1 00:00:00,040 --> 00:00:03,600 Welcome to the M365 FM podcast, where we explore the people, 2 00:00:03,600 --> 00:00:06,120 ideas and technologies shaping the future of work. 3 00:00:06,120 --> 00:00:07,759 What a trigger actually is. 4 00:00:07,759 --> 00:00:09,240 Most people picture an AI agent 5 00:00:09,240 --> 00:00:10,320 as something you talk to a 6 00:00:10,320 --> 00:00:11,599 chatbot that sits there until 7 00:00:11,599 --> 00:00:12,320 you type. 8 00:00:12,320 --> 00:00:15,160 Triggers are the part that lets it move first. 9 00:00:15,160 --> 00:00:16,519 They turn a polite question and 10 00:00:16,519 --> 00:00:17,640 answer bot into something that 11 00:00:17,640 --> 00:00:18,039 reacts. 12 00:00:18,039 --> 00:00:21,160 When a file lands in SharePoint, a task gets assigned or a new 13 00:00:21,160 --> 00:00:22,440 person joins your team. 14 00:00:22,440 --> 00:00:23,800 In this episode, we're breaking 15 00:00:23,800 --> 00:00:25,519 down copilot studio triggers in 16 00:00:25,519 --> 00:00:26,199 plain language. 17 00:00:26,199 --> 00:00:28,199 No jargon, no admin details. 18 00:00:28,199 --> 00:00:29,800 Just what they are, how they 19 00:00:29,800 --> 00:00:31,179 work, and where they can cost 20 00:00:31,179 --> 00:00:32,159 you money. 21 00:00:32,159 --> 00:00:35,799 Welcome to the M365 FM podcast where we explore the people, 22 00:00:35,799 --> 00:00:38,200 ideas and technologies shaping the future of work. 23 00:00:38,200 --> 00:00:39,719 So here is the plain version. 24 00:00:39,719 --> 00:00:42,200 A trigger is the condition that decides when your agent should 25 00:00:42,200 --> 00:00:43,280 do something at all. 26 00:00:43,280 --> 00:00:44,479 The clever conversation and the 27 00:00:44,479 --> 00:00:45,560 final answer are not the 28 00:00:45,560 --> 00:00:46,240 trigger. 29 00:00:46,240 --> 00:00:48,640 It's the small rule that says this is your moment. 30 00:00:48,640 --> 00:00:51,560 An image helps here, and it's the one I keep coming back to. 31 00:00:51,560 --> 00:00:55,439 Think of a topic inside Copilot Studio as a room, and inside 32 00:00:55,439 --> 00:00:58,000 that room sits everything the agent actually does. 33 00:00:58,000 --> 00:01:01,359 It asks the questions, sends the messages, runs the flow, and 34 00:01:01,359 --> 00:01:03,079 updates the record in Dataverse. 35 00:01:03,079 --> 00:01:05,599 The trigger is the doorway into that room, and it decides 36 00:01:05,599 --> 00:01:07,879 whether the person standing outside gets to walk in. 37 00:01:07,879 --> 00:01:10,560 This matters more than it sounds when your triggers are vague. 38 00:01:10,560 --> 00:01:11,920 One of three things happens. 39 00:01:11,920 --> 00:01:14,079 No topic starts at all and your agent looks lost. 40 00:01:14,079 --> 00:01:15,239 The wrong topic starts and it 41 00:01:15,239 --> 00:01:16,879 answers a question nobody asked, 42 00:01:16,879 --> 00:01:18,480 or the user gets a flat generic 43 00:01:18,480 --> 00:01:19,719 fallback reply that tells them 44 00:01:19,719 --> 00:01:20,680 nothing useful. 45 00:01:20,680 --> 00:01:22,239 Microsoft's own documentation 46 00:01:22,239 --> 00:01:23,879 describes triggers as the way an 47 00:01:23,879 --> 00:01:25,200 agent intercepts and handles 48 00:01:25,200 --> 00:01:26,620 different events or incoming 49 00:01:26,620 --> 00:01:27,519 activities. 50 00:01:27,519 --> 00:01:28,640 That's a long way of saying the 51 00:01:28,640 --> 00:01:29,879 trigger notices that something 52 00:01:29,879 --> 00:01:31,200 happened and then decides to act 53 00:01:31,200 --> 00:01:31,760 on it. 54 00:01:31,760 --> 00:01:33,000 A trigger is small, but it does 55 00:01:33,000 --> 00:01:34,359 a lot of quiet work behind the 56 00:01:34,359 --> 00:01:35,040 scenes. 57 00:01:35,040 --> 00:01:37,959 It's the difference between an agent that feels alive and one 58 00:01:37,959 --> 00:01:40,620 that just sits there waiting for a human to push it, and some 59 00:01:40,620 --> 00:01:43,640 triggers fire when nobody is typing at all, which is the part 60 00:01:43,640 --> 00:01:45,359 beginners never see coming. 61 00:01:45,359 --> 00:01:46,760 The two families of triggers. 62 00:01:46,760 --> 00:01:48,959 One you start and one that starts itself. 63 00:01:48,959 --> 00:01:52,040 Every trigger in Copilot studio belongs to one of two families. 64 00:01:52,040 --> 00:01:55,040 And once you see the split, the whole tool gets easier to hold 65 00:01:55,040 --> 00:01:55,719 in your head. 66 00:01:55,719 --> 00:01:57,920 The first family is the one you already understand. 