Sign up for our newsletter and get the latest big data news and analysis.

Interview: Kobi Stok, Vice President, Product at WalkMe

I recently caught up with Kobi Stok, Vice President, Product at WalkMe, to give an insiders view for how his company is using AI to predict how/when you’re using a mobile app, and how companies can use this info to push customized campaigns to prevent someone from leaving the app. Kobi is VP, Product at WalkMe, a leading digital adoption platform that simplifies user experiences and drives action using insight, engagement and guidance capabilities. Prior to joining WalkMe he was the Cofounder and CTO of Abbi.io, an award-winning technology that learns and understand user usage behavior. A serial entrepreneur, Kobi has over a decade’s experience in building mobile and web-based software from design to customer deployment.

Daniel D. Gutierrez – Managing Editor, insideBIGDATA

insideBIGDATA: You’re promoting your product as “AI enabled.” Can you provide a high-level overview of how AI is being used?

Kobi Stok: WalkMe’s digital adoption platform sends user behavior events to our back-end. These events include all the actions that the user is performing within the product, such as page changes, mouse clicks, session durations, etc.

We run a set of machine learning algorithms multiple times per day to build prediction models, per a customer use case. These prediction models can then be used to determine specific end user scenarios. For example, using our predictive analytics, the organization can anticipate when a customer is likely to churn, if a customer is likely to use a feature, if a customer is like to need help within the app, and so on.

The prediction models can be used in two ways. Firstly, they can provide insights about a group of users (i.e. a cohort). This output is the most significant information about a user group. For example, we can segment a group of users based on their likelihood to use a feature within the app, so that companies can then target them directly based on the duration and frequency they are using these features.

Secondly, we can engage users in real-time based on predictions (i.e. AI segments). To simplify the work that our customers need to do to create customized campaigns, and help them realize their KPIs faster, we provide our customers with an “AI Autopilot”, which targets the right users at the right moment without the need to drill down and perform complex data analysis. For example, they can create a segment and target “Users that are likely to churn” and publish this. Since our models get updated multiple times a day, the system is always up to date and the AI segment will hold the right set of users.

insideBIGDATA: Are you using Deep Learning, which is most commonly associated with AI, or are you using more traditional machine learning?

Kobi Stok: In production we use traditional machine learning. At WalkMe, we currently do our research using Deep Learning and experimenting with a few new technologies..

insideBIGDATA: How is deep learning being deployed for your solution? Are you using convolutional neural networks (CNNs), recurrent neural networks (RNNs), etc.? What deep learning framework do you use, e.g. TensorFlow?

Kobi Stok: In our research labs we use Deep Learning, we are testing TensorFlow currently.

insideBIGDATA: How is machine learning being deployed for your solution? What kind of statistical learning are you using: regression, classification, ensemble methods like random forest, gradient boosting?

Kobi Stok: WalkMe leverages various machine learning technologies to analyze hundreds of user experience data points collected every session every second. We use statistical learning techniques such as decision trees, random forest and a number of other combinations to ensure our customers have the insights they need to anticipate whether a customer is likely to churn, complete a certain call to action or utilize WalkMe content to some extent and from here, what the best approach would be to prevent this from occurring.

insideBIGDATA: How large are the data sets you use for training your models?

Kobi Stok: At WalkMe we have more than 1,500 customers – including over 25% of the Fortune 500 – that send tens of thousands events per day, which brings hundreds of millions of records per day.

With our most recent product announcement, WalkMe AI, our solution sits on top of any enterprise software or mobile application to understand a variety of user and external device factors, so that organizations can identify user segments and target them with specific campaigns to prevent user churn. We use such data to train our models.

On mobile for example, we collect over 300 parameters of each session including all the device sensors so we can calculate what’s the users physical state, the entire user journey – using our elements engine we’re able to understand the entire path and journey as well as all the external parameters such as device, day, time, country, locale, etc. Using all these parameters, we’re able to determine which user is right for each piece of walkme content and avoid showing irrelevant guidance and help to users who do not need it.

insideBIGDATA: Do you employ the use of GPUs for training or for inference?

Kobi Stok: Currently, we’re only using GPUs in our research lab, not yet in a production environment.

insideBIGDATA: How much of your AI deployment resides in the cloud? What cloud solutions do you use?

Kobi Stok: All of our AI deployments are in the cloud. We use a variety of solutions, including Redshift, EMR, and DynamoDB on top of AWS. Additionally, we use Spark, Python, Scala, Parquet and many others.

 

Sign up for the free insideBIGDATA newsletter.

Leave a Comment

*

Resource Links: