Are Districts Ready for Enterprise AI Models

On Episode 168 of the K12 Tech Talk podcast, Josh and Mark discussed Enterprise AI for School Districts. Side Note – Chris was at the K12TechPro Southwest Meetup in Dallas, Texas during recording. If you are reading this, in the Southwest region, and didn’t attend, you missed out!

Mark – Google has come out with their Workspace Gemini licensing plans, which are similar, if not identical, to Microsoft’s Copilot as well. Also, pricing wise, similar to ChatpGPT Plus. They’re being released to school districts and being touted more as enterprise AI models for the first time. Up until now, these have really been consumer tools, independent user tools. Now, these companies are building them into their enterprise environment and offering enterprise pricing.

I did get a chance to attend Google’s launching of Gemini. It was what they touted as a live webinar, but it was highly suspicious of whether it was actually live in the chat. It was very polished. Also in the chat, there was a little bit of kerfuffle regarding the pricing. Now, I mean, we’re going to talk about this in terms of Google Gemini, but really, Microsoft Copilot and ChatGPT Plus pricing is basically the same.

In each environment, there is a free tier where you can go to the website, you can use the tool, you get a very good model, it’s not their best, but you get a good model. And you can sometimes have a limited number of interactions, but really just for the casual user to go to the page and use it. In the upper tiers, though, the paid versions are about $20 a month or $30 a month across the three, which equates to a good amount of money per year per person. Yes, it gives you faster, higher quality models, sometimes multimodal meaning video, audio, and text. With some of these new enterprise tools, they’re really embedding them into the tool. So, you don’t just go to a website, it’s actually in your email tool, your document tool, your spreadsheet, etc.

So, Josh, these enterprise models are out. What do you think? Where’s the place where they live? Is this something that you’re going to buy into? Are you considering it? If you’re not considering it, what’s their market? Because clearly, people who are marketing these tools, they’ve done their research.

Josh – I don’t see my district buying into it. Just because I don’t think there is a need for it. I think you and I have talked about districts that are larger districts that can point an AI tool at their board policy, and say, you know, read all of these board policies, and then more or less create, like a virtual assistant, that staff can ask questions, and it would be able to pull that information out of the policy bank. I think that’s brilliant. I think you could save an FTE or a couple FTE out of the benefits department or help the HR department by doing that. But again, I think that threshold, not super large school districts, three quarters of the state of Missouri for example, don’t need this because they’re small. I love the idea of having it wrapped into Docs and email to help you with grammar and creating content and maybe tone of a document or tone of an email. But, it’s not 20 bucks. It’s not worth 20 bucks to me for that.

Mark – It’s a hard one because I was talking to a colleague today about just this exact topic of whether or not we would pay for it or not. And, if I look at how much time this is going to save me, if it’s emails, if you can summarize these long emails, I think if I were to spend a day with these Enterprise models, going through emails, I know I would see 10% to 20% time increase between reading and replying. At the very least, I’m going to see like a 10% increase in my productivity. If I looked at my hourly wage in my time over a month, that $20 a month is beyond worth it, it would pay for itself the first day. At the same time, I’ve got to look at if I buy this for the entire district, am I going to see that same 10-20% performance boost? I don’t. I don’t know if we’re mature enough to be able to say that or see that yet.

And I think there’s a difference between instructional use of these generative AI models, as well as back office.

If I do the math, I would be asking just for staff, just for staff, I’d be asking for $3 million. I have to do the math, the number of employees monthly, annually, it just doesn’t line up. And I think the hard part of everything is that this is a bad budget year for a lot of people. A lot of school districts are making cuts. They’re making reductions. Very, very few people are making major investments. Because as we talked about the podcast, ESSER is going away. I think Google was trying to sneak in this announcement and same with Copilot – Can we get some of that federal funding, like schools are just throwing and trying to spend before the deadline? I don’t think it’s going to work, though.

These are expensive models. These AI models are incredibly expensive to run. And, you know, it’s something like 7 to 12 times the processing power to use an AI search versus just a plain old web search. They have to cover their bases. I think if I could look at these folks critically, I would say you need to start to market to smaller audiences in terms of don’t just leave it up to districts to imagine if they’re going to buy this. They should not recommend that all staff use it. Buy it for maybe your superintendent and your business person and your finance person and maybe a school secretary or to try it out and see how it’s used. That will sell itself if you get it to the right people.

My final thoughts. One is, let’s say I buy this for everybody. How many of your email conversations for the next week are going to be entirely generated and responded to with AI? Everybody’s going to be so excited to use it that they’re going to be sending and replying to emails via AI. And then you just step back and watch the chaos.

Also, in that Google webinar, talking about data privacy, they did say that you are getting a model that is specific to your environment, and it’s not going to look through other domains, other districts. Your data isn’t going to train to model.

Listen to the full episode here.

Leave a Comment