67 00:01:57,920 --> 00:02:01,055 These are toxic triggers and they're user led, which means 68 00:02:01,055 --> 00:02:04,030 they start a topic or an action based on what a person says. 69 00:02:04,030 --> 00:02:06,609 Someone types a question and the trigger works out which topic 70 00:02:06,609 --> 00:02:07,629 should handle it. 71 00:02:07,629 --> 00:02:09,949 The second family is the one that surprises people. 72 00:02:09,949 --> 00:02:12,629 Event triggers are system led and they start agent behavior. 73 00:02:12,629 --> 00:02:13,909 When something happens outside 74 00:02:13,909 --> 00:02:15,389 the chat, nobody types anything 75 00:02:15,389 --> 00:02:16,030 at all. 76 00:02:16,030 --> 00:02:17,750 A file shows up in OneDrive. 77 00:02:17,750 --> 00:02:20,710 A task gets marked done in planner, a new row appears in a 78 00:02:20,710 --> 00:02:23,430 table, or a schedule ticks past a certain time. 79 00:02:23,430 --> 00:02:26,150 The event itself wakes the agent up with no person involved. 80 00:02:26,150 --> 00:02:27,669 So the clean way to hold it is 81 00:02:27,669 --> 00:02:29,110 this one family is someone asked 82 00:02:29,110 --> 00:02:30,030 and the other family is 83 00:02:30,030 --> 00:02:31,150 something happened. 84 00:02:31,150 --> 00:02:33,189 If a person is the reason your agent runs. 85 00:02:33,189 --> 00:02:34,389 You want a topic trigger. 86 00:02:34,389 --> 00:02:37,189 If a system event is the reason you want an event trigger. 87 00:02:37,189 --> 00:02:40,189 That distinction sounds simple and it is, but it changes how 88 00:02:40,189 --> 00:02:41,949 you design everything. 89 00:02:41,949 --> 00:02:43,110 With a topic trigger, you're 90 00:02:43,110 --> 00:02:44,590 mostly thinking about words, the 91 00:02:44,590 --> 00:02:45,870 phrasing people use, and how to 92 00:02:45,870 --> 00:02:46,870 recognize it. 93 00:02:46,870 --> 00:02:47,990 With an event trigger, you're 94 00:02:47,990 --> 00:02:49,469 thinking about systems instead, 95 00:02:49,469 --> 00:02:51,069 what fires the event and what 96 00:02:51,069 --> 00:02:52,270 the agent should do the moment 97 00:02:52,270 --> 00:02:52,710 it lands. 98 00:02:52,710 --> 00:02:54,590 There's a third tool worth naming here, too. 99 00:02:54,590 --> 00:02:57,669 When a connector is hard for the agent to call directly, or when 100 00:02:57,669 --> 00:03:00,629 you need strict step by step logic around an approval, an 101 00:03:00,629 --> 00:03:02,349 update, or a notification. 102 00:03:02,349 --> 00:03:05,990 Microsoft's guidance is to reach for a power automate flow. 103 00:03:05,990 --> 00:03:07,229 Think of it as the dependable 104 00:03:07,229 --> 00:03:08,750 back office behind the agent, 105 00:03:08,750 --> 00:03:09,990 while the trigger is the front 106 00:03:09,990 --> 00:03:10,750 door. 107 00:03:10,750 --> 00:03:12,550 The two families are not equal, though. 108 00:03:12,550 --> 00:03:13,710 A topic trigger always has a 109 00:03:13,710 --> 00:03:14,590 human in the loop because 110 00:03:14,590 --> 00:03:15,849 someone had to ask for something 111 00:03:15,849 --> 00:03:16,550 first. 112 00:03:16,550 --> 00:03:17,909 An event trigger can run on its 113 00:03:17,909 --> 00:03:19,189 own at three in the morning with 114 00:03:19,189 --> 00:03:19,990 nobody watching. 115 00:03:19,990 --> 00:03:21,030 That is where the power sits, 116 00:03:21,030 --> 00:03:22,030 and it's also where the risk 117 00:03:22,030 --> 00:03:22,550 lives. 118 00:03:22,550 --> 00:03:23,710 Hold on to that risk. 119 00:03:23,710 --> 00:03:24,789 There's a cost side to running 120 00:03:24,789 --> 00:03:26,270 on its own, and we'll get to it 121 00:03:26,270 --> 00:03:27,590 soon. 122 00:03:27,590 --> 00:03:29,550 How a topic trigger figures out what you mean. 123 00:03:29,550 --> 00:03:32,349 Let's take the user led family first, because this is the one 124 00:03:32,349 --> 00:03:33,349 beginners build first. 125 00:03:33,349 --> 00:03:34,569 Inside a topic, you write 126 00:03:34,569 --> 00:03:35,669 trigger phrases, which are 127 00:03:35,669 --> 00:03:37,849 example sentences that stand in 128 00:03:37,849 --> 00:03:39,469 for how people actually ask for 129 00:03:39,469 --> 00:03:39,909 something. 130 00:03:39,909 --> 00:03:41,870 Say you're building a leave request topic. 131 00:03:41,870 --> 00:03:44,330 Your phrases might be apply leave request, vacation book, 132 00:03:44,330 --> 00:03:46,550 time off, I need leave or can I take leave tomorrow? 