Product Development Industry Insight

Why One Curriculum Provider Is Banking on AI for the Future of Teacher Support

The CEO of Collaborative Classroom Reveals Its New AI Assistant Tool, And How The Organization Is Strategizing Its Investment
By Emma Kate Fittes — October 04, 2024 12 min read
2 15 23 mb k12 insider ai 1245955129
  • Save to favorites
  • Print

When leaders at Collaborative Classroom began to learn about the possibilities with generative AI, they faced a critical question: Is the investment worth the risk?

It’s a question that all companies — especially smaller ones — face given the uncertain legal and regulatory environment with the technology.

There’s no guarantee district administrators will react positively to the development of an AI product. And creating one under any circumstances can be expensive and time-consuming.

For the nonprofit literacy curriculum provider Collaborative Classroom, investing millions of dollars in generative AI represents a significant chunk of its budget.

About These Analysts

kelly stuart

Kelly Stuart serves as president and chief executive officer for Collaborative Classroom. Stuart has worked with educators in schools and after-school sites in every state. Before coming to Collaborative Classroom, she worked in literacy and research-focused organizations (Success for All, WestEd, Education Partners). She began her career as an elementary school teacher and coach in a small rural community in Northern California.

Weiermiller Liz

Liz Weiermiller serves as the digital learning manager: AI innovation for Collaborative Classroom, where she is responsible for managing the development and maintenance of AI support and the Collaborative Classroom Support Center. She joined the organization in 2019. Previously, Weiermiller spent more than 15 years as a classroom teacher, reading recovery teacher, reading interventionist, instructional coach, and adjunct professor.

The nonprofit expects to release its new generative AI-powered chat feature, CC AI Assistant, to teachers using its curriculum in the spring, after months of testing that is already underway.

The tool will allow educators to type in any question, whether it’s a simple troubleshooting issue or a complex question about a specific sticking point for students, and get a detailed answer within a few seconds.

The AI’s responses are pulled from all of Collaborative Classroom’s resources, including things like implementation guides, example lesson plans, and internal knowledge support teams have gathered from years of fielding questions and concerns from teachers.

It will be added to the organization’s suite of support and PD offerings, which includes a learning portal and optional in-person trainings.

For Kelly Stuart, Collaborative Classroom’s CEO, the coming months will be about navigating all of the uncertainties that come with the decision to bank on AI. Her team is preparing to combat questions over the feature’s accuracy, potential for bias, and reliability.

But she maintains that it is worth the risk, given the need for support she’s seen in schools, at a time when funding for education is shrinking.

“Publishing companies ... have to do more than just give them new materials,” Stuart said. “They need the support to go along with it. As much as people can be thinking about how to support every single teacher once they actually get the curriculum, the better the whole system is going to do. And that’s why we’ve stepped into this world.”

EdWeek Market Brief recently spoke with Stuart and Liz Weiermiller, who is leading the CC AI project, about the decision to invest in generative AI for professional development, how the initiative has been received, and why they believe it is the best way to meet districts’ needs in a post-ESSER market.

This conversation has been edited for length and clarity.

Tell me about the new professional development system you are working on.

Stuart: One of our challenges — and a challenge that I think every organization creating curriculum has — is supporting teachers at scale ... with professional development. With big contracts, you still only reach a handful of teachers in that process, and it’s very expensive.

We’ve been at this for a long time trying to support teachers in the curriculum itself. That is very educative, that teachers learn as they’re teaching it. Then we’ve had this live chat going on for a long time [where] people can come to our learning portal, which everyone has access to if you have our curriculum, and ask questions. So we’ve built this huge bank of responses.

Basically, last year, we decided to make a pretty big investment in creating our own well-trained chat bot. The name is CC AI. So we’ve been hard at work, doing all of that work and testing how accurate CC AI is — and it’s wildly accurate.

How does using generative AI change the experience for teachers?

Stuart: Now we see a whole layer of support that any educator at any time can come to — in our portal, that’s already very safe and secure — and get a high level of response. Our goal is that we can support probably 60 to 70 percent of most educator needs in our curriculum with [the CC AI tool] alone.

Can you explain how this is different than the basic chat bot that many people are already familiar with?

Stuart: A lot of chat bots, historically, that we interact with work on an “if, then” system: If somebody says this, then this happens ... and then you get stuck and everybody gets frustrated.

The whole power of generative AI is that there’s so much data in there that it can be a lot more helpful and responsive. So that’s basically what we’ve been able to build because we’ve spent years fielding all these questions that [educators] have and banking them.

See Also

Image of woman using laptop computer.
Getty

One of the things we found is that [Weiermiller’s] team hasn’t answered a new question for quite some time. Which tells us we probably have a very extensive data set on the kinds of needs that our educators have. Without that investment of running this live chat and all of this ticketing for so many years, just starting fresh with none of that content, it wouldn’t be a very powerful chat bot. But because we have all of this work, we’ve been able to get a really great data set together. That’s the big advantage.

