VitalSource Leans on GenAI to Reimagine Content Discoverability for Higher Ed Faculty

Tribe

About VitalSource

VitalSource is the leading education technology solutions provider committed to helping partners create, deliver, and distribute affordable, accessible, and impactful learning experiences worldwide. As a recognized innovator in the digital course materials market, VitalSource is best known for partnering with more than 1,000 publishers and resellers to deliver extraordinary learning experiences to millions of active users globally– and today they’re also powering new, cutting-edge technologies designed to optimize teaching and learning for maximum results.

VitalSource is the central point between content providers (publishers, courseware platforms, and other online learning tools) and over 300,000 US higher education faculty members.

VitalSource’s Challenge

Nick Brown, VP of Product, has witnessed many industry innovations during his 15 years at VitalSource, but none as fast and gamechanging as Artificial Intelligence (AI).

“There’s an AI announcement weekly in education that moves the goal posts on what AI can do for education,” said Brown. 

VitalSource isn’t new to AI innovation. They’ve had AI-enabled offerings (e.g. practice questions as a supplemental support to course materials) in production since 2018. Brown knows, though, that their current offerings are just the tip of the iceberg in how AI can further innovate the industry. GenAI has been used frequently in content creation by other EdTech solutions providers, but that's not an area VitalSource wishes to play in. 

Rather, Brown and his team at VitalSource have a three-pronged approach to determining when and how to implement an AI strategy. The strategy must be centered on:

  1. Improving the student learning experience
  2. Increasing the productivity and proficiency of internal teams
  3. Innovating the course materials eco-system

Brown and his team felt they could touch on all three goals by prioritizing their B2B channel, specifically through improving the content discoverability experience for higher ed faculty.  

From ongoing research, the VitalSource team knew that the current faculty search experience was fragmented, varied, and inefficient. When it came to determining course materials, instructors used a variety of research methods like asking their peers for recommendations, performing google searches, and even calling publishers directly. Faculty, themselves, confirmed what VitalSource had been suspecting: there must be a better way!

“Our goal was to completely reimagine the faculty experience. We believed AI would be a necessary part of that future state, dramatically improving the experience of faculty members by making our content more accessible and the interaction more engaging and conversational,” explained Brown.

Brown surmised that an improved faculty search experience would enable the best selection of course materials which in turn would deliver improved learner experiences and ultimately position VitalSource as the premier EdTech solutions provider. A title like that undoubtedly comes with increased market share and revenue.

Why Tribe AI?

VitalSource has a strong technology team on staff with solid AI experience, however Brown felt confident that they needed a third-party AI expert to help validate and fastrack this initiative. 

“The pace of progress in AI is rapidly increasing. We needed to prioritize this work and lean on experts who live and breathe in this world daily,” said Brown.

VitalSource was introduced to Tribe AI by Francisco Partners, a reputable tech investment firm.

Proposed Solution

Brown’s team began a proof of concept (POC) engagement with Tribe AI to explore a conversational faculty interface, using an Amazon Bedrock-powered GenAI strategy deployment, that was aimed at offering a more reliable and efficient experience for faculty to discover course materials. 

“We did not have a preconceived notion of what this solution would look like when we first met with Tribe AI. Their team was instrumental in bringing clarity to what was possible,” said Brown.

Tech Stack

The project utilized a frontend chat experience served using a public Amazon S3 bucket. The backend server was hosted on Amazon Web Services (AWS) EC2 (Elastic Compute Cloud). Github Actions were used to set up Continuous Integration/Continuous Deployment (CI/CD) pipelines that built Docker images and updated the application on the EC2 instances. AWS Elastic Container Registry (ECR) was used to hold all historical Docker images, allowing for rollback functionality if necessary. With this setup, Tribe AI was always able to push the latest code changes automatically to the live app.

The backend was written in Python using Django. PostgresSQL was delivered  using Amazon Relational Database Service (RDS) to store conversation history for users, allowing them to revisit previous course material searches.

AWS Bedrock was used as the backbone for the GenAI work. The VitalSource catalog was turned into embeddings using Cohere's models on AWS. The embeddings were then stored in Pinecone. When users asked for course material, the closest documents would be retrieved, and an LLM-reasoning step would construct a consultative message for the end user. Claude was the model of choice on AWS Bedrock for this reasoning step, because of its accuracy and speed.

Developing the Roadmap

The engagement began with the VitalSource and Tribe AI teams coming together to perform a knowledge share centered on prioritizing the work, defining success metrics, and uncovering data challenges/limitations.

Three success measures were defined to guide the POC work:

  1. The AI interface must feel truly conversational
  2. The search results must be viewed as trustworthy by faculty
  3. The cost per search should be closely monitored to justify the business model

Faculty insights drive conversational interface

The teams knew that delivering a truly conversational interface would be a heavy lift. It would require a strong understanding of faculty, the seasonality of their work, what drives their search, what challenges they’re facing, and more.
“We knew that faculty cared deeply about finding the right course materials and technology add-ons to ensure the best learning outcomes for their students. But, this discovery experience needed to be made easier and more intuitive for them,” said Brown.