133 00:03:46,550 --> 00:03:48,550 You're not writing a script the user has to follow. 134 00:03:48,550 --> 00:03:51,129 You're giving your agent a handful of realistic ways to 135 00:03:51,129 --> 00:03:52,509 recognize the same request. 136 00:03:52,509 --> 00:03:54,550 People get the wrong idea about this step. 137 00:03:54,550 --> 00:03:57,129 Trigger matching isn't keyword search, where the user has to 138 00:03:57,129 --> 00:03:58,949 type the exact phrase you saved. 139 00:03:58,949 --> 00:04:00,509 Copilot studio reads the intent 140 00:04:00,509 --> 00:04:01,990 behind the message so someone 141 00:04:01,990 --> 00:04:03,150 can type something you never 142 00:04:03,150 --> 00:04:03,590 wrote. 143 00:04:03,590 --> 00:04:04,949 And the right topic can still 144 00:04:04,949 --> 00:04:06,389 start as long as your examples 145 00:04:06,389 --> 00:04:07,349 captured the shape of the 146 00:04:07,349 --> 00:04:08,389 request. 147 00:04:08,389 --> 00:04:10,669 A good phrase list is short, natural, and varied. 148 00:04:10,669 --> 00:04:14,110 You want the words real people use, so mix, leave, vacation 149 00:04:14,110 --> 00:04:15,030 time off, and holiday. 150 00:04:15,030 --> 00:04:16,189 If those are common where you 151 00:04:16,189 --> 00:04:17,310 work, you don't need hundreds of 152 00:04:17,310 --> 00:04:18,629 phrases and near duplicates 153 00:04:18,629 --> 00:04:19,269 don't help. 154 00:04:19,269 --> 00:04:22,230 You need to cover the realistic ways people ask and then test 155 00:04:22,230 --> 00:04:25,350 them, including partial phrases, typos, and longer questions. 156 00:04:25,350 --> 00:04:26,990 If you're stuck for ideas, raid 157 00:04:26,990 --> 00:04:27,949 your support tickets or your 158 00:04:27,949 --> 00:04:28,430 chat logs. 159 00:04:28,430 --> 00:04:29,769 Real questions from real people 160 00:04:29,769 --> 00:04:31,029 are the best phrase list you'll 161 00:04:31,029 --> 00:04:32,550 ever find, and they'll tell you 162 00:04:32,550 --> 00:04:33,790 which words your organization 163 00:04:33,790 --> 00:04:34,790 actually uses. 164 00:04:34,790 --> 00:04:37,829 When two topics look equally good, Copilot studio has a built 165 00:04:37,829 --> 00:04:39,189 in way to handle it. 166 00:04:39,189 --> 00:04:42,149 It can show a multiple topics matched list, so the user just 167 00:04:42,149 --> 00:04:43,670 picks which one they meant. 168 00:04:43,670 --> 00:04:45,509 That's much friendlier than a wrong guess. 169 00:04:45,509 --> 00:04:46,750 And it also tells you that two 170 00:04:46,750 --> 00:04:47,670 of your topics might be 171 00:04:47,670 --> 00:04:48,589 overlapping. 172 00:04:48,589 --> 00:04:50,149 There's an order to things as well. 173 00:04:50,149 --> 00:04:51,910 Some trigger types fire before others. 174 00:04:51,910 --> 00:04:53,230 And if two triggers of the same 175 00:04:53,230 --> 00:04:55,189 type both qualify, the older one 176 00:04:55,189 --> 00:04:55,790 runs first. 177 00:04:55,790 --> 00:04:58,110 You can set a priority yourself to override that. 178 00:04:58,110 --> 00:05:01,045 You can also add conditions, so a trigger only fires when the 179 00:05:01,045 --> 00:05:02,060 situation is right. 180 00:05:02,060 --> 00:05:03,259 For example, only when the 181 00:05:03,259 --> 00:05:04,500 conversation is happening inside 182 00:05:04,500 --> 00:05:06,139 Microsoft Teams, when a plain 183 00:05:06,139 --> 00:05:07,620 condition isn't enough, there's 184 00:05:07,620 --> 00:05:08,779 a formula editor for anything 185 00:05:08,779 --> 00:05:09,779 more complicated. 186 00:05:09,779 --> 00:05:12,660 One thing changed recently, and it's worth knowing before you 187 00:05:12,660 --> 00:05:14,100 follow an old tutorial. 188 00:05:14,100 --> 00:05:15,740 Newer agents use a mode called 189 00:05:15,740 --> 00:05:17,740 Generative Orchestration, and 190 00:05:17,740 --> 00:05:18,060 they're. 191 00:05:18,060 --> 00:05:19,980 The default trigger is the agent chooses. 192 00:05:19,980 --> 00:05:21,259 Instead of matching phrases, the 193 00:05:21,259 --> 00:05:22,660 agent looks at a topic's name 194 00:05:22,660 --> 00:05:24,019 and its description and decides 195 00:05:24,019 --> 00:05:25,259 whether it fits what the user 196 00:05:25,259 --> 00:05:25,819 asked. 197 00:05:25,819 --> 00:05:27,060 Older style agents use. 198 00:05:27,060 --> 00:05:28,779 User says a phrase which is the 199 00:05:28,779 --> 00:05:30,139 phrase matching version we just 200 00:05:30,139 --> 00:05:31,060 walked through. 