Developed by ex-Xerox Labs exec, VideoKen uses AI to make videos data rich and consumable
VideoKen uses AI-based techniques to automatically index videos, create table-of-contents and phrase cloud, which summarises key topics in a video.
From home DIY (do-it-yourself) skills to learning a new recipe, we have all resorted to YouTube to learn a new concept or skill. While these videos are informative and unique, many are too long and most people end up losing interest after a few minutes of watching it.
To make these videos more rich and consumable, Manish Gupta, Ashish Vikram and Vishnu Raned started VideoKen in 2017. Manish then was the Vice President at Xerox Labs he was joined by his classmate from IIT Delhi, Ashish, who is an ex-VP of Engineering at Flipkart and Yahoo. Ashish brought on-board Vishnu who he knew from his days at IBM Rational.
The platform uses Artificial Intelligence (AI) techniques to automatically index videos, create table-of-contents and phrase cloud, which summarises key topics in a video. VideoKen’s embedded video player provides navigational capabilities to jump within a video to points of interest to the viewer. It also provides insights on which topics within a video received more views, and where the viewers stopped watching the video.
The beginning
VideoKen initially started as a research project in 2015 at Xerox Labs.
“We noticed a shift in preferences in the learners. The younger generation was keen on videos rather than textbooks as a primary learning source, yet most disengaged quickly and didn’t have the patience to watch a lecture end-to-end,” says Manish.
Taking inspiration from the kind of indices the textbooks provide, the team developed algorithms to automatically generate table-of-contents and a phrase cloud (analogous to the glossary) of a video. Both these helped summarise the key topics being covered in a video and enable the viewer to navigate directly to the points of interest.
Manish says their data later showed that it improves the level of engagement of viewers typically by 2-4x. The team realised that both instructors and learners often want to utilise just a specific segment of a long video as the learning material.
“We provided sophisticated capabilities to curate clips, and add further annotations like notes, bookmarks and embed questions into video clips to utilise them more effectively as a learning material,” says Manish.
When Xerox Labs decided to split into two companies and pursue different strategic priorities, the trio convinced the company to sell the IP (Intellectual Property) for this project . After buying the IP and software rights to the research prototype from Xerox in January 2017, the team spent four months to harden that code into a real product that could be used by clients.
The workings
“Since April 2017, our team has continued to improve the capabilities for automated creation of table-of-contents and improving the quality of phrase selection for the phrase cloud,” says Manish.
VideoKen has two kinds of users - content publishers and viewers. A content publisher can submit a video using different sources like API end point, web form for the URL, or they can even upload a new video to the VideoKen platform.
The platform then processes the video content using an AI-based video processing pipeline. The content publisher gets an API call back or an email/push notification when the video processing is complete.
The notification includes an embedded video URL for the submitted video, which can be embedded to the content publisher’s website. The content publisher can publish this video URL on their website or Learning Management System (LMS) or Content Management System (CMS) to enhance the video viewing experience for their viewers.
The viewers can use various video indices (table of content, concept summary/phrase cloud) while navigating through a video.
Manish says the learning platform can be used by enterprise users in two modes - continuous learning and course-based learning. In the continuous learning mode, the user searches for videos based on their topic of interest, navigate to the points of interest and select clips to be saved for further consumption.
Model and revenues
VideoKen is able to add notes and bookmarks to videos clips and not only learn from them personally, but also share them with other colleagues to support peer-to-peer learning.
For a course-based learning, an instructor creates a course using the platform by combining video clips with presentations and documents, adding assignments and quizzes, and then registering learners for the same.
The learners go through the course material and participate in discussion forums and get assessed, and the instructors are able to track their activities.
VideoKen has been granted five US patents and has a few more pending applications. It offers a freemium product to enterprises (large), companies (mid-size), and SMBs (small).
VideoKen’s primary target customers are marketing and HR/Learning leaders within enterprises, corporates and SMBs.
It has a subscription-based revenue model, with a per-learner price for the learning platform and a per-video price for the video indexing solution. VideoKen also offers deep indexing services for free to not-for-profit organisations, individuals and students.
Clients and future plans
The SaaS pricing for SMBs and enterprises starts at $5,000 per year. The team claims to have grown over 100 percent year-on-year. In March last year, VideoKen raised a seed funding of $1 million led by angel investors.
The team claims to work with the likes of Walmart, Bosch, Samsung and TCS. In the past year, the team says it has acquired close to 20 paying customers. It is also a part of the Microsoft Accelerator programme and has been selected by the Karnataka government’s Elevate100.
Some of the other players in the space include Kpoint, OnTrack International, DuPont Sustainable Training Solutions, Orange Tapir and several others.
For businesses, VideoKen focusses on aligning people skills with business needs in scenarios ranging from training, corporate-wide deployment of the platform to support self-learning for all employees.
Videos today account for 75 percent of the internet traffic. Globally, the enterprise video platform market is expected to grow from $16.34 billion in 2017 to $40.84 billion by 2022, at a compounded annual growth rate (CAGR) of 20.1 percent.
The team is now integrating the video indexing platform with marketing automation software such as HubSpot and adding support for more detailed insights and call-to-action capabilities to support the needs of marketing organisations.
VideoKen is adding capabilities for video hosting and search to provide a complete video platform solution. The team is also extending the algorithm to generate automatic table-of-contents to support a broader class of videos and going beyond videos with rich visual information such as slides.
“On the business front, we are targeting customers in the North America region via a combination of inside sales campaigns and business visits. As we gain traction in that market, we are looking to establish an office in North America and then expand to other regions,” says Manish.