TechBytes with Ivaylo Bahtchevanov, Head of Data Science, ForgeRock

 Tell us about the role and team/technology you handle at ForgeRock.

I currently head up our data science initiative, which encompasses everything from the data collection process to organizing, structuring, and storing it for later use to building and deploying Machine Learning (ML) features that augment our existing product line.

Our core use cases are focused on developing and deploying predictive capabilities to improve the authentication and authorization journey.

Why are B2B marketing teams steadily moving toward Applied Data Science for Sales and Marketing initiatives?

An important part of Sales and Marketing is understanding the key customers of your product, and data is the mechanism that really develops and takes that understanding to the next level.

For the B2B segment specifically, this approach manifests in two ways. The first is applied data science to enhance the performance of marketing campaigns. A combination of optimization functions, decision-making, and planning functions allows businesses to better understand and personalize content and offers for their target audience.

The second is something called Robotic Process Automation. There are plenty of repetitive, manual, and error-prone processes that exist, and data science can be leveraged to fully automate those tasks so that the marketing teams can focus their efforts on where they add the most value — creativity, strategy, calculated risk-taking, and partnership development.

Tell us more about the changing nature of digital identity and how it affects data management initiatives.

Generally, people agree there needs to be a “strong” digital identity that can be comprehensive and difficult to impersonate. Currently, the definition of a digital identity is evolving to be more mobile and dynamic. However, this introduces a wider range of vulnerabilities. We’ll start to see increasing demands for robust security measures and consent, and the way data is managed and processed will ultimately be subject to tighter control.

Typically, you’ll see legal teams closely embedded with the product managers and tech leads, and they’ll play a heavy hand in designing the system from the start. Added functionality like “given a user, provide me with all logs associated with recorded actions and events of that user” or “given a set of users, delete all data associated with these accounts” will influence how large-scale computing systems are implemented. Involving these considerations upfront usually steers the design since attempting to add data management functionality afterwards usually means rebuilding most of the system.

Data residency will also be a huge component for data centers. Vendors such as Google and Amazon have guarantees on their residency so they can take on international clients such as HSBC.

How do you see Big Data and Customer Intelligence coming together to solve identity-related issues?

Previously, digital identities had become these static systems of record that represented a generic ledger for attributes and previous actions. For identity and security vendors, historically there has always been a direct trade-off between user experience and security. If you want to make an application more secure, you bar it with multiple authentication steps and make it as painful as possible for a user to authenticate and access anything.

If you want to make an application with a great user experience, you remove all user friction and allow the application to remember devices, integrate with services that directly authenticate, etc. and allow the compromise of any of these devices or adjacent applications to also compromise your own application.

Applied data science allows these identities to become dynamic entities that can easily be monitored. This removes the need for excessive authentication while keeping a close eye on identity behavior on the platform. The result is a better experience that’s more secure.

Tell us about the various steps of the AI-powered identity process journey.

The journey begins the second a user signs up. The creation of an account kicks off a training process that learns the behavioral trends that tie a specific action to a user.

The models learn things like:

• What devices does this user typically use to authenticate?
• What devices are most commonly associated with specific locations or actions on the platform? (for example, a user might only perform wire transfers or transactions on their laptops and not on their cell phone)?
• What are a user’s keystroke patterns and typical mouse movements?
• Given the functional role and their industry, what sort of policies or access permissions do users of similar demographics hold?

From there, every action a user performs includes a ping to the ML engine that asks “how likely is this action performed by the intended user?” If the engine finds an inconsistency, the user would need to re-authenticate or verify their identity in some way.

The ML engine retrains on new actions and feedback, becoming smarter and more fine-tuned to the individual identities.

What are the different scenarios AI could be applied to in Marketing, Sales and Customer Service?

The concept of being “customer-centric” has been defining categorical winners and losers in the digital economy. Having one data-complete source of truth on customers is a necessity, especially when Sales and Marketing teams are interacting with users across a variety of different channels. Structuring this first-party data is what empowers these organizations to deliver a great customer experience.

AI helps in determining how and when to display content to users — displaying the right ad to the right user at the right time that will maximize the probability of a click and a conversion is an exact science.

Customer service represents a huge green-field window of opportunity — across industries, support teams will repeatedly see the same issues come up over and over again.

Companies such as Solvvy are attempting to solve this problem by building a large knowledge graph that captures all of these problems and ties together their respective solutions into an intelligent self-service chat tool. The knowledge graph will develop a deep understanding of customer intent and can provide both an intuitive, guided experience for customers to resolve their issues and  effortless automation of complex workflows (integrate backend systems to external APIs to execute these workflows, etc.)

This does not mean AI will replace support teams — it will empower them. By delivering intelligent self-service and automating repetitive tasks and tickets, agents will have a lightened workload so they can focus only on the issues that require significant attention and detail.

How do you build analytics around Cloud and Enterprise Mobility platforms?

It starts with understanding the customers and then designing a platform that will best analyze leading indicators of behavior and convert that into valuable insights. For ForgeRock, we focus primarily on authentication and authorization, so we design our systems at a high level around how we can capture the entire identity journey.

Building an enterprise cloud analytics platform means thinking about performance, scalability, and rapid elasticity.

A unified system should be used for all your enterprise customers, so you want a platform to effectively “plug and play” different customer instances without rebuilding or redesigning the system to meet their needs.

You know your customers and should create a one-size-fits-most system to capture the most important insights that will bring your business the most value.

Tell us about your customer experience products. How can MarTech customers benefit from your AI/ML products at ForgeRock?

As of now, we focus primarily on creating a frictionless and seamless authentication and authorization experience and ensure the identity behind the application is, in fact, the person they claim to be. We are developing a unified dashboard for enterprises to monitor and identify their vulnerabilities and understand where they come from and what causes them.

On the side of authorization, we are building models to develop strong recommendations that will ultimately, automate the extensive manual policy-setting for enterprises.

Finally, we are leveraging traffic patterns to predict how enterprise customers can deploy and allocate resources to meet demand. These models will connect to their infrastructure and IT tools to automate their deployment or reduction as needed.

As of yet, we do not provide analytics to our customers from the standpoint of customer segmentation, optimal monetization opportunities, recommendations, etc. While we are well-positioned to do so, we have not yet determined how these efforts play into our long-term strategy.

What is the current state of Machine Learning and AI for digital transformation?

It’s becoming a pivotal part of organizations in making data-driven decisions and in delivering personalized customer experiences.

What are the major pain points for Product Management and Innovation teams in building/scaling analytics for customer experience?

A lot of work goes into developing and designing a system that actually solves the right problems. That starts with orchestrating the data processing and collection, including understanding how much data there is and how frequently users interact with the core product to build the streaming mechanisms. Then they need to think about how they embed that process with legal and privacy regulations of the end users.

You want these analytics systems to be flexible — if a customer adds a new feature, you want to be able to track it without rebuilding core components. Similarly, you want to track metrics across customers, and you want that to be part of a plug-and-play aspect.

How do you work with AI/Machine Learning at ForgeRock?

We use a lot of unsupervised learning that tackles unstructured and seemingly random data on usage and structures it to make sense of patterns invisible to the human eye.

We leverage the most cutting-edge deep learning techniques to learn from our data to integrate intelligent actions around how we authenticate and authorize. These models can directly map a set of actions or time series behavior back to a specific identity and vice versa.