201 00:05:31,060 --> 00:05:33,779 Both are still alive in the product, so if a tutorial looks 202 00:05:33,779 --> 00:05:37,019 different from your screen, that's usually the reason how an 203 00:05:37,019 --> 00:05:40,300 event trigger lets your agent act on its own. 204 00:05:40,300 --> 00:05:41,420 Now let's look at the family 205 00:05:41,420 --> 00:05:42,500 that runs without anyone 206 00:05:42,500 --> 00:05:43,180 watching. 207 00:05:43,180 --> 00:05:44,339 Before you can use an event 208 00:05:44,339 --> 00:05:45,740 trigger, your agent needs 209 00:05:45,740 --> 00:05:47,180 generative orchestration turned 210 00:05:47,180 --> 00:05:47,620 on. 211 00:05:47,620 --> 00:05:50,019 Orchestration is just the agent's ability to decide which 212 00:05:50,019 --> 00:05:51,379 tool or topic to use. 213 00:05:51,379 --> 00:05:52,660 And in plain terms, it's the 214 00:05:52,660 --> 00:05:54,420 part doing the choosing event 215 00:05:54,420 --> 00:05:54,740 triggers. 216 00:05:54,740 --> 00:05:56,939 Build on that because the agent has to figure out what to do 217 00:05:56,939 --> 00:05:58,740 with the event once it arrives. 218 00:05:58,740 --> 00:06:01,620 The flow has three steps, and they're easier than they sound. 219 00:06:01,620 --> 00:06:04,019 First, an event happens somewhere else, like a new item 220 00:06:04,019 --> 00:06:05,379 appearing in SharePoint. 221 00:06:05,379 --> 00:06:08,180 Second, the trigger sends what Microsoft calls a payload, which 222 00:06:08,180 --> 00:06:10,740 is a small bundle of information about the event, plus any 223 00:06:10,740 --> 00:06:12,500 instructions you want to attach. 224 00:06:12,500 --> 00:06:14,939 Third, the agent reads that bundle and picks an action or a 225 00:06:14,939 --> 00:06:15,819 topic to run. 226 00:06:15,819 --> 00:06:17,939 The library of events is bigger than you'd guess. 227 00:06:17,939 --> 00:06:19,139 A new item in SharePoint. 228 00:06:19,139 --> 00:06:22,279 A new file in OneDrive, a task completed in planner, or a set 229 00:06:22,279 --> 00:06:24,620 amount of time passing, which is the recurrence trigger. 230 00:06:24,620 --> 00:06:26,939 You pick the event and then you narrow it down. 231 00:06:26,939 --> 00:06:30,620 So for a table in Dataverse, you choose which tables changes 232 00:06:30,620 --> 00:06:31,860 should wake the agent. 233 00:06:31,860 --> 00:06:33,259 One thing trips up beginners here. 234 00:06:33,259 --> 00:06:34,980 The agent doesn't invent an action on the spot. 235 00:06:34,980 --> 00:06:36,100 You have to build the action or 236 00:06:36,100 --> 00:06:37,740 the topic first, and then your 237 00:06:37,740 --> 00:06:38,259 instructions. 238 00:06:38,259 --> 00:06:39,980 Tell the agent which one to reach for. 239 00:06:39,980 --> 00:06:41,300 If you expect the trigger to do 240 00:06:41,300 --> 00:06:42,660 the work by itself, you'll be 241 00:06:42,660 --> 00:06:43,579 disappointed. 242 00:06:43,579 --> 00:06:44,720 You also have two places to 243 00:06:44,720 --> 00:06:46,060 write instructions, and keeping 244 00:06:46,060 --> 00:06:47,060 them straight saves you a 245 00:06:47,060 --> 00:06:47,660 headache. 246 00:06:47,660 --> 00:06:48,620 There are the agent's general 247 00:06:48,620 --> 00:06:49,899 instructions which cover how it 248 00:06:49,899 --> 00:06:51,420 behaves overall, and there are 249 00:06:51,420 --> 00:06:52,620 the instructions inside each 250 00:06:52,620 --> 00:06:54,180 trigger payload which cover this 251 00:06:54,180 --> 00:06:55,019 one event. 252 00:06:55,019 --> 00:06:57,899 When those two disagree, you get errors or strange behavior. 253 00:06:57,899 --> 00:07:00,420 So the guidance is to keep them consistent. 254 00:07:00,420 --> 00:07:01,560 If an agent has several 255 00:07:01,560 --> 00:07:03,060 triggers, put the event specific 256 00:07:03,060 --> 00:07:04,459 detail in the payload instead of 257 00:07:04,459 --> 00:07:05,420 overloading the general 258 00:07:05,420 --> 00:07:06,180 instructions. 259 00:07:06,180 --> 00:07:07,379 There's a testing step here that 260 00:07:07,379 --> 00:07:08,699 beginners skip, and it's worth 261 00:07:08,699 --> 00:07:10,139 slowing down for before you 262 00:07:10,139 --> 00:07:10,579 publish. 