Weiermiller: When you think about educators, they have students who have very individual needs ... but just based on what we’re able to provide, we’re able to help teachers support their students. So maybe I have a student who’s struggling with [a particular skill], what should I do? We’re able to mine all of our resources and provide the best resource possible for a certain scenario.

Was the AI tool trained using only your content, or does it pull from other sources?

Stuart: Only our content. We feel like if you feed it a very healthy diet, it will give healthy things back. So it’s only trained on our stuff. It’s our programs itself — it’s all these years of Q&A, it’s the knowledge base that our professional learning folks have had in the field all of these years. That’s what it’s built on.

You mentioned the tool is testing as very accurate. What has your process has been like to evaluate that?

Weiermiller: Our first phase was internal — where we just use our internal, small group of people who knew about what we were going to be doing and asked questions and then evaluated the responses ourselves based on three categories: “accurate yes,” “accurate no,” or “accurate yes, but.” With “yes, but” something may be misleading. Based on how we evaluated that, then we added additional context for the knowledge base of our AI.

Once we were comfortable with that, we moved down to another phase, broadened our scope of people who were testing, followed that same process, but got some additional data. Each time the data is improving. Now we’re up to 25-30 people [testing the tool], all affiliated with our organization, but some are full-time colleagues, some are our cadre members who are working in schools and districts.

One of the things we found is that [Weiermiller's] team hasn't answered a new question for quite some time. Which tells us we probably have a very extensive data set on the kinds of needs that our educators have.

Based on that process, we are at a really high level of accuracy. I believe, in the AI world, 60 percent accuracy is a good number. We’re hovering around 90 percent.

Based on your expression when you said 60 percent accuracy, I take it that wasn’t your goal?

Weiermiller: Well, yeah, especially when we’re dealing with like educators and students, right? And we want our educators to feel supported. We don’t want them to feel like they’re coming to us and getting inaccurate information. It is super important to us.

What made your organization decide to make this investment, and what was the relative scale of that investment for Collaborative Classroom?

Stuart: Just as a reminder, we’re 100 percent nonprofit. Almost everyone in our space is a for-profit company. So for us to make an investment like this, it’s a very big decision. We only have a small pile of cash that we can invest each year, and it’s all based on how successful we are. We don’t get a lot of money from foundations, we don’t have venture capital, we don’t have private equity.

We’ve always said: How do we support the hundreds of thousands of teachers? And we’re never going to get there with our humans. School districts can’t afford it.

We had been working with a group called Javelin Learning for a few years, and they helped us build a coaching platform. And they have been really leading some of our thinking around what’s possible with generative AI in learning. They come out of healthcare learning, they’re psychometricians, psychologists.

All last year, we started to work with them and see examples of what was possible. By April, I had worked with my board and said, “We’re going to make an investment in this.” It’s a couple million dollars investment for us — which for us is huge. It’s a very big deal, but it’s all to try to support teachers and leaders. It’s not to try to build something to sell to another firm at some point. It’s really, how can we support teachers?

Why focus on teachers versus trying to implement AI into something student-facing?

Stuart: We really see a lever of change with teachers. It’s why we develop the curriculum that we do in the ways that we do. And I also think there’s a lot of fraught things right now with student-facing AI. We are seeing what’s happening, and we feel like, if we can support teachers really well, then they can support their kids really well. And if we can help them at the moment that they need it in small chunks of learning, that could be really helpful.

We also see this as a safe space to ask questions. Sometimes teachers have a curriculum for a couple years and might not be comfortable saying, “Gosh, how do I actually get my kids placed appropriately in certain parts of the lessons?” This gives them a way to go to a very safe place and get some answers.

As we’ve been showing this to our district leaders, they’re also seeing a big time savings with their own work because those district literacy coaches often are answering the same questions over and over again. So if we can kind of deploy the humans to the more complicated things and use something like this to answer the types of questions we know people have when they get new curriculum, when new teachers come into a system, that this can just provide a huge level of support in a school system.

Can you give me an example of how this works?

Weiermiller: [Using a test version of the tool,] I’ll just populate like a quick question that is something that an educator would ask: “What if one of my students does not pass a SIPPS mastery test?” And we’ll see what CC AI has to say.

For a new educator, they could find this answer in our program materials, but it would take a lot of digging, maybe some talking with a coach. However in just a matter of five seconds, we have a super accurate response that tells me that I need to target the phonics patterns and the sight words and that the passing criterion is 80 percent. [It also] talks to me about slowing the pace of instruction, and I can even ask a follow up question.

I would spend hours reading through the materials, trying to find the answer. I had two-week check-ins with a consultant, so oftentimes I would wait for those two weeks to be able to get answers.

