WorkAI and Your ChMS: Safe UsageHow to use AI with your ChMS and keep the keys with your staff. CITN 2026.Oct 4-6, 2026Church IT Network, LouisvilleConference siteRelatedOther talks and videos from this site.Nov 3-5, 2026Ministry By DesignNov 3-5, 2026Seven Systems That Turn Curious Visitors Into Church FamilyRX25 · Rock CommunityFrom SEO to GPO: Unleashing Your Congregation as Digital MissionariesAI has disrupted SEO, but the opportunity for the church has never been greater. Learn how to shift from chasing algorithms to equipping God's people with content that is helpful, human, and shareable.Dec 22, 2022 · 20 min · TriumphCheck-In Celebrations & Achievement LabelsA Triumph video on using Rock Streaks, Achievements, and Celebrations to help kids (and adults) engage.RX2022 · Rock Community · with Lee PetersonCheck-In Celebrations & Labels: How Do You Get Kids to Engage?A short tour of Rock Streaks, Achievements, and Celebrations, and how they can help your ministry encourage kids (and adults) to engage. RX2022, with Lee Peterson.TranscriptSelect a line to jump there.0:03Welcome. Come on in.0:04We're just getting started.0:06I'm glad that you guys are here.0:08This is my first CITN conference.0:11So how many of you is your first CITN?0:14Hey, there we go. I'm not alone.0:16Awesome.0:17How many of you have been coming for five years?0:21Seven?0:22Six?0:24Okay, five years. Oh, okay.0:27I would expect you to keep your hand up.0:29So ten years? Fifteen?0:31Uh, yeah. 20? 9. Sorry, just congratulations on home prize up here.0:37But, I'm going to talk today about AI and your ChMS0:42your church management system.0:44There's a lot of great sessions around AI from some of you0:47learning0:50We'll talk about maybe some different resources that we have available to you0:54but I really just want to be helpful.0:58As you guys are listening, if there's anything that you guys have questions1:02about, just let me know.1:03I'll leave some space at the end.1:04And hopefully we'll have enough time.1:07And then hopefully my battery on my computer so I have power up here.1:11So, if you don't know who I am, my name is Brad Erb.1:14I am a ministry analyst at Triumph Tech.1:17And Triumph Tech is a provider for Rock RMS.1:20We help churches get on to Rock.1:22So be supported, websites, mobile apps, TV apps, that sort of thing.1:26I spend most of my week inside of Rock with different church teams1:29helping them navigate different solutions.1:31I also lead our AI Agentic Cohort, which is all about translating a workflow1:36that you might have into something that is an agent-type workflow to create some1:40more margin for your team.1:42And so most of what you hear today is going to be like tested for real1:46it's things that we've talked about in our cohorts or things that we're actively1:49working through with churches.1:51When I'm out in Rock you'll find me tinkering, working out1:53hanging out with my family, I've got a wife and two kids1:55with a third one on the way1:58so you can2:00understand that our life is a little bit hectic.2:02So two toddlers, five and three.2:04This is what we're going to be talking about today.2:06So it looks like a lot, but we're going to start talking about why AI in your2:09church management system2:11matters right now.2:12We'll talk about what good AI software development can look like.2:16And then there's like six things that I think we can look at Rock for as a case2:20study that you can take back to your other platforms and your conversations.2:23When they built their AI agents.2:26And then How do you make it practical for your church?2:29I think this is the thing that is most important to make.2:31How do you make it practical so that you can understand what is safe?2:37You know, what would a one-page policy be like to take into your leadership?2:40And then how do you stay accountable when AI doesn't work and actually put into2:44practice?2:45And then those questions I mentioned earlier that we can kind of take back to2:48our different platforms.2:50So before we go any further, I just want to make sure that I've defined two2:54things, chatbot and agent.2:58So chatbot answers, so you'll think about chatbot or chatbot answering a3:02question, it's time to go back and forth, an agent will ask3:05you give it a job, you plan the next steps, and you still use tools to search3:11your different platforms or things like Rock.3:14You could draft an email and it keeps going until the job is done or it hits the3:17guardrail or stopping point.3:19The AI underneath can be the same in both those cases3:23but the risk is very different, right?3:25When a chatbot gets it wrong, you get a bad answer that you can't ignore3:29or a bad answer that you can't ignore, you can just say3:33okay, I don't really need that.3:34I can keep the board when you When an agent gets it wrong3:37you have to clean up.3:39There's an action that needs to be taken from your team.3:41Maybe the wrong person got removed from a group.3:44Maybe an email went out to 40,000 church members.3:46I don't know.3:47We're not speaking from experience up here.3:50Agents are what are showing up inside of our church management system3:53right? And they're happening right now.3:55We hear, I mean, talk to any vendor.3:56They're figuring out ways to apply these different solutions.3:59So how do we navigate that today?4:01So AI is here, whether we like it or not.4:04Most of us have tried AI and liked it.4:07Not like it. Very few churches have changed how they work with their team.4:12Watching. Here agents are what are showing up inside of our church4:18in Q2 by KPMG, they're like a consulting team.4:21They surveyed senior business leaders every quarter about AI4:25and 76% said that AI is already delivering real value to them.4:30That is crazy, three-fourths of the organization.4:33Except, it's not three-four the organization4:36only 22% said that they reached a4:37point where they're actively rolling it out across the whole business.4:42That's a disaster, right?4:44Everyone in the survey had access to the same AI tools that you and I do.4:49The difference comes out to the rollout.4:51A lot of people tried it, liked it, and maybe stopped at the pilot or the4:55experiment.4:56What do we do with that?4:57Well, rollout is the bottleneck.5:00I think this is the number I would write down, this 