including Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP). Some enterprises use these building blocks to create their own IVA solutions and typically need developer expertise to do so, but the IVA players may also exploit these tools in building out their solutions. So while the cloud providers have increased competition in the market, they also provide opportunities to both enterprises and vendors.

The market for chatbots and IVAs continues to grow and improve, but the top end of the market—the sector satisfying the needs of large enterprises—must present solutions offering human-like performance and high scalability, which require more sophisticated capabilities, and this report assesses such capabilities. The companion GigaOm report, “Key Criteria for Evaluating Intelligent Virtual Assistants,” delves into the architecture of an IVA solution and the features that are critical for evaluating its capabilities.

The post GigaOm Radar for Intelligent Virtual Assistants appeared first on Gigaom.

Unintelligent legacy virtual assistants could only manage a single intent of the end user and were not able to conduct a natural conversation. They were disliked by the public for their limited capability and could drive away customers. The latest generation of IVAs that use artificial intelligence (AI) can separate different intents, handle them in turn, and have a natural conversational ability; overall, these solutions score high in customer satisfaction and can operate efficiently on par with human agents. For end users, an IVA offers a rapid response instead of a long queue waiting to speak with a human agent. As such, the IVA is a win for both end users and businesses.

End users are characterized as having a journey, from the initial query to its resolution. An example of a complex journey would be calling an agent to book a flight, during which the IVA must account for several variables:

• There could be multiple stages with stopovers.
• Hotel and car bookings may be included.
• There may be outward and return bookings and different departure and arrival points.
• Weather may be a factor, such as the likelihood of snow shutting down an airport.
• Preferences over ticket class and position of seats must be taken into account.

The IVA agent must navigate the multiple requirements while retaining context, managing changes if the end user needs to modify any details, and taking into account multiple knock-on effects.

Less complex journeys can be handled by chatbots, which involve question-and-answer interactions with minimal continuity or memory between the chatbot responses. In this report, we focus on solutions designed to handle complex end-user journeys.

This GigaOm Key Criteria report details the criteria and evaluation factors for selecting an effective IVA platform. The companion GigaOm Radar report identifies vendors and products that excel in those criteria and metrics. Together, these reports provide an overview of the category and its underlying technology, identify leading IVA offerings, and help decision-makers evaluate these platforms to make a more informed investment decision.

The post Key Criteria for Evaluating Intelligent Virtual Assistants appeared first on Gigaom.

With the use of AI/ML now commonplace, differentiation in this space instead becomes a question of how the ML is used. For example:

• Is the ML model supervised or unsupervised?
• Does the vendor create a custom ML model specific to the customer or supply a ready-built model, which is then fine tuned and retrained for each customer over time?

Customer data privacy is strictly respected, and with AI/ML vendors are able to see trends appearing in their customer base and can alert customers not yet affected by a new fraud. Vendors with large customer bases and those who subscribe to fraud data alert services have an advantage here.

Some of the newer types of fraud relate to the digital economy, such as exploiting voucher schemes and return policies (which are more generous than ever before as retailers aim to compete with Amazon). Unlike a transaction fraud—which is either allowed or stopped—the newer types of fraud require a more nuanced approach because a good customer might have picked up a bad behavior and it is the behavior that needs to be stopped while retaining the customer. For example, it’s a gray area merging into abuse when a customer buys 10 items online (which are free to return) to only keep one.

One common approach that many vendors take is to model normal behavior and then train the ML model to detect any deviation from the norm—being open ended in this manner allows new forms of fraud to be detected earlier. Combined with various risk assessments, the abnormal incident may then be flagged for escalation. For example, more stringent ID checking may be performed before bringing the incident to the attention of human fraud investigators.

AI/ML has brought significant benefits to the automation of fraud detection, greatly improving banks’ and merchants’ abilities to deal with the rise in fraud in the wake of economic digitization. For more background information on this topic, we recommend readers refer to the GigaOm report “Key Criteria on Intelligent Fraud Detection Solutions.” In this Radar report, we perform an in-depth side-by-side evaluation of the leading intelligent fraud detection (IFD) products in the market.

The post GigaOm Radar for Intelligent Fraud Detection in Financial Transactions appeared first on Gigaom.

Back in the days when banking transactions required three to five working days to process, there was a period of opportunity to detect fraud; however, with digital finance enabling real-time transactions, modern IFD solutions need to operate within millisecond time windows.

Anomaly detection involves understanding what is normal and triggering alerts when abnormality occurs. The earliest attempts at anomaly detection involved establishing simple thresholds for normal behavior by the person or process under observation and monitoring for threshold crossings. With AI, a more sophisticated model can be created that goes further to exploit multiple data channels and complex data relationships, leading to more accurate results with fewer false positives and false negatives.

The varied methods of fraud are a testament to the diversity of finance industry methods for transacting payment. The growth of digital finance and internet-based payment systems have propelled an increase in fraudulent activity across the various payment systems, with the fastest rise happening in identity-related fraud.

Frauds can be categorized in two ways:

• Perpetrator is internal to the victim organization (i.e., fraud committed by a person against the organization for which they work).
• Perpetrator is external to the victim organization.

Of course, it’s possible for external and internal agents to be acting in unison.

According to the 2020 Global Study on Occupational Fraud and Abuse (that is, internal fraud) reported by the Association of Certified Fraud Examiners (ACFE), proactive data monitoring and analysis (encompassing both traditional and AI-based solutions) was used in only 38% of victim organizations across 2,504 cases examined by the ACFE. This statistic indicates considerable growth potential for AI fraud detection technology.

We use the term intelligent fraud detection (IFD) to describe systems that primarily use AI to monitor and reveal fraud.

Figure 1 shows a high-level view of an IFD solution.

Figure 1: Architecture of an Intelligent Fraud Detection Solution: High-Level View.

These are the typical steps followed by an IFD solution:

• Step 1: A customer or employee initiates a transaction with the organization, such as a payments system. The IFD begins its verification process, which will use its fraud-detection capabilities.
• Steps 2 and 3: Real-time data is gathered, which typically involves retrieving records from back-end systems.
• Step 4: A fraud risk score is produced using an ML model, which fraud analysts and financial investigators can monitor, and intervene if necessary.
• Step 5: A decision is made as to whether to allow the transaction to proceed.
• Steps 6 and 7: The ML model that is used in the fraud risk assessment (Step 4) is produced in an ML studio by experts and deployed into the production process. ML model development may use external ML resources and application accelerators, both hardware and software.
• Step 8: Periodically, the ML model is retrained with the most up-to-date data to improve accuracy and eliminate “data drift” when the environmental data has changed.

Note: In this report we will refer to financial institutions (FIs), but we embrace any organization with a payment system.

The post Key Criteria for Evaluating Intelligent Fraud Detection in Financial Transactions appeared first on Gigaom.

In this GigaOm Radar report, we cover the vendors that are the most active in edge AI computing. The GigaOm Radar chart provides an immediate positioning of the solutions, and the report provides an in-depth analysis of each of them based on our key criteria and evaluation metrics.

We recommend first reading the accompanying GigaOm report “Key Criteria for Evaluating Edge AI Processors,” which provides a foundational resource for making decisions about selecting an edge AI processor for use by product manufacturers building edge technology solutions. GigaOm has also published Radar editions for Ultra-Low Power Edge AI Processors and Automotive Edge AI Processors.

The post GigaOm Radar for Edge AI Processors: Low Power Edition appeared first on Gigaom.

In three GigaOm Radar reports, we cover the vendors that are the most active in edge AI computing. The GigaOm Radar chart provides an immediate qualitative positioning of the solutions, and the report presents an in-depth analysis of each of them based on our key criteria and evaluation metrics. This edition covers edge AI processors targeting the automobile market for advanced driver-assistance systems (ADAS) and autonomous driving (AD), with an emphasis on low-power devices. Vendors who provide low-power chips for other automobile functions, such as in-cabin infotainment, are not included here.

We recommend first reading the accompanying GigaOm report “Key Criteria for Evaluating Edge AI Processors,” a foundational resource that details the criteria and metrics we used to evaluate vendor offerings. GigaOm has also published Radar editions for Ultra-low Power Edge AI Processors and Low Power Edge AI Processors. It can help product manufacturers building edge technology solutions select an edge AI processor.

The post GigaOm Radar for Edge AI Processors: Automotive Edition appeared first on Gigaom.

We recommend first reading the GigaOm report “Key Criteria for Evaluating Edge AI Processors,” which provides a foundational resource for making decisions about selecting an edge AI processor to be used by product manufacturers who are building edge technology solutions. GigaOm has also published Radar editions for Low Power Edge AI Processors and Automotive Edge AI Processors.

In this GigaOm Radar report, we cover the vendors that are the most active in edge AI computing. The GigaOm Radar chart provides an immediate positioning of the solutions and the report provides an in-depth analysis of the solutions based on our key criteria and evaluation metrics.

The post GigaOm Radar on Edge AI Processors: Ultra-Low Power Edition appeared first on Gigaom.

This GigaOm Key Criteria report is designed to be a foundational resource for product manufacturers building edge technology solutions, to help them make decisions about selecting an edge AI processor. In this report, we provide an overview of edge AI processor capabilities, including those that are commonplace and those that stand out. We define specific criteria and metrics for selecting an edge AI processor, enabling engineering organizations to make better decisions from among available options.

