Q2 Earnings Roundup - EdTech Generative AI

A roundup of LLM and AI discussions from Q2 EdTech Earnings

With August almost behind us, most public education companies have released their earnings. In this piece, we’ll be breaking down how a few of edtech’s most important companies are thinking about AI developments.

• Duolingo

• Powerschool

• Coursera

• Docebo

• Instructure

• Nerdy

---

Duolingo

Since its inception, Duolingo has been at the forefront of leveraging ML/AI to make language learning more personalized, accessible and engaging. They have been building their AI team for more than 10 years and have created proprietary systems that now drive many aspects of their products and business.

Duolingo currently deploys AI in both user-facing and internal capacities.

• Birdbrain: proprietary learning model that makes Duolingo more personalized and effective. With data from 1B+ exercises completed each day, Birdbrain can figure out which exercises in a particular lesson are going to be the best match for a learner’s level of knowledge.

• Creation of course content: instead of experts spending time creating thousands of sentences and exercises, Duolingo works with GTP-4 to speed up script-writing for Duolingo Stories and efficiently scale course content.

  • They are focusing on (1) more advanced content - getting all of their courses to teach through high levels of proficiency, and (2) more interesting content by adding narratives and storylines with all of their characters.

• Duolingo English Test (DET): uses AI in every step of its process - from creating content at scale, to administering, scoring, and securing the test. DET has been working with OpenAI since 2021.

• Duolingo Max: the higher subscription tier that they are testing. Features are aimed at teaching better by emulating a human tutor experience.

  • Explain my Answer: explanation when you get a wrong answer

  • Roleplay: allows users to dynamically practice conversation

  • In-lesson Coach: before you submit an answer, a coach will give you hints

  • They are continuing to test Max and its price on a very small fraction of their users. Right now, it’s roughly 2x the price of their Super offering

Other notes:

• Despite some vendors beginning to build customized LLMs for their own content instead of bearing the cost on a per interaction basis, Duolingo is sticking with OpenAI. “We're very happy with it. Our belief internally is that the cost of these models is going to go down.”

Using new technologies to make excellent products takes time to get right, but we think we are uniquely well-positioned to do that—we have experience with AI and an enormous user base that will help us test new features and products. - Luis von Ahn (CEO, Duolingo)

AI was mentioned 20 times in Duolingo’s Q2 Earnings Call.

---

Powerschool

Powerschool has used AI for the last couple of years in their systems on risk interventions predictability, as well as in their learning navigation/personalized learning pathways. They recently announced collaboration with Microsoft Azure, OpenAI Service and the usage of their open AI large language models.

• Connected Intelligence DaaS: the industry’s first turnkey, fully managed “Data as a Service” platform for education and government agencies built on Snowflake Data Cloud. This will be connected with their Unified Insights analytics to provide customers with limitless capabilities to understand the trends, factors, details and gaps to improve education outcome across K-20 longitudinal data systems.

  • A talk-to-data functionality that will help non-technical users offer solutions that provide powerful data insights easily.

• Performance Matters Assessment: an automated generator that creates assessment questions and lessons that are aligned to a desired learning objective, grade level, subject and standard, which the teacher can use to deliver more frequent formative assessments at scale.

• MTSS and learning navigation: to deliver personalized intervention and learnings based on individual student needs and learning goals.

• Other notes:

  • “We are seeing a lot of interest in education actually to present one of the biggest opportunities for generative AI. A lot of the school districts actually see a clear need…we are the system of engagement, and we have the system of intelligence with a product, we are in a unique position to be the platform for all these school districts and state to leverage these and monetize that effectively.”

You're actually making their teacher’s life easier and you're [giving] a student help right on their day-to-day homework. So that's a very unique opportunity and we expect again the – being able to really have a huge – a market advantage of this by – as we grow all these systems. - Hardeep Gulati (CEO, Powerschool)

AI was mentioned 27 times in Powerschool’s Q2 Earnings Call.

---

Coursera

AI has the potential to impact an entire new class of knowledge workers, unleashing a new wave of reskilling imperatives. Coursera is at the forefront of providing a powerful blueprint for the types of innovation that will be required to keep pace with our rapidly changing world. Ongoing product innovation across Coursera’s platform:

• Coursera Coach: a virtual learning partner powered by generative AI and grounded in expert content. The coach will also be able to provide quick video lecture summaries and resources, such as a recommended clip, to help learners better understand a specific concept. It is designed to allow learners to ask questions and receive personalized explanations and answers, get personalized evaluations and feedbacks on their submissions.

  • Beta version of Coach has been launched to Coursera Plus subscribers. They continue to be excited about the early feedback from these technologies when paired with the trusted authoritative content from Coursera partners

• Coursera ChatGPT Plug-in: enhanced personalization and discovery across the Coursera catalog. Like an academic counselor, the ChatGPT plug-in allows learners using GPT4 to identify recommended content and credentials based on the subject or career field the learner says they're interested in exploring.

• Machine Learning Translation: uses technology to dramatically reduce time and cost of producing high-quality trust and content at scale, including localization. In Q2, they delivered the first milestone with subtitle translation for 2,000 courses in 7 different languages. In Q4, they will begin to roll out the full course translations to learner's around the world.

• AI-assisted Course Building: based on a few simple inputs from a human author, a new set of AI-powered features can auto-generate course content  — such as overall course structure, readings, assignments, and glossaries — to help educators dramatically reduce the time and cost of producing high-quality content.

• In addition to degrees and artificial intelligence, Coursera’s leading educator partners have been focused on creating new projects, courses and specializations to meet the demand for Generative AI learning content.

• On M&A in AI, “frankly, we will probably do it more for the talent on the actual technology because these large language models are moving very quickly. We see early-stage companies, whole products get subsumed into the capabilities of a new base language model. So we're kind of holding back on that a little bit.”

Today, we're witnessing this time horizon compressed dramatically from several years with the Internet and mobile computing to a matter of just months with ChatGPT. We believe that AI will be a general purpose technology, representing the next major technological shift that will profoundly change how we live, learn and work. - Jeff Maggioncalda (CEO, President & Director, Coursera)

AI was mentioned 18 times in Coursera’s Q2 Earnings Call.

---

Docebo

Docebo’s learning suite offers tools to span the entire enterprise learning lifecycle, leveraging AI innovation to transform the delivery of personalized learning at scale. Docebo’s innovation roadmap is focused on a few strategically critical areas:

• Docebo Shape: rapidly automating the creation of e-learning course content - automatically turn internal or external knowledge sources and content into short, engaging “learning pills” in multiple formats. Content is kept fresh with versioning and leverage automatic, AI-based translations to publish multi-language content within minutes.

  • “Docebo Shape allows customers to reduce [course creation] from 40 days to like 20 minutes, [with] the speed of deploying faster learning live in multiple languages”

• Docebo acquired Edugo.AI: Docebo’s acquisition of Edugo is focused on two main objectives: enhancing its existing AI capabilities with Edugo’s team of expert AI engineers on board and adding new capabilities to the Docebo platform to better serve its customers.

  • They are exploring the possibility of building their own LLMs to run AI inside a government cloud without going outside non-federal compliant environments. “We bring the AI in an environment which is highly protected without going and connecting APIs outside the FedRAMP-certified environment.”

• Virtual Interactive Capability – introduce capability to provide solution-specific AI simulations, allowing learners to practice soft skills with real time AI feedback.

• AI-based Skill Tagging – leveraging AI to automatically match ontologies and ensure consistency across systems, as skills ontology varies from one customer to another.

• Hyper-personalized learner experience – learning paths can be adapted to the learner based on their content interaction in the platform.

• Deep semantic search – improved content accessibility and embedded search functionality is expected to enhance the dependability and rapidity of search results based on the learner’s needs.

We started investing on AI 4 years ago. And what we have learned is that the biggest problem of AI is not creating a nice product. But first of all, make it scalable - to build AI that can serve a million of learners. And this is more complicated than creating an appealing content generator. - Claudio Erba (CEO, Docebo)

AI was mentioned 26 times in Docebo’s Q2 Earnings Call. More updates about Docebo’s AI development roadmap to come at Docebo Inspire in September.