3X.5:04Organizations where one leader is clearly responsible for AI are about three5:09times more likely to report on that real return on investment.5:13And I want to be careful here, because this conference is full of IT folks5:17right?5:18And I'm not saying that you're the problem, alright5:20I'm not.5:21The reality is, when AI gets handed to IT as a project5:26it tends to be treated like a software to install5:28And installing software doesn't change how anybody works.5:32The church has seen the results of one leader whose name is on it.5:37Like they're responsible, they're owning this move5:39this transition.5:41Sometimes that's you.5:42Sometimes your job is helping your executive pastor make the change.5:46How many of you are5:49in a leadership role or executive team member at your church?5:52See a couple hands, okay, cool.5:54How many of you are actually doing the implementing.5:56The IT professional, the operator, a few more hands.5:59Okay, cool.6:01Any other roles I didn't get?6:03Shout them out.6:05Okay, great. So what do we do?6:09I think the good news is you probably6:11have kind of like owned more AI in your infrastructure than what you might I6:16feel like you have a better thought at first.6:19If an executive pastor is tasking you with rolling this out6:22you probably have a suite of products that already have some sort of an AI6:25infused within it.6:26And I'm not saying you should go and just use that6:28but I am saying that it gives you a place to experiment.6:31And so before buying anything new, just take inventory.6:34So you got Microsoft 365, they've, of course, features on several plans.6:38Maybe your church system is adding AI agents.6:41I bet that probably already has a ChatGPT or CLOG3 account that they're actively6:46using.6:47So having the tools was never6:49the hard part. is getting people to use them well and to use them together6:54and that takes leadership and a little bit of work.6:57So here's what we've been experiencing.7:00We've already had people in this7:03environment who have said, hey, I went through this AI agenda cohort.7:06I'm gonna throw up some QR codes on the screen.7:08Don't feel the need to scan them.7:10I just tend to go through them really quickly because I'm just kind of warning7:12you now.7:14But I think about a guy named Jesse McComb, he came into our AI Agentic Cohort7:19with one monthly app and came in an entire afternoon.7:22And now, by the second session, he could get his agentic workload to run under7:2810 minutes. That saves him a bunch of time.7:30What was even cooler is he- handed it off to somebody else.7:32He doesn't have to worry about it anymore because he made it simple enough in7:36terms of process.7:38Rachel Bobs built a cloth setup that catches her constant pains and7:41interruptions so that she can stay focused on the work in front of her.7:44Tim Newton went and showed his team what.7:45He learned and they started using it that same week without waiting for another7:48training. They started implementing it right away.7:50He called it the game changer for his team.7:53None of these three are developers.7:55They're not IT professionals.7:56They're administrative staff.7:59And 75 Church staff have already kind of been through the cohort now.8:02And we've got another one coming up.8:03But Regardless whether it's this one or something else8:07I want you to take the time, be intentional, about creating training for your8:10teams to go through because that's gonna be an important part of the rollout.8:13John Edmondson, the founder of Rock, he said as he's talking with the core team8:19to8:21as part of this AI summit that happened earlier this summer8:25he said this quote that stuck with me, the goal of AI is not to take the person8:28out of ministry, it's to put more people in the ministry.8:30And the reason why I said that was because people were spending all of their8:34time just like solving their own busy work and workflows.8:36And so he's really, challenging, you know, hey, every hour that we hand back to8:41AI, it should be an hour that we were able to hand back to a person in the8:44ministry. And that's like a test that I'll personally use on some projects.8:47You know, if I'm working through some things for some church partners8:50or you know, serving with my church, I'm asking, did this free somebody up to8:54spend more time with people?8:57So be asking that question for your team.8:59To know that well, you've got to make sure that you .re able to trust the AI in9:02the first place. You can't be handing off something that you can't trust.9:06This is where I think it's helpful to understand how Rock built their9:10you know, approach their AI development.9:12Not just for the product side, but also for their AI agents that9:17you would have access to if you were using the Rock platform.9:20AI has made it much simpler to build.9:22I think a lot of us, I think I heard somebody who shared earlier where somebody9:25was like, hey, who in here has built a website with Claude and I think every9:28cold room raised their hands?9:30It's made it easier to build, including software that touches your people data9:35right? want to challenge that.9:38I think if we don't really take the time to understand the different ways that9:42things can be built, we might run into a problem of security or trust down the9:47road. So here's like three different ways that we can think about developing9:51something. I'm gonna keep this really high level, this is not like a technical9:54talk.9:55So the three ways that we see people building software with AI right now would9:58be Vibe Coding. You may have heard that term.10:00It means that you describe what you want, you see what comes back and you're10:03matching it until you get that final result.10:05It feels like magic.10:07And then you've got something like Spec Coding.10:09It means that you write down exactly what you want first in a plan called a spec10:12or something similar.10:14You check everything that AI builds against that plan.10:18You give it a source of truth or a record to verify.10:21And then you've got artisanal coding, which I would say is the old way10:24where people kind of write line by line with their hand.10:27And some people might miss those In days.10:28We might, be nostalgic if you're a more technical person in the room.10:32I