263 00:07:10,579 --> 00:07:13,259 The agent won't react to the trigger automatically, so you 264 00:07:13,259 --> 00:07:16,060 run the event once to create a payload, then use the test 265 00:07:16,060 --> 00:07:18,540 trigger icon to send that payload into the test chat and 266 00:07:18,540 --> 00:07:19,540 watch what happens. 267 00:07:19,540 --> 00:07:22,939 The activity map shows you step by step how the agent responded, 268 00:07:22,939 --> 00:07:25,779 which is how you catch a missing input before real users do. 269 00:07:25,779 --> 00:07:27,060 Event triggers authenticate 270 00:07:27,060 --> 00:07:28,060 using the maker's own 271 00:07:28,060 --> 00:07:28,779 credentials. 272 00:07:28,779 --> 00:07:30,500 The account of the person who built them. 273 00:07:30,500 --> 00:07:32,500 That's what lets the agent run unattended. 274 00:07:32,500 --> 00:07:33,860 And it's also why Microsoft 275 00:07:33,860 --> 00:07:34,939 shows a warning before you 276 00:07:34,939 --> 00:07:36,540 publish, your agent might reach 277 00:07:36,540 --> 00:07:37,740 data and systems that you can 278 00:07:37,740 --> 00:07:39,259 access on behalf of people who 279 00:07:39,259 --> 00:07:41,199 can't, so the power and the risk 280 00:07:41,199 --> 00:07:42,459 arrive together. 281 00:07:42,459 --> 00:07:45,779 An agent that can act on its own at any hour is useful, and that 282 00:07:45,779 --> 00:07:48,660 same agent is something you have to point carefully. 283 00:07:48,660 --> 00:07:50,779 Three everyday examples that make triggers. 284 00:07:50,779 --> 00:07:51,300 Click. 285 00:07:51,300 --> 00:07:54,379 Three examples will make this concrete because the idea lands 286 00:07:54,379 --> 00:07:56,180 better when you can picture it. 287 00:07:56,180 --> 00:07:59,139 Example one is the leave request and this one is a topic trigger 288 00:07:59,139 --> 00:07:59,980 an employee types. 289 00:07:59,980 --> 00:08:01,569 I need to book time off next week. 290 00:08:01,569 --> 00:08:02,730 The trigger recognizes the 291 00:08:02,730 --> 00:08:04,050 intent and starts the leave 292 00:08:04,050 --> 00:08:05,050 request topic. 293 00:08:05,050 --> 00:08:06,329 From there, the agent walks them 294 00:08:06,329 --> 00:08:07,430 through it, asking for the leave 295 00:08:07,430 --> 00:08:08,730 type and the dates and maybe a 296 00:08:08,730 --> 00:08:09,449 reason. 297 00:08:09,449 --> 00:08:12,170 Then it submits the request through a flow and confirms that 298 00:08:12,170 --> 00:08:13,089 it went through. 299 00:08:13,089 --> 00:08:15,649 The person never has to learn the menu or know the name of the 300 00:08:15,649 --> 00:08:18,250 process, because they just ask in their own words. 301 00:08:18,250 --> 00:08:21,730 Example two is the new hire and this one is an event trigger. 302 00:08:21,730 --> 00:08:23,410 Picture a new team member being added. 303 00:08:23,410 --> 00:08:26,370 An event trigger fires on that moment, and the agent sends a 304 00:08:26,370 --> 00:08:29,050 welcome message with the onboarding resources that person 305 00:08:29,050 --> 00:08:29,889 needs on day one. 306 00:08:29,889 --> 00:08:31,290 Nobody in it has to remember 307 00:08:31,290 --> 00:08:32,330 because the event is the 308 00:08:32,330 --> 00:08:33,090 reminder. 309 00:08:33,090 --> 00:08:35,730 The same pattern works for a new order coming in, where the agent 310 00:08:35,730 --> 00:08:38,929 confirms availability, checks shipping timelines, and sends 311 00:08:38,929 --> 00:08:40,009 the customer an update. 312 00:08:40,009 --> 00:08:42,610 Example three is the quiet one and it's my favorite. 313 00:08:42,610 --> 00:08:45,169 An event trigger watches the table in Dataverse, and every 314 00:08:45,169 --> 00:08:47,409 time a new row shows up, it checks whether the account name 315 00:08:47,409 --> 00:08:48,210 is a duplicate. 316 00:08:48,210 --> 00:08:51,049 When it finds one, it creates a to do task to investigate with 317 00:08:51,049 --> 00:08:52,330 the details attached. 318 00:08:52,330 --> 00:08:54,370 Nobody is staring at a spreadsheet and the agent just 319 00:08:54,370 --> 00:08:55,929 keeps watch on its own. 320 00:08:55,929 --> 00:08:57,809 Notice what all three have in common. 321 00:08:57,809 --> 00:09:00,970 The person gets a fast, natural experience and the maker spent 322 00:09:00,970 --> 00:09:03,529 their effort on the trigger and the instructions rather than on 323 00:09:03,529 --> 00:09:04,529 the chat itself. 324 00:09:04,529 --> 00:09:05,690 That's the trade you're making, 325 00:09:05,690 --> 00:09:06,769 and most of the time it's a good 326 00:09:06,769 --> 00:09:07,210 one. 327 00:09:07,210 --> 00:09:08,309 But there's a bill attached to 328 00:09:08,309 --> 00:09:09,529 that third example, the one that 329 00:09:09,529 --> 00:09:10,450 keeps watch. 