In education, new supplemental course materials and digital tools (like chat forums, quiz makers, video components, etc.) are launched regularly, making it of utmost importance for the interface to know when and how to recommend the right  offerings to faculty as they navigate through their research process. Relevance will be key here so the suggestions don’t appear as unwanted upselling but rather as reliable and helpful suggestions to optimize teachers’ course materials.

Another important piece of the conversational interface will be VitalSource’s ability to provide insights on what their peers (other teachers) are using. As a central purchasing point for faculty in the US, VitalSource certainly has access to this data. Through ongoing historical data mapping, VitalSource hopes to uncover and share data to help guide faculty through the course materials decision-making process.

Of course, updating course materials requires additional legwork for faculty not just during the search and discovery phases but also in the area of course prep. Introducing new materials means editing the syllabus, studying new textbook editions for updates, as well as learning new content, add-on tech platforms, and more. To reduce this friction, VitalSource enlisted the help of Tribe AI to develop a digital syllabus builder. This tool makes updating syllabi with new course materials much faster and easier. Additionally, VitalSource proactively reaches out to publishers to gain in-class teaching resources like powerpoint decks that can be passed on to faculty and they also offer demos and trials of new technology to make adoption easier. These are just some examples of how VitalSource contributes to the inertia behind content upgrades.

There are more alluring aspects to updating course materials regularly than simply keeping faculty and students up-to-date on the subject at hand. Consistently looking for refreshed materials and approaches ensures the best learning outcomes for students. For example, add-on digital offerings like ‘Quiz Me’ have shown to keep students more engaged in class reading and better prepare them for class time.  Another example of a technology add-on for course materials is a social posting forum called ‘Hypothesis’ where students can come together virtually to share insights and answer/ask questions outside of class. When students are asked to be more engaged in their learning in these ways they are more likely to purchase the required class materials and have better long-term outcomes. When these add-on offerings are not required by instructors, students are less engaged and some opt to not purchase the required course materials, some even buying second-hand, outdated editions of materials. In fact, research indicates that student textbook sales increase from 40% to 100% when instructors utilize a digital learning add-on.

Tribe AI has an arsenal of ways to utilize indexed historical purchase data to map relational insights that can drive the trustworthy and conversational interface required to improve content discoverability for faculty.

“Improving content discoverability indicates improved outcomes for the entire ecosystem. Faculty can more easily stay on top of what’s new, best, and most impactful. Students have the resources to stay engaged and learn more efficiently. Publishers gain sales on textbooks, course materials, and digital add-ons. Everyone reaps the benefits of this scenario, including VitalSource,” said Brown. 

Trustworthy search results rely on robust, quality data

BISAC codes are used as a book industry classification that demonstrates where the book’s content sits on a tree of subjects. BISAC codes are obviously one data set to utilize for building out a GenAI-enabled interface; however these codes aren’t always reliable and publishers don’t always provide them up front. The teams knew they would need additional data sets to feed the AI in order to achieve the goal of delivering trustworthy search recommendations.

Another dataset utilized to improve search trustworthiness was higher ed institutional naming conventions like department abbreviations, course names, and term information. One challenge with these datasets is that institutions rarely adhere to a consistent nomenclature.

“Some of the data we provided to Tribe AI needed help, for sure. Although not originally part of the scope, they were able to refactor the data and begin some really intensive work surrounding historical data relationship building. They stretched our data in ways we didn’t even consider. This proved to be really impactful toward the goal of delivering trustworthy content discoverability,” said Brown.

Increased sales offsets cost per search

An important consideration when implementing a GenAI strategy is monitoring operating costs and evaluating the right business model going forward to account for them.

Throughout the POC engagement, Brown and his team saw consistent signs that confirmed the business model VitalSource had been contemplating all along. VitalSource believed that if they could achieve a truly conversational and trustworthy content discoverability experience, they would have more than sufficient growth opportunities to offset the operating costs of implementing their GenAI strategy.

Increase market share

Based on VitalSource’s current market share (roughly 50% of all US higher ed institutions), there was an opportunity for 100% growth in market share. Their research revealed that the industry had roughly one billion dollars annually to spend on course materials and it was VitalSource’s plan to capture as many of those dollars as possible through the new conversational search interface.

Encourage material updates & add-ons

Coupling VitalSource’s existing offerings, like the syllabus builder, with the new conversational search interface would make updating course materials quicker and easier for faculty. Building a healthier ecosystem –centered around easily finding what’s new, best, and most impactful– will garner more sales to offset the additional server costs. 

“I think we can all agree that the likelihood of one textbook being the best way to teach a subject year after year after year is extremely low. Even if faculty dont change textbooks each year, there are so many other supplemental materials and digital learning opportunities that can be added to improve learner engagement and outcomes. Each of these components contributes to improving the entire ecosystem and provides the publishers –and VitalSource– the chance to capture more sales, said Brown.