would just encourage you, you know, as we think about how do we resource our10:35team, you know, I can bi-code little helpers for myself10:38for things that only I use, but with software such as member for member records10:43or church data10:44I don't think that live coding sets a church up for success.10:50You want to understand what's going into it.10:53So Rock uses AI to build, especially nowadays, to accomplish more.10:57Triumph, our team uses it too.10:59And the written plan is what makes the difference.11:02John has talked about this on a podcast that the Rock community does called11:06Rockcast. I know we're real original.11:08The Rock team spends about 95% of their time writing detailed specs before AI11:13writes in the code.11:14The AI breaks the spec in the phases and builds one phase at a time.11:18And when they need a new feature for their AI gateway11:22John had actually spent a whole day using AI to kind of guide this spec to be11:28developed, it actually turned out to be like a 30-page11:30For workday, honestly it would have taken them about a month to code by hand.11:35He says that those specs were blueprints, and because the AI had access to them11:39It was able to bring the speed, but and the team were able to bring the wisdom.11:44You know, AI is very intelligent11:47but I would not equate that intelligence with wisdom.11:50You know, that's how humans, we've been designed to bring that to these11:55different tools.11:56And you can see that here in the agents that have been developed in Rock.12:00So every tool has an agent that could be used and defined ahead of time.12:06Every new agent starts locked down so nothing runs outside of your permissions12:11and at Triumph we work the same way with clear specs and personal reviews before12:14anything gets delivered in the church.12:16So Rock wrote a spec for their AI gateway and by the end of this time I hope12:21that we can actually work on spec together for your church.12:23Alright so here's what I think Rock got right.12:28We're going to look at some of the specs up to close a personal so i'm going to12:33walk through six choices that Rock made.12:35And for each one, I'm going to...12:37To talk about, as a leader, this is how you should you know take this kind of12:42take this next step.12:42As an operator, or the person who's doing the implementing12:45this is what this means for you.12:46So if you're just on another system, I don't want you to feel left out.12:51So six things down.12:52Because I think it would be good vendor questions that you can start.12:56And honestly, us too.12:57Pepper us down at the vendor hall with questions.13:02We'd love to engage in that conversation.13:04So AI agents of VOP kind of arrived in version 19 .413:08one of the latest versions of Rock.13:10And so if you're interested in what Rock does, I would encourage you to mark on13:15your calendars tomorrow 10 .45.13:17There's a session called Rock-RMS, the Innovation Toolkit for Churches13:20and that's gonna give you more information about Rock as a whole.13:23One of the interesting things this summer for the AI Summit was 670 churches had13:27registered for this event.13:28And I think, what I think is cool is plenty of churches want to get this right.13:33They're being tasked with the responsibility of their databases13:36and my guess is that this is how many people were kind of quietly worried that13:40we're falling behind13:41and they wanted to see what safe approach looks like from an AI and church13:45management database.13:46Choice number one13:48is where I'd start if a senior pastor asks you, what does it mean to put AI in13:53your church management system safely?13:55The first thing is the AI runs as you.13:58It's not as some other separate entity that you've given every possible power14:03and permission to.14:04Every time14:06our AI agents will look something up, Rock will check the permission of whoever14:10is logged in at the time, whether that's using the chat built into Rock or14:14connecting via MCP through CLOD or ChatGPT.14:19The AI can't see anything that the person couldn't already see.14:22On screen, a staff member names somebody like Alicia.14:26Let's just take Alicia for example.14:27She asked for a summary of Ted Decker and she would only get back exactly what14:31she'd be able to see if she was logged in on her own account.14:34For leaders, for your executive leaders, this means that AI never gets more14:38access than the person using it, that's the principle.14:41For operators, if your security roles are messy, if the way that you handed out14:45permissions are messy, then one of your first AI projects might be a permissions14:51cleanup that you've been putting off.14:53Next, choice number two.14:54Thank you. is how outside tools will actually connect with Rock.14:59Some vendors will tell you to paste an API key, which is basically a password15:02for your software, into a chat tool.15:04And once you do that, the key lives somewhere that you don't control with every15:09permission that you gave it, and almost nobody can change it unless you rotate15:13that key on a regular basis.15:15Rock uses MCP instead.15:17MCP is a standard called statistical model context protocol.15:21It lets tools like Claude or ChatGPT connect to Rock and with it you sign in the15:26way that you always do.15:27That's how it knows the permissions that you have access to.15:29The tool gets a short-lived path, the token, that keeps renewing and AI itself15:34will never see the password.15:36So John and another developer, Daniel Hazelmaker on the core team15:40they walked through this on the Rock-CAD which I have queued up there.15:44For leaders, as they're asking, as you're helping them navigate this15:48you can help them as you want, as you implement AI in your church management15:54system. You want it to sign in the way that you would to your system.15:59You should choose MCP whenever you can.16:02Rock does have an option to put a personal key in the connection link for tools16:06that can't handle the sign in screen.16:09If you turn that on, I would just treat those like passwords that you want to16:12rotate or keep safe on a regular basis.16:15Choice number three, every action has its own tool.16:20And this is what makes policies enforceable.16:22So adding a group and deleting a group, think of