330 00:09:10,450 --> 00:09:12,289 And it's the reason a perfectly 331 00:09:12,289 --> 00:09:13,250 innocent trigger can get 332 00:09:13,250 --> 00:09:13,809 expensive. 333 00:09:13,809 --> 00:09:15,250 Let's talk about that next. 334 00:09:15,250 --> 00:09:16,289 What triggers cost? 335 00:09:16,289 --> 00:09:17,570 Because they are not free. 336 00:09:17,570 --> 00:09:19,250 The cost catches almost everyone off guard. 337 00:09:19,250 --> 00:09:20,409 So let's start there. 338 00:09:20,409 --> 00:09:23,210 Copilot studio builds in units called copilot credits. 339 00:09:23,210 --> 00:09:24,850 And the counting rule that catches people. 340 00:09:24,850 --> 00:09:25,970 Is this a message? 341 00:09:25,970 --> 00:09:27,409 Is any request or message that 342 00:09:27,409 --> 00:09:28,889 triggers an action or a response 343 00:09:28,889 --> 00:09:30,570 from the agent and payload sent 344 00:09:30,570 --> 00:09:31,850 from event triggers count as 345 00:09:31,850 --> 00:09:32,529 messages. 346 00:09:32,529 --> 00:09:34,090 Read that part again because it 347 00:09:34,090 --> 00:09:35,490 means the event itself is 348 00:09:35,490 --> 00:09:36,049 billable. 349 00:09:36,049 --> 00:09:38,889 Every time your trigger wakes the agent up, that's a message. 350 00:09:38,889 --> 00:09:40,529 Now think about the recurrence trigger. 351 00:09:40,529 --> 00:09:41,929 It fires on a timer every time 352 00:09:41,929 --> 00:09:43,009 the clock reaches the interval 353 00:09:43,009 --> 00:09:43,690 you set. 354 00:09:43,690 --> 00:09:44,850 If you set it to every ten 355 00:09:44,850 --> 00:09:46,330 minutes, your agent gets a 356 00:09:46,330 --> 00:09:47,610 payload, a message every ten 357 00:09:47,610 --> 00:09:47,970 minutes. 358 00:09:47,970 --> 00:09:50,169 That's one hundred and forty four messages a day for as long 359 00:09:50,169 --> 00:09:51,169 as you leave it switched on. 360 00:09:51,169 --> 00:09:52,970 Whether anything useful happens or not. 361 00:09:52,970 --> 00:09:54,250 It's easy to build that on a 362 00:09:54,250 --> 00:09:55,450 Friday and forget about it by 363 00:09:55,450 --> 00:09:56,210 Monday. 364 00:09:56,210 --> 00:09:58,690 The rates themselves are set in credits, and they vary by what 365 00:09:58,690 --> 00:09:59,610 the agent does. 366 00:09:59,610 --> 00:10:00,850 A classic answer, one that you 367 00:10:00,850 --> 00:10:02,450 wrote by hand, costs about one 368 00:10:02,450 --> 00:10:02,889 credit. 369 00:10:02,889 --> 00:10:05,769 A generative answer, one the AI writes on the fly is about two 370 00:10:05,769 --> 00:10:08,370 an agent action, which is a step like a trigger or a topic 371 00:10:08,370 --> 00:10:10,409 transition is about five. 372 00:10:10,409 --> 00:10:11,409 Grounding an answer in your 373 00:10:11,409 --> 00:10:12,830 company's own data costs more 374 00:10:12,830 --> 00:10:14,610 around ten and running agent 375 00:10:14,610 --> 00:10:14,929 flows. 376 00:10:14,929 --> 00:10:17,929 The predefined sequences of steps is built at about thirteen 377 00:10:17,929 --> 00:10:19,409 credits per one hundred actions. 378 00:10:19,409 --> 00:10:21,210 Those numbers are illustrative and they change. 379 00:10:21,210 --> 00:10:23,929 So check the current licensing page before you plan a budget. 380 00:10:23,929 --> 00:10:26,409 There's a piece of good news hiding in here, though. 381 00:10:26,409 --> 00:10:28,529 If the person using your agent has a Microsoft three. 382 00:10:28,529 --> 00:10:30,889 Sixty five copilot license employee facing usage is 383 00:10:30,889 --> 00:10:32,129 included at no charge. 384 00:10:32,129 --> 00:10:33,409 The exception is agent flows 385 00:10:33,409 --> 00:10:34,409 that are triggered some other 386 00:10:34,409 --> 00:10:35,690 way, which bill at the normal 387 00:10:35,690 --> 00:10:36,169 rate. 388 00:10:36,169 --> 00:10:37,250 And there's a hard edge worth 389 00:10:37,250 --> 00:10:38,610 knowing if your environment 390 00:10:38,610 --> 00:10:40,570 blows past its prepaid capacity 391 00:10:40,570 --> 00:10:41,649 at one hundred and twenty five 392 00:10:41,649 --> 00:10:43,090 percent, agents get switched 393 00:10:43,090 --> 00:10:43,409 off. 394 00:10:43,409 --> 00:10:46,210 Users see a message saying the agent is unavailable or that 395 00:10:46,210 --> 00:10:47,289 there's a billing issue. 