---

Instructure

In the last 18 months, Instructure has worked with customer user groups to understand what areas that they should make investments and innovations - new solutions to save educators time, personalize learning experiences for students and simplify complex tasks for admin. At InstructureCon, they previewed some of these AI strategies:

• Partnership with Khan Academy: to bring together Instructure’s structured learning platform with Khan AI-powered student-tutor teaching assistant Khanmigo, leveraging generative AI to empower educators to meet students where they are in their educational journey. They are working on an AI writing coach in partnership with Khan Academy to deliver human-driven, technology-enhanced essay feedback and grading, lesson planning and rubric creation.

• Emerging AI Marketplace: which gives the educator visibility and access to the AI solutions that are integrated into the Instructure learning platform. “We are working with our customers to define privacy and security standards to ensure AI solutions are safe and each partner in the marketplace is committed [within standards]. In just the last two years, we have nearly doubled our partner base. Our investment in the Instructure Learning Platform gives our customers access to innovation across the [entire] landscape.”

• Panda Bot: AI-powered support chatbot to deliver instant support to all Instructure Community members in a personal and personable way.

• AI-assisted Course Design and Templates: gives educators and instructional designers the power to quickly create elegant page layouts using natural language instead of HTML.

• Semantic Search: simplify finding course materials on contextually relevant subjects using natural language.

We had a lot of sign-ups [or] customers that want to be part of the early adopter program as we work through some details on these. In general, very positive [feedback]. I'd say, it's still early days as far as what impact this is going to have on the business model long-term and we're working through those with our customers as we work to understand which of this functionality, the value that is bringing results and changes in pricing or what the cost structure looks like. - Steve Daly (CEO, Instructure)

AI was mentioned 10 times in Instructure’s Q2 Earnings Call.

---

Nerdy

Over the last 6 years, AI has been foundational to Nerdy’s platform. They have been using AI to identify the highest quality Experts, assess Learners’ foundational knowledge, help ensure the right Expert-Learner match, and drive operational efficiency. They have actively been infusing generative AI into their products to supercharge human interaction. Nerdy’s teams are encouraged and pushed to leverage AI in their daily work to support high volume activities at scale.

• AI for HI: “Artificial Intelligence for Human Interaction”, significantly reduce the labor needed to operate their platform.

• Nerd AI: Nerdy’s dedicated generative AI infrastructure team has built out and deployed shared capabilities and APIs that enable multiple different engineering teams to leverage common generative AI capabilities via APIs with established best practices.

• Nerdy AI Playground: an exploratory environment that embodies Nerdy’s innovative spirit, the “AI Playground” serves as an internal platform enabling teams to experiment, learn, and prototype AI applications.

Customer Facing AI Enabled Innovation:

• Varsity Tutors AI Content Generation: rapidly deploying learning content across 200 of their most in-demand subjects with 66K+ AI-generated practice problems, answers, and explanations. By year end, they expect to be able to 10x the quantity of learning content generated by the new AI system.

• AI Tutor Chat Improvements: employed an upgraded AI model and refined metrics to evaluate each conversation, in order to sharpen their understanding of the AI Tutor’s strengths and areas for improvement.

• Moderation and Redaction Services: leveraging state of the art AI, they implemented monitoring and safeguarding of customer information and harmful content through PII identification, redaction, and AI classification techniques. This will allow them to have complete audit-ability and control over AI products and services.

• Speech Transcription: using speech recognition AI, they are automating Nerdy’s rich audio data between customers, sales, support, and Experts to more comprehensively predict insights, understand customer satisfaction and engagement.

• Summarization: AI-based summarization capable of distilling precise and accurate summaries from interactions across all organization levels (Expert-Learner engagements to customer service)

• My Learning Hub AI-generated Learning Recommendations: predict and suggest the next product interaction across learning formats and subjects that are most likely to drive engagement and customer value.

AI was mentioned 1 time in Nerdy’s Q2 Earnings Call.

---