those as like two separate16:26switches, right, and so Rock, so it would be like drafting an email versus16:33sending it. So Rock lets you lock down a whole agent16:36lets you lock down a set of tools or skills, and then itavi locks lets you lock16:42down a single tool or action that the agent has access to16:48and it still respects every security on the record so for leaders this means16:52that the rules in your policy live in the church management system or rock16:57itself nobody has to remember something that they committed to from the staff17:01meeting so if you train your staff on17:04hey as a team we will never allow AI to send an email well that sounds sometimes17:09the eyes go grow. It doesn't matter many instructions that we give it.17:12But if we prevent that tool from ever being accessed by a certain team based on17:17permission, that's why we can truly lock it down.17:21For operators, I'd recommend turning on this graph data for your whole staff and17:26then leave sending it or deleting something off.17:29You know, don't give people that power.17:31That way an agent can write an email, but only a person can send it.17:35They have to have some sort of review in place.17:36And on the screen you can see every action, you know17:40within, you know, an MCP tool.17:42It has its own all or off switch for the different tools.17:45I think that's what you can show to an executive pastor or a leader.17:49Choice number four17:52lock it down. I have seen or heard some stories about MCP servers or different17:58AI agents being switched on for different products18:00and everybody has access.18:02It doesn't matter if you still even attend the church18:04you might have access.18:05A new agent in Rock is only available to Rock administrators until someone on18:10your team grants access to the other staff.18:12The AI settings for different tools, you can start to lock them down even more18:17granular. As an example, there's another feature in Rock called prayer18:21modulation. You can use AI to help categorize and classify prayer requests18:28and you might want to only allow certain categories to take certain actions with18:34AI. For example, if you identify somebody with self-harm tendencies within the18:39prayer request, you probably want to categorize that immediately and launch some18:43sort of follow-up to a local campus pastor so that they can address the issue18:47and take care of the individual18:50being able to connect that with AI in the right way with the right permissions18:54and set those granularly on the right categories gives you the flexibility that18:59I think you'll need in this next age for leaders.19:02Nobody should get access until somebody decides that they should.19:05Right? Feels like a pretty easy statement that you can make to your executive19:09leader. For operators, open access one role at a time.19:14And ideally this is paired with some sort of rollout plan.19:17But whether it's like turning on a prayer feature one category at a time or19:20you know, allowing certain tools within the staff's tool belt to be turned on19:26just be really intentional about not getting everybody access19:29but being intentional about the rollout.19:31Choice five is false.19:33This is like the second biggest topic at the AI Summit.19:37So, Rock IQ brought more stands for Rock Intelligence.19:40This is a service that powers AI inside Rock.19:43You can set one spending limit for your whole church.19:46You can pick a monthly amount that works kind of like a prepaid gift card.19:50The AI pauses if you run through it and every conversation reports how much it19:55used. For a leader, you can set a ceiling and see where the money goes.19:59For operators, you can build a report on that usage.20:03And remember, it only covers the chat features with inside of Rock in this case20:08or is built into your church management system.20:12Your system allows for an MCP, like Rock does, that's a whole separate bill20:16which gives you even more granular20:18permissions depending on the type of platform20:20and plan that you have.20:22Alright, so Rock IQ has two plans.20:25Standard uses a solid, low-cost model from one of the major AI companies.20:30And that's where we typically recommend somebody starts if they turn on in the20:33future I like this.20:35There's another standard that's going to be coming down the road called Econ and20:40that's going to use a cheaper open-source model that's slower and weaker at20:43using the tools20:44and using the tools is pretty much the agent's main job.20:48So you probably want to go with the standard to start.20:51However, there are going to be some churches who want that extra cost savings20:54Rock wants to give them the flexibility of managing that.20:57When one of our developers had taken a look at this20:59this is Daniel Casey-Maker21:01one of the people who builds.21:02Rock's AI features21:03when he was looking at the AI bill21:06he actually was just surprised, wow it's really not that much when we look at21:09the green screen.21:10So before you set an AI budget21:13maybe run that standard model or whatever it is that means for your church21:16management system for about a month and start to capture a fetch mark so that21:21you can start to evaluate that over time.21:23Most of the time it's a lot smaller than what people expect.21:26Choice number six is where your data lives.21:29So, conversations in the chat built into Rock are saved to your Rock database so21:34that you can actually look them up.21:35You can audit them and you can see how your chat is using the tools.21:38I promise you it will be enlightening.21:40If someone connects with Claude or ChatGPT.21:44Instead, that history lives in that tool.21:46So it's just another reason why your approved tools list matters21:50especially if churches or staff are trying to use their own tools21:54I would really discourage that.21:56Spark Development Network, the team behind Rock, they actually say that even22:00with the Rock IQ service, they can't actually see the data that's being sent22:04across, they're actually just monitoring the billing.22:07What they really wanted to do was create a safe way for churches who are not22:11technical to be able to connect to AI, while giving people the flexibility of22:15power users to be able to do more down the road.22:18For leaders, when the board asks, where does your data22:24go? Because that will happen.22:27You can help them to understand that there's safeguards in place and it's really22:32between you and the chat service that you're working with.22:37For operators, those conversations are a good to actually go through and22:41reevaluate with your staff to see how people are using AI and I think that's22:45gonna help to shape the next policy update and can allow you to help to define22:50better training for your team, For example, if people are constantly looking for22:53certain types of documentation, and you have that documentation in your system22:56but it's not clear, maybe you can help to fill that gap.23:00So, AI models Rock uses, they never learn from your church's data.23:05This is kind of like a really big point that we wanted to protect churches from.23:08Our core team has been upfront that some of the best models will still hold on23:13to some conversations first off time for safety checks.23:17So we encourage you to read the term, but they eventually let them go.23:20Most of the models, the way that the Rock IQ service is set up23:23they've got zero data retention turned on and do not use the data for training23:27and stuff like that to try and protect the church.23:30So most chats are going to hit the AI, it's going to respond23:34and then it's going to lose memory of the conversation.23:36What they can't protect against is a staff member who stops reading what the AI23:41writes, or what it's doing, and just hits send.23:45And so, we wanna protect against that.23:47We wanna protect against your people.23:49So,. Rock wrote a spec for its software.23:52Let's actually write a spec for your people, for your church23:55and how people can actually use it.23:58One of the best places or best ways that I've found in addressing this idea is24:03finding a place. of where you need to start so maybe you're nervous about AI and24:08your staff is and you're trying to figure out where everybody's at that's a24:11normal place to start where most of the churches that we work with right now are24:15trying to figure out so we built a free AI.24:17Fluency rubric to help.24:20It describes five levels from, like, I've tried it a couple of times24:23to a marker across my entire day, kind of everything onto it.24:27And each person can kind of find where they land.24:30And you can actually take those answers, put them together24:32and it gives you a picture of your whole team24:36plus a plan for helping people take that next step.24:39So you'll know who's ready for more of those agentic workflows down the road.24:45Let's talk about card rails.24:47So, we built these lanes on our team because our team kept asking the same24:52question, are we allowed to it24:55and the policy only works, as good as people24:59if people remember without looking it up, they don't need to constantly go to25:02some sort of a document.25:04So if they go to the document freeze, or guess, sorry about that.25:10And so we decided that we give them three easy colors to remember.25:15So whether or not they should progress with allowing AI to ask.25:20There's five factors that will decide easily so I'll show you what these colors25:24are in a second you probably guess But I would encourage you25:28before you put AI tasks into a specific way and I would ask these five questions25:32what is this data the touch or what is AI gonna touch what happens if you get it25:37wrong, or if the data is wrong can anyone do it25:40or whether it's reversible whose name is unresolved so she's owning it Can25:45and can the person check if it's actually, they can actually spot a mistake?25:51Do they have the domain expertise or are they able to actually review these25:54things? If your communication director signs off of an AI-built giving report25:58that they can't read, and nobody really checked it26:01that's obviously gonna be a problem.26:03So any one of these answers can kind of like take us further down the lane of26:06deciding whether or not we want to go, hold, or stop.26:11So green, yellow, red makes it really easy.26:14Everybody understands what these colors are, so we don't need to spend a whole26:18lot of this other than what defines each26:23So green, this is going to be improved data that you can easily undo.26:27So if you hand the AI a task and it makes a bunch of changes that you can still26:31pull back those changes and it's with approved data26:34then we would consider this probably a green choice.26:37So, this is where people can get to work, drafting26:39summarizing, reformatting.26:40A first pass is something that always gets reviewed.26:43That's kind of our goal.26:45But a green example in Rock is is asking an agent to set last month's connection26:49requests so that you can review them.26:52Connection requests are a way that people can stop from falling into cracks26:56within your church.26:58So green means like two things, an owner and a way to catch mistakes.27:04You can automate the review once you trust it, but a person is always27:07responsible for the result.27:09Now we've got yellow, all right?27:11This is one of those things where, like, as you go through these lanes27:15you will find yourself and you'll find your staff constantly questioning and27:18bargaining with themselves, oh, I think this is actually a green27:20or I think this is a yellow, or I think this is not a red.27:23But this is where I think of the interesting work lives.27:27I think most of us are going to spend our AI time early on.27:30So use approved tools only.27:33Let's say we try it on a copy of our church database.27:35So we copied our data.27:37And we want somebody to go through and update a field27:42like a background check date on like 50 records.27:46It's a test copy, right?27:48So the yellow work, or like this lane, can still move forward.27:52We're using it in a safe environment, we're not updating it27:55we're making sure it works before we execute it in production.28:00If you happen to be arguinging yourself whether something is green or yellow28:04I would just default to it's yellow.28:05Don't try to argue with yourself.28:07Stay, as you're building out your objective workflow28:10if you stay in this yellow lane a couple extra weeks just to make sure that28:13everything is working appropriately.28:14I