396 00:10:47,289 --> 00:10:48,809 Nobody likes that surprise. 397 00:10:48,809 --> 00:10:51,649 So a trigger that runs forever deserves a second look before 398 00:10:51,649 --> 00:10:53,450 you walk away. 399 00:10:53,450 --> 00:10:54,610 The limits and the mistakes 400 00:10:54,610 --> 00:10:56,370 beginners make money is one 401 00:10:56,370 --> 00:10:58,289 limit, reliability is the other, 402 00:10:58,289 --> 00:10:59,409 and this is the one they hit 403 00:10:59,409 --> 00:11:00,169 first. 404 00:11:00,169 --> 00:11:01,210 Start with the words because 405 00:11:01,210 --> 00:11:02,460 most trigger problems live 406 00:11:02,460 --> 00:11:03,000 there. 407 00:11:03,000 --> 00:11:05,559 Too few example phrases and your agent misses requests. 408 00:11:05,559 --> 00:11:08,720 It should catch duplicate phrases across two topics and it 409 00:11:08,720 --> 00:11:10,000 can't tell them apart. 410 00:11:10,000 --> 00:11:13,200 Long unnatural sentences like initiate employee absence 411 00:11:13,200 --> 00:11:15,879 management transaction confuse the matching because nobody 412 00:11:15,879 --> 00:11:17,200 actually talks that way. 413 00:11:17,200 --> 00:11:19,899 And when two topics are simply too similar, the agent has to 414 00:11:19,899 --> 00:11:22,720 guess the fix is boring and effective, which is to read how 415 00:11:22,720 --> 00:11:25,919 real people ask from support tickets or chat logs and write 416 00:11:25,919 --> 00:11:27,279 your phrases the way they do. 417 00:11:27,279 --> 00:11:29,840 There's a quick sanity check you can run on yourself. 418 00:11:29,840 --> 00:11:32,159 If two of your topics could both plausibly answer the same 419 00:11:32,159 --> 00:11:33,600 message, they'll fight over it. 420 00:11:33,600 --> 00:11:36,279 So either merge them or sharpen their descriptions until only 421 00:11:36,279 --> 00:11:37,559 one clearly fits. 422 00:11:37,559 --> 00:11:38,960 Then there are instruction problems. 423 00:11:38,960 --> 00:11:40,600 Vague agent instructions leave 424 00:11:40,600 --> 00:11:42,279 the agent guessing and payload 425 00:11:42,279 --> 00:11:43,679 instructions that contradict the 426 00:11:43,679 --> 00:11:43,960 agent. 427 00:11:43,960 --> 00:11:46,799 Instructions produce errors or odd behavior. 428 00:11:46,799 --> 00:11:49,600 So the advice is to say what you mean in one place and keep the 429 00:11:49,600 --> 00:11:51,480 two sets of instructions agreeing with each other. 430 00:11:51,480 --> 00:11:54,000 Now the harder truth, and I want to be fair about it. 431 00:11:54,000 --> 00:11:57,240 Some makers report that event triggers fire inconsistently. 432 00:11:57,240 --> 00:12:00,159 One person on Microsoft's own question and answer site 433 00:12:00,159 --> 00:12:03,480 described an email trigger that worked, then stopped, then 434 00:12:03,480 --> 00:12:06,639 worked again after republishing with Power Automate reporting 435 00:12:06,639 --> 00:12:08,039 success the whole time. 436 00:12:08,039 --> 00:12:09,480 The likely reasons are messy and 437 00:12:09,480 --> 00:12:11,080 they include preview status in 438 00:12:11,080 --> 00:12:12,679 some environments, connection 439 00:12:12,679 --> 00:12:14,759 and identity mismatches, a flow, 440 00:12:14,759 --> 00:12:15,720 and an agent sitting in 441 00:12:15,720 --> 00:12:16,840 different environments 442 00:12:16,840 --> 00:12:18,559 throttling under load and the 443 00:12:18,559 --> 00:12:20,879 way email triggers batch their 444 00:12:20,879 --> 00:12:22,399 events. 445 00:12:22,399 --> 00:12:24,679 Treat that as a community report rather than an official 446 00:12:24,679 --> 00:12:27,879 guarantee, and design like the event might not fire. 447 00:12:27,879 --> 00:12:30,960 That last point matters if the work really has to happen. 448 00:12:30,960 --> 00:12:33,879 Don't let a single trigger be your only safety net. 449 00:12:33,879 --> 00:12:35,039 Microsoft's own guidance 450 00:12:35,039 --> 00:12:36,440 suggests using a power automate 451 00:12:36,440 --> 00:12:37,600 flow when a connector won't 452 00:12:37,600 --> 00:12:38,879 cooperate, and building the 453 00:12:38,879 --> 00:12:40,139 whole thing so it can retry 454 00:12:40,139 --> 00:12:40,840 safely. 455 00:12:40,840 --> 00:12:43,000 There are some plain guardrails worth carrying with you. 456 00:12:43,000 --> 00:12:45,580 Keep the number of actions and topics an agent calls in a row 457 00:12:45,580 --> 00:12:48,399 under fifteen because long chains get unreliable. 