It can also be a technical-related question, too, because all of our resources are on our digital platform. So, it will give me some help. You can see here now, it’s asking me if I want to connect to a live agent if something doesn’t work. And so we’re developing a flow for how this will then escalate to a person if the needs aren’t met.

Are there any features you are still debating? I saw a document upload symbol, is that part of this?

Weiermiller: Yes. So if I wanted to upload something like, I could add something here, like a file from my computer. [CC AI could say,] this looks like the handwriting stroke sequence. And it might refer me to where in the implementation handbook I could find it, in what particular section.

We’re not [sure] whether that feature is going to be included, just because we imagine a lot of educators might upload student data that we don’t necessarily need to see. We don’t want to see actual student names or anything like that. So the icon that is purely there right now for a testing purpose, and it’s to be determined if that would be included.

What are you hoping that educators get out of it?

Weiermiller: I was a coach in a school district using Collaborative Classroom materials before I was working full time for a Collaborative Classroom, and I just remember I would have so many questions coming at me from the educators I was supporting that I didn’t know the answer to as a coach.

I would spend hours reading through the materials, trying to find the answer. I had two-week check-ins with a consultant, so oftentimes I would wait for those two weeks to be able to get answers. And [then] the answers are really no longer relevant to the teachers, because so much time has passed.

I just think about how our teachers will be supported, which will translate to a higher level of student achievement. For me, that is what is most exciting about this.

Have you had to navigate any concerns related to the use of AI, either from district clients or internally from employees worried about its impact on their job?

Stuart: We’re just starting to work and talk with our districts. Before we got started, we interviewed a lot of our district partners and showed them some things. It’s going to be really important that people understand that they are interacting with AI. So we’re going to be super upfront about that. We’re also going to be really upfront about where the data is sourced from. It’s all Collaborative Classroom data.

We’re also going to be using some of our humans to be constantly checking what the what the tool is giving back to people. So we are shifting people’s internal roles to start to look at that. Some of our agents now may not be answering as many live questions, they may be actually monitoring what’s happening with CC AI’s responses. So there’s some redeployment there.

Because we weren’t an ed-tech organization or ed-tech forward, you can imagine some of the internal discussions about it.

How have you successfully eased people’s fears about AI?

Stuart: One of the things we’ve been able to do is kind of bring people along with us, show them everything, be really upfront about everything.

The other big piece is, because this is all going to be happening in our learning portal, we’ve already met all the security standards that districts have. This is already where teachers come to access our curriculum and their materials. So it’s in a very protected space.

Post-ESSER, what kind of demand are you seeing for PD from districts, and how do they want it delivered?

Stuart: This is our biggest year for professional learning, so we’re busier than ever. I think districts who’ve made big investments in making shifts in their curriculum have also aligned a lot of their PD purchases in the same way.

One of the things I think we’re going to see, obviously, is cost [being a big factor in district purchasing decisions], so having something like CC AI available, having something like our asynchronous coaching — which is a much lower cost than some of our in-person work. I think we’ll always have a blend, but it’s going to get harder in these coming years, for sure, with the loss of ESSER funding. For now, we’re still very busy with professional learning.

Webinars & Quarterly Briefings

Education Market Webinar Key Trends to Watch in the K-12 Market in 2025
We'll look at emerging district demands, changes in state and local policy that will affect school spending, and more.
Sales & Marketing Quarterly Briefing Data Dive: How Districts Are Making Decisions About Academic Resources, Post-Stimulus
An in-depth look at what EdWeek Market Brief’s survey data reveal about district decision-making heading into 2025.
Education Market Webinar The State of the K-12 Industry in 2024—and Where the Market Is Headed Next
In this webinar, we’ll break down highlights from our annual State of the K-12 Industry Report.

EdWeek Market Brief offers actionable K-12 intel for business leaders powered by original data, research, and reporting.

Read Next

Product Development Market Analysis HMH And NWEA: An Inside Look at What’s Happened Since the Deal
The major companies have made an ambitious bet on creating an integrated curriculum and assessment platform.
10 min read
Contemporary art collage. Hands holding puzzle pieces and trying to connect it against blue background.
iStock/Getty
Product Development Industry Insight Why Accessibility Will Be the Next Big Demand From K-12 Districts
Many innovations are likely to focus on helping teachers focused on disabilities improve their crafts, a prominent advocate predicts.
10 min read
Artificial intelligence concept. Ai child's face made with circuits.
iStock/Getty
Product Development Market Analysis With Few Standards on AI to Guide Them, Education Companies and Districts Stare Down Big Risks
Companies and districts alike are trying to hone in on what makes up responsible AI.
9 min read
Conceptual illustration using icons and vectors: AI ethics and legal concepts artificial intelligence law and online technology of legal regulations Controlling artificial intelligence technology is a high risk.
iStock/Getty