think that you're gonna be feeling a lot better.28:18And if you move to green too28:20early, it can cost you a member's trust.28:23All right, everybody knows what this means.28:25Red means what? Stop.28:27The list is short on purpose.28:29Data that should never go into AI, anything that reaches a member before a28:33person needs it, and anything that you can't undo.28:35This is all red. We should not do this.28:37We should not hand this over to AI.28:39An agent texting first-time guests without anyone reviewing the messages.28:43That would be free.28:45No, I'm just kidding.28:46Red.28:47We'll keep people bargaining with red, too.28:49We'll just spot check a few.28:50It's only for one event.28:52They sound reasonable28:54but they're gonna put unreviewed message in front of a guest and that's gonna28:57cost you trust. So how do we manage this?29:01Well I would encourage you to write it down and get your leadership to read it29:04in a policy. A policy written right after a scare usually comes out too strict29:10too structured, and you'll spend you know a lot of time actually trying to29:14loosen it after the fact.29:16Our free AI policy template that we've just given away29:20this actually has the lanes and kind of like these different movements built29:25into the policy for you.29:27We keep it really simple and it's less about do not do this and it's more about29:30how do we think about using this as another tool.29:33So before we get keep going, here's where we're at.29:36You know how to find out where your team is with the fluency rubric.29:41You know how to sort the different AI agentic workflows into specific lanes.29:46And then we're going to add a line to this card after each section.29:49But by the end, you'll have a spec for your church.29:51That's the goal.29:52So that one-page policy that we were talking about29:55this is the part that I think you can take home and use this month.29:59We're at the beginning of October.30:00How great is that?30:01How crazy is that?30:02We're already in October.30:03But that means we can make progress before the end of the year.30:07So let's be action oriented.30:08Let's take this policy and let's work through it with our leadership.30:12And I think that if you present it in a way like this30:16I think it's going to be easy for the leadership to digest and that you can30:19influence to make this decision.30:22Sometimes people are worried about all slowing them down many of us feel like30:27churches wrote their first social media policies.30:30What actually slows people down is just not knowing what's allowed or not30:33like not having those guardrails30:35because then every question comes to you and you become the bottleneck.30:39If you provide bigger lines to your staff30:42it'll allow them to kind of say yes on their own, and for most churches30:45this will be the first AI guidance that anyone has ever written down.30:49Hopefully that's not the case, but we have seen that in the case.30:52So the policy is kind of like four moves.30:54These first three are going to cover how the work gets done30:57and then we've got this hard line we'll work in at the end.31:00So the first one, nothing ships unchecked.31:04That means nothing is sent out without some sort of review.31:12So one specific person has to read it.31:14They have to check it.31:15They have to own it.31:17Sometimes people treat AI like they're a traitor, and I would just say31:22hey, it's actually, use it to have your report, you just have to be the one to31:26put your name on it.31:28So make sure it was reviewed31:30you don't want to send it out.31:32Here's what this could look like in a day-to-day example31:36draft, don't send.31:37And so an agent, this is an example within like a voice agent within about31:41drafting a communication.31:43So it can draft a follow-up email, a reminder, a thank you note31:46and then even one for every person on your outreach list31:50just don't allow it to click send.31:52The day the agent sets its own31:56you know, kind of like message without approval31:59is that a member gets a message that wasn't clearly written by someone who knows32:02them and that trust is hard with the back so on screen someone is using the32:06rocks voice agent which is another and to write a note to a man whose mom is32:11sick and writes a draft then goes32:14you know to the person to check before anything actually goes out so the choice32:20that we're really addressing and I'm really trying to drill in here is like we32:23really can't send out anything that has not been reviewed.32:26Another thing as we talked about the policy is to allow AI to sharpen but not32:30outsource. Allow to poke holes in your ideas, in your strategy32:35in your thinking. argue with the outline, find one of this.32:39For example, ask what a what would a skeptical elder ward member push back on32:44this plan to drill in here, or for another thing.32:47Once it's produced, the idea is that you haven't been able to stop becoming the32:52author and become a forwarding address, so people can usually tell.32:57We don't want that.32:58Next, in our policy, we've got approved tools for members33:01so we're not doing free chat to key plans, free cloud accounts.33:04We want to kind of bring anything under one33:08organizational plan.33:10So it comes down to kind of like two rules.33:12So for general work, like brainstorming, a title for a sermon series33:17you know, you can use what's helpful to you.33:19For anything with member data, like a list of first-time guests or giving33:23history33:25use only the tools on your approved data list.33:28Because those are the ones that your church has agreements with33:31you know, with those members.33:33With those first-time guests.33:34If you don't publish an approved list, your church will build out on whatever is33:38free or whatever they have access to, and you'll be left trying to clean up the33:42list later.33:44Anybody who's moved and migrated systems will understand what that's like too.33:48Alright, so here's the one absolute in the policy.33:51Everything's kind of old up until this point, But here's what we recommend and33:56strongly urge for people, is to not include these types of data within your AI34:02prompts. So government IDs, passwords34:06health and care34:07details, anything that a member has shared with