This evaluation extends to table stakes that are common to virtually all products in this sector, key criteria that define differentiating features to focus on, and emerging technologies that point to ongoing innovation in this space. Finally, we describe a set of evaluation metrics—high-level characteristics that help determine the impact processor choice can have on implementation and are useful in assessing specific products. Whether you are looking to extend existing capabilities in AI edge computing or have yet to adopt edge AI processors, this report lays the groundwork for informing the selection and implementation of an edge AI processor for your needs.

Findings reached in this report include:

• The edge computing market is set to grow on the back of a number of converging technologies: AI, 5G, digital transformation, cloud native technology, and expansion of IOT.
• A useful criterion for determining an edge (and local) application from one that is merely distributed between edge and cloud is that latency is less than 20ms, meaning the application does not rely on a network to the cloud to perform its main function; instead, it has local resources that allow it to fulfil its prime function and operate in near real time. Many edge applications are connected to the cloud or gateways mid-way for additional functionality and services.
• The requirements for an AI processor at the edge are markedly different from accelerating AI in the cloud or data center: In the latter case, AI applications are typically trained and run large data sets for large numbers of users or they process complex numerically heavy applications in high-performance computing (HPC). At the edge, in contrast, the AI applications are typically running in inference mode, and are constrained by limitations in power, chip size, latency, bandwidth, and cost.
• AI training at the edge is a differentiator across the edge AI processors, with local continuous learning. This is in contrast to wired or over-the-air model updating, for which the edge processor is always operating in inference mode.
• Selecting an edge AI processor simply on its tera operations per second (TOPS) rating or even TOPS per watt misses important considerations. We provide a full and rounded set of assessment criteria that considers preprocessing optimization, latencies, software maturity, and more.
• We take the view that the type of processor architecture is less important than its characteristics and suitability for edge AI applications.

The post Key Criteria for Evaluating Edge AI Processors appeared first on Gigaom.

The other domain is the focus of this report and concerns software used for products and services that are safety-critical or mission critical, and highly regulated. In this case, the developers are experts in the domain and are working in line-of-business departments. Driven by digital transformation, the size of this market segment has grown massively, creating an opportunity for ALM vendors.

The goals of ALM remain relatively constant: to offer an end-to-end software development lifecycle (SDLC) management platform that supports integration with IDEs, testing, and other development tools; and to enable reporting to management about project and delivery progress.

Digital transformation is taking place at a time of accelerated pace in the rate of technological change. The smartphone today has the processing power of supercomputers from not very long ago. It is possible to embed fully functioning computers in products that run applications with many millions of lines of source code. The use of code in products has gone beyond basic microcontroller firmware and progressed to advanced driver assistance systems (ADAS) in automobiles, high-end medical devices such as for diagnostic imaging, and sophisticated applications in pharmaceutical drug discovery and financial investment trading.

With digital transformation, the need to manage the development of high-quality software has grown and along with it the compliance burden, as auditors demand, for example, full traceability from high-level requirements to implementation in code and deployment. To serve this need, a new generation of ALM tool suites has emerged, able to handle thousands of requirements, configurations, and parameters across multiple products and variations, at a scale that a typical agile project management tool is not designed for. Through an integrated set of tools, these ALM solutions are able to produce compliance reports in an instant, demonstrating traceability and showing the testing history of each component.

Given the extent of divergence between tooling for traditional ALM or agile project management and the new ALM targeting highly regulated industries, we are relabeling the latter as regulated software lifecycle management (RSLM).

The post Key Criteria for Evaluating Regulated Software Lifecycle Management (RSLM) appeared first on Gigaom.

Application lifecycle management (ALM) suites (often under the label of agile project management) have been used to manage the development of highly complex software and/or large software projects with multiple teams. Over the years, they have evolved and split to meet the needs of both enterprise IT and of highly regulated markets for software applications—from advanced engineering manufacturing to the pharmaceutical industry, financial services, and beyond. Given how the tooling for enterprise IT ALM and the new ALM targeting highly regulated industries has diverged, they are now two distinct categories and therefore, we are re-labelling the latter as regulated software lifecycle management (RSLM).

If you’re a product manufacturer or service provider considering the purchase of an RSLM solution, we recommend first reading the GigaOm Key Criteria report, which serves as a foundational resource for making decisions about selecting such a solution. In this GigaOm Radar report, we cover the vendors that are the most active in this sector. The GigaOm Radar chart shows a visual positioning of the vendors, and the report provides in-depth analysis of the solutions based on our key criteria and evaluation metrics.

The post GigaOm Radar for Regulated Software Lifecycle Management (RSLM) appeared first on Gigaom.