458 00:12:48,399 --> 00:12:49,879 Be careful with very frequent or 459 00:12:49,879 --> 00:12:50,759 never ending recurrence 460 00:12:50,759 --> 00:12:51,759 triggers, since they burn 461 00:12:51,759 --> 00:12:52,860 resources and can get you 462 00:12:52,860 --> 00:12:53,480 throttled. 463 00:12:53,480 --> 00:12:56,159 Keep an eye on how many triggers are active in an environment, 464 00:12:56,159 --> 00:12:58,960 and remember that availability isn't only up to you, because an 465 00:12:58,960 --> 00:13:02,240 administrator can block certain triggers through data policies 466 00:13:02,240 --> 00:13:04,960 and adding event triggers requires a specific sharing 467 00:13:04,960 --> 00:13:06,679 setting to be turned on. 468 00:13:06,679 --> 00:13:08,360 So when a trigger simply isn't 469 00:13:08,360 --> 00:13:09,720 there for you, it's often a 470 00:13:09,720 --> 00:13:11,159 setting rather than a missing 471 00:13:11,159 --> 00:13:12,360 feature. 472 00:13:12,360 --> 00:13:14,399 Where all of this lives in Microsoft three hundred sixty 473 00:13:14,399 --> 00:13:15,320 five and Azure. 474 00:13:15,320 --> 00:13:16,840 Step back and it helps to see 475 00:13:16,840 --> 00:13:17,960 where triggers sit in the bigger 476 00:13:17,960 --> 00:13:18,600 picture. 477 00:13:18,600 --> 00:13:21,159 There are really three layers and you can hold them as people. 478 00:13:21,159 --> 00:13:22,440 Platform and power. 479 00:13:22,440 --> 00:13:25,200 Microsoft three hundred sixty five copilot is the layer for 480 00:13:25,200 --> 00:13:29,519 people, a license that puts AI into the tools they already use. 481 00:13:29,519 --> 00:13:31,279 Copilot studio is the platform 482 00:13:31,279 --> 00:13:32,679 layer, the low code place where 483 00:13:32,679 --> 00:13:34,279 you build the agent, wire up its 484 00:13:34,279 --> 00:13:35,639 triggers and connect it to the 485 00:13:35,639 --> 00:13:37,000 channels where people meet it, 486 00:13:37,000 --> 00:13:38,419 like Microsoft Teams or a web 487 00:13:38,419 --> 00:13:39,120 page. 488 00:13:39,120 --> 00:13:40,639 Azure is the power layer where 489 00:13:40,639 --> 00:13:42,480 heavier custom AI work runs and 490 00:13:42,480 --> 00:13:43,879 where the pay as you go meter 491 00:13:43,879 --> 00:13:44,360 lives. 492 00:13:44,360 --> 00:13:45,879 That meter is a nice detail for 493 00:13:45,879 --> 00:13:46,919 beginners because it lets you 494 00:13:46,919 --> 00:13:47,519 start small. 495 00:13:47,519 --> 00:13:50,480 You can build inside Copilot Studio and never touch Azure. 496 00:13:50,480 --> 00:13:51,840 If you do grow, you can link an 497 00:13:51,840 --> 00:13:53,039 Azure subscription for pay as 498 00:13:53,039 --> 00:13:54,840 you go, which also means overage 499 00:13:54,840 --> 00:13:55,919 bills to that subscription 500 00:13:55,919 --> 00:13:57,039 instead of switching your agents 501 00:13:57,039 --> 00:13:57,480 off. 502 00:13:57,480 --> 00:13:58,639 If you bring your own model, 503 00:13:58,639 --> 00:13:59,759 including the ones from Azure 504 00:13:59,759 --> 00:14:01,399 Foundry that's built separately 505 00:14:01,399 --> 00:14:02,710 outside the copilot studio 506 00:14:02,710 --> 00:14:03,350 credit rates. 507 00:14:03,350 --> 00:14:04,350 So the shape is simple. 508 00:14:04,350 --> 00:14:05,029 Microsoft three. 509 00:14:05,029 --> 00:14:07,029 Sixty five copilot is your license. 510 00:14:07,029 --> 00:14:10,389 Copilot studio is the place you build and the heavy lifting plus 511 00:14:10,389 --> 00:14:11,389 the billing flexibility. 512 00:14:11,389 --> 00:14:12,309 Live in Azure. 513 00:14:12,309 --> 00:14:14,029 You don't need all three to get started. 514 00:14:14,029 --> 00:14:16,470 And that's the encouraging part. 515 00:14:16,470 --> 00:14:18,830 A quick recap and one question for you. 516 00:14:18,830 --> 00:14:20,029 Let's pull it together. 517 00:14:20,029 --> 00:14:22,909 A trigger is the small condition that decides when your agent 518 00:14:22,909 --> 00:14:25,230 acts the doorway into a topic. 519 00:14:25,230 --> 00:14:26,190 There are two families. 520 00:14:26,190 --> 00:14:27,549 Topic triggers that fire when a 521 00:14:27,549 --> 00:14:29,309 person asks, and event triggers 522 00:14:29,309 --> 00:14:29,710 that fire. 523 00:14:29,710 --> 00:14:32,629 When a system event happens, you feed topic triggers with 524 00:14:32,629 --> 00:14:35,509 realistic example phrases and you point event triggers at 525 00:