you34:10like confidential pastoral care type of information.34:13Someday someone will want to paste a member diagnosis into a chat window to help34:18write a care government.34:19And I would encourage you to remind them that that's not allowed.34:23It's really just like a guardrail to protect yourself and your team.34:28This policy covers church data and church-based tools.34:31What it doesn't cover is when your Worship Pastor pulls up that G .T .T at home34:37and if you try to reach that you're just going to use your staff's trust so make34:40sure that you are giving them access to the tools that they to do their campaign34:45for them, but setting the proper boundaries in place.34:48AI is changing so fast that I think if you're not reviewing a policy on a34:54regular basis, your tools are going to be out of date34:57so will your policy in the end of months.34:59Alright, so that's your policy.35:01So writing it can be the easy part, making sure that a real person knows the35:04result once an agent does the work takes a little bit more thought35:07so let's talk about that.35:09Keeping Accountability.35:11This section is for anyone who gets the phone call at 11 at night when something35:16breaks, which I think we're at a good conference for.35:19That's typically the type of audience where people are getting handed those35:23types of things. So handing work to an agent is easy.35:26Keeping your name on the outcome when you do it I think that's the part that you35:30need to transition for your staff and I think doing that goes back to spec35:35coding but just a the size of the task, not like a full application.35:39So what is it that you want the workflow to do?35:43I think it's a habit that I tend to lean on when I'm gonna work for other35:46churches as well. Anytime that you could let let AI do something simple35:52you swap something with unpredictable.35:55AI agents, they're not deterministic is a word that we use.35:59They're unpredictable sometimes you can kind of guide them and give them36:03instructions to get very specific results But it doesn't mean that you will36:06ultimately get the exact same result every single time.36:09So when it comes to your AI36:13make sure that you're giving it narrow focus, specific tools36:19to be able to execute workloads provided structure around.36:23And I've got a resource for for for this36:28but when it comes to how do you navigate from allowing36:35AI to complete these different tasks for you, and being able to step back and36:41kind of, like, Re-evaluate the judgment piece.36:43Here's how I encourage you to handle an agent.36:47Tell an agent to handle uncertainty.36:51And honestly, it's not even just an agent thing, like an AI agent thing.36:54If I had a new hire, I'd be telling them this exact same advice.36:58If a choice is cheap and easy to undo, I would just say make your best guess.37:04You know, try to follow what you assume.37:07But if it's costly and undoable, and you can talk to your AI agents about this37:11you can set up these guardrails.37:13Show me the options before you pick one.37:15So if I've got a teammate who is competing on my behalf37:19ultimately I have extreme ownership over what they produce.37:22And if they're nervous about something that has bigger repercussions37:28I want them to bring me that decision for me to be able to help make it.37:32And that's what we want to do with our agents37:35right? Daniel Hazelbaker, he has this, he said this thing during one of the37:40broadcast points, and he says the bigger concern is the people behind37:46the wheel. Are they outsourcing their work to it, or are they actually using it37:50as a helpmate or in a system?37:53So you might run into the situation where people are hesitant to use AI for37:57their work.37:59You might just find people who are using it really well and treating it like a38:03true partner, and then you might have people who have just delegated their38:06entire job38:07to AI. And I think that that's somebody who you need to address38:13with some training and coaching, because ultimately they're putting your38:16organization at risk.38:18Work with these different churches on a regular basis38:21and they're trying to figure out how to use AI.38:24And honestly they create margins so they can put their resources to other38:28projects. And so a church partner letting ministry build its own data using rock38:34data views are features like seeing a list of entities over certain list of38:38people with certain criteria.38:39That's like the old and these are like safely right38:45but what ended up happening over time is they these ministries have to38:52find you know what a certain definition was like for them who is active38:58what does a like a small group attendee look like39:02what is a39:05a consistent giver you know in their39:08And over time, they didn't intend for this to happen39:11but over time, nobody was really owning the process.39:14And it became a question where everybody began to have their own definitions for39:19things. In fact, this is even more...39:23True now with ai where we saw that same church partner end up building a39:27dashboard for themselves.39:28It looked really pretty.39:30It39:31honestly looked great.39:32It was servicing upright information.39:34But what we found out later was a lot of these key performance indicators that39:37they were relying on and pushing to the leadership were never fully defined by39:41their ministries. And so I would encourage you39:45as you're figuring out how to roll out AI, Make sure that you agree with your39:48definitions or ministry dictionaries39:52with your teams or with your leadership so that you can agree on one definition39:56for things like active, engaged39:59first time giver and then write them down where everybody can see them and being40:03able to have that easily accessible I think will serve your teams well.40:08All right, so, if someone challenged40:13you know, this tomorrow, could you explain what you approved?40:17This being the agentic workflow that you put into place40:20why you approved it, what you verified, and what would you do if it was proved40:26wrong? You know, if you could ask yourself these questions.40:29I think it would be really eye opening to whether or not we should be handing40:33this work off to different agents.40:35So that