The

time for convening has arrived, so we'll start this evening's uh work session.

Uh item one is the data strategy analytics and AI integration update by Mr.

Keith McMillan.

All right.

Uh good evening, Mayor and Council.

Uh thanks for the opportunity to speak this evening and provide an update

on

the data, AI and innovation efforts we've been working on.

So I'll share some current projects, uh modernization of Lucidy and some of our next steps.

So there we

go.

And okay, sorry about

that.

So

late in 2025, we ran a pilot for 20 users to begin using AI.

And as word got out, it quickly expanded, and we now have about 37.

Um, but we wanted to rather than let users run free with AI, we chose to license everyone.

Um the benefits included that prompts weren't used for model training.

We were in our own space.

Um we can share our content to support queries and provide more curated answers.

We could also turn off some features like certain connectors or models that would be deemed dangerous.

We also did a customized orientation class for all users on how to more effectively use AI, what a bad prompt is, what is a good one, what you can safely share or not, and don't ask it for legal advice.

That's what Dan's for.

Um we chose Perplexity, which is one of the large commercially available uh models.

It's a blend of options.

Um it has its own, but it'll default to pick the best one based on your prompt.

We also implemented what's called retrieval augmented generation if you hear the term RAG using external sources to generate answers.

We loaded some of our own information like policy documents, Municode, some HR policies, and so on, so that users could ask detailed questions about ordinances or summarize meetings or validate policy and procedures and receive sourced answers.

Um however, the downside of that was it was actually a very manual process.

But these are some of the outcomes that we found.

Um a benefit of having that structured pilot was we were able to perform

uh

run performance management against it.

We sent out surveys at the 30, 60, and 90 day mark.

The picture on the right is actually from the dashboard I created to track this.

We track perception, how often people were using it, and if it was saving them time.

Um the 28 hour stat has been reported previously, but it's worth noting again.

Um we projected that based off of the pilot group for a full year based off their survey feedback.

Um document review, content creation, and data analysis were the most common uses.

Um examples included reviewing RFP responses, others used it to refine and create presentation or other documents.

Um it was kind of weak in the area of creating like pictures or images we found.

Um but these foundings taught us that we could scale this across the city, um, which leads us to enterprise AI.

So we wanted to scale up that success.

Um we're implementing in the process right now of uh making Gov AI available to all staff.

Um part of our enterprise agreement is training, um, introductory and advanced each one-hour sessions.

Um basic training actually started yesterday.

We had our first one.

Over about a hundred people attended.

We actually ran out of room in the WebEx.

Um the next one is Friday, and then we have two advanced scheduled in early July.

We're already seeing great engagement.

Yesterday morning we only had 2028 active users.

Uh by the end of the day, um, that jumped to 68, and um we went from 34 to 167 prompts in one day.

So we're seeing a 25% increase in in terms of usage over the past couple weeks, and that's actually jumped 100% this week.

Um we're also gonna have some office hours for more one-on-one interaction.

So if somebody has an idea or wants to flesh it out over how to use features, we'll do that.

Um we evaluated the reason we chose Gov AI is we evaluated multiple options, commercial and government based.

Um, and after talking with multiple cities and doing some demos, we found this this was the best fit.

Um, it also picks up your context from your profile, like title, seniority areas to help it tailor your its answers.