VitalSource’s Experience Working with Tribe

“Tribe AI’s team members are smart and truly care about understanding our users and the problems we’re trying to solve for them,” said Brown.

From Brown’s perspective, the partnership with Tribe AI has been crucial for expanding the VitalSource team’s AI capabilities especially in areas outside of their expertise, like performing really intense historical data mapping to improve the search trustworthiness. Specifically, Brown was impressed with how Max regularly added value by pushing the teams to jump straight to the finish line and explore potential viable business models. Additionally, Will was commended for the big picture insights he shared with Brown and his team regarding other AI opportunities VitalSource should consider exploring. It was obvious to Brown, too, that Nick had a background in education because he kept the teams really focused on delivering trustworthy search functionality for faculty members.

“It's really unique to have partners who can wrap their heads around not just the technology but also the business; and reflect on how one can leverage the other,” said Brown.

Faculty User Testing

The next biggest piece of learning in this work will come after the user testing is complete. During the testing, faculty will ask specific questions to the GenAI-enabled bot to guide their search for course materials. The questions will likely vary from person to person, but taking note of the questions and determining which can presently be answered with the GenAI interface will be important.

Because the scope of the POC work did not include indexing the raw textbook content, there may be questions asked by faculty that the GenAI-enabled interface cannot yet answer. However, this is something that Brown and his team are considering tackling if the user testing shows it's of great enough importance to faculty. For now, faculty do have the ability to click a link and see a digital copy of a textbook themselves to help answer any content-specific questions they may have.

Impact

The POC work was successful in producing a conversational AI interface. Tribe’s extensive GenAI deployment experience coupled with Amazon Bedrock’s managed services allowed for the rapid pace of development. Although faculty demos and testing have not yet been completed, initial internal feedback on the prototype is extremely encouraging.

“We know that improving faculty content discoverability was the right focus and that GenAI is the medium for innovation. Now, we need to keep testing and iterating until we’ve reached the point where faculty are delighted,” said Brown.

Trustworthiness of search results, which has been a major focal point throughout the project, is a needle that Brown and his team now know how to move in the right direction. Utilizing BISAC codes and performing historical data mapping both showed improvements to trustworthiness. Brown predicts that additional data engineering –including the indexing of course materials and textbooks– may provide the extra push needed to hit the desired level of trustworthiness. 

Cost per search is an element Brown has been keeping his eye on throughout the POC engagement. Although real cost per search data may not be known until after user testing when the solution goes live, Brown feels confident in moving forward with a business model where VitalSource foots the bill for the AI-enabled interface. Through the work he and his team have completed with Tribe AI, he knows the combined improved user experience and content discoverability –including increased sales and upselling opportunities – will outweigh the predicted operating costs of the solution.

The Future

According to Brown, having AI as a core piece of the VitalSource business strategy will be essential in future-proofing their business, in addition to supporting their partners. 

As the user testing results are underway, Brown and his team may pursue additional data engineering work to improve data reliability and perform additional historical data mapping to improve the efficiencies of their AI strategy. Once complete, Brown and his team believe this work could fuel other AI initiatives for VitalSources, including those centered on student insights. 

“I’m looking forward to learning more from our user testing. The faculty feedback – like how they search and what strings of questions they ask – will play a huge role in helping us shape a truly conversational, trustworthy content discoverability tool. Beyond that, there's just so much more we can do with AI. It’s very exciting,” said Brown.

Related Case Studies

Case Study

Building a GenAI Roadmap for Educational Content Creation

Case Study

GenAI Solutions: How Bright Transformed Workforce Training with Tribe AI

Case Study

Taking a Data-Driven Approach to Women's Fertility with Rita

Case Study

GoTo Revolutionizes Contact Center Quality Management with AI

Case Study

How Tribe AI Built a Model on GCP That Increased Security Questionnaire Auditor Efficiency by 55%

Case Study

Native Instruments Leverages Amazon Bedrock for Smarter, More Intuitive Search and Discovery for Music Creators

Case Study

How Tribe AI Shaped Truebit’s AI Strategy

Case Study

How Nota Built a Roadmap for AI-enabled Journalism with Help from Tribe

Case Study

Boomi Leverages Amazon Bedrock for Faster Help Desk Responses

Case Study

Togal.ai powers the construction industry into the age of machine learning

Case Study

Building a Proprietary Investment Engine Using Public Data for a Top PE Firm

Case Study

Insurance Company Uses ML to Optimize Pricing

Case Study

How Togal AI Built the World's Fastest Estimation Software on AWS

Case Study

Tribe AI & Venture Labs: Accelerating Startups with Tailored AI Expertise

Case Study

How Wingspan built a machine learning roadmap with Tribe AI

Case Study

How Tribe Helped Reservoir Bring Finance Infrastructure to NFT Trading

Case Study

Francisco Partners Accelerates Portfolio AI Efforts with Tribe AI

Case Study

Kettle uses machine learning to balance risk in a changing climate

Case Study

How Fantasmo is using machine learning to make GPS obsolete

Tribe helps organizations rapidly deploy AI solutions that have real business impact.

Close
Tribe