comes back to approval, making sure that you have one owner on that40:39on And so if you can answer those four questions, I think you've got a really40:43good start to an agentic workflow, and you're ready for your first pilot.40:47Everything up until this point is about how you think about AI.40:50So we actually have an article that will grow.40:53If you're wondering how do we think about putting these steps together for40:56agentic workflows40:58we've got an article around seven steps41:01that you can use to define your own agentic workflow.41:05And one of the questions I love from the Rockcast recently was about vacation.41:10You know41:11if you ever want to go on vacation, you need to make sure that you provide your41:15team with the right resources to cover you while you're gone41:18or else you're going to be on call.41:21So, whether that's a defined person on your team to handle this for you41:26or whether that's the different skills and tools, like I talked about Jesse41:29earlier, who was able to handle your team, I think you need to find the lowest41:35hanging fruit of the bottlenecks that are currently on your plate start to work41:39on offload then you need to make sure that your I'm not saying to41:46offload big parts of your job or things that require a lot of attention to41:50detail.41:51You can even keep it pretty simple in terms of not making changes to different41:55data, but really focus on surfacing up data for you41:59whether it's a daily brief or who needs your attention within your church42:02management system.42:04But don't let it send a message or change a record within your church management42:08system. Really try to surface information, figure out what's missing42:12and then you can start to move over.42:15So track these types of things for two weeks.42:18Keep a shared. And write down these different types of problems that you kind of42:22notice within your ministry or in your area.42:25What's missing in context?42:26That's kind of the most common problem.42:29When you're building these different agency portfolios.42:32It's missing the right database, missing a training or a resource that you put42:37together in an SOP or a standard operating procedure.42:40A summary can look complete and sound42:43Because the AI can do that sometimes.42:45But it might be wrong.42:47It might not have the data it needs to make those types of things.42:50And so they can produce those wrong conclusions.42:52Maybe the facts were right, but the reasoning went sideways.42:56Like finding a family that disconnected because they're from a 10:45 service to43:01a nine a service and you're tracking people based on their preferred service43:04times.43:05Permission gaps are another thing that you can because AI can do that43:11sometimes. So if you implement an agentic workflow and you find yourself running43:15into different barriers, you might need to re-evaluate your permission gaps43:19whether that be your church management system or a different tool.43:22And I would encourage you to do this for the next two weeks and try to build an43:26agentic workflow for your team that you can hand off to somebody else.43:29So the next question, how do you know it's ready to grow?43:31If a ministry leader who owns this work can explain what it means43:35what has changed? whether that how to turn it off without IT translating for43:40them, I think it's right.43:41If they can't, I will just keep it in the pilot a little bit longer.43:44I find myself keeping my own more close in pilot a lot longer than I want43:48because I really want to make sure it's good because I want to hand it off to a43:50team member, and once you hand it off to a team member43:53they're going to run with it.43:53It's going to be a lot harder to update that process for your team.43:58Alright, so you have this idea of a pilot.44:01You know, if you're on Rock, you have what you need to run it.44:04If you're on another church management system, you might need to evaluate with44:07your vendor what that looks like.44:09So here's six questions, I talked about this at the end44:12that you can ask your church management system vendor.44:17You should probably know what a good answer sounds like after we've covered all44:21this ground in the presentation, or based on your experiences of things not44:24going well. within your own environment.44:27I can't speak to how other systems have built their AI in patients44:31but I think that ROC has done it really well.44:33And if they can't seem to answer the question in a way that seems suitable to44:36you, Don't rush into a pause and ask more questions.44:40I found that to be very, very helpful.44:42So here's the whole spec for your church44:46so Rock wrote a spec for their AI, and this is where you can actually start44:50yours, and I would encourage you to do that, and try to do that this month.44:55I mentioned earlier about that article about building an agentic workflow44:59we included that for you here.45:01The last slide will45:07cover all45:09So here's the whole you grab that one I know I'm faster so a moment saw some45:13different things pop up So I'll put them all in one slide for having me.45:16And then I can answer any questions.45:18Every resource for today is on this slide, but I'm also around today and45:23tomorrow,OCK, so I would love to talk with you too.45:25And then answer any questions that you might have.45:29Just a reminder that we do have another AI Agenda cohort.45:3275 organizations have gone through it so far45:35starting on November 17th, every other Tuesday45:39in the afternoon, would love to have you.45:42It's been a lot of fun working through a lot of those cohorts45:45and we've all learned a lot.45:46And they keep changing.45:47It's a lot of fun.45:48Another reminder, make sure you hit up Rock RMS.45:51The Innovation Toolkit for Churches, tomorrow.45:53If you're not familiar with Rock and you want to know more45:5510:45.45:57Well, I really appreciate you guys being here.46:00It's been fun. If you guys need to find me, I'll be at the Triumph booth most of46:04the time. The rest of the conference.46:07And I really would like to get to know you and just come say hi.46:11You know, just hear about your journey, and your church.46:14I can give you some time back.46:15Thanks again for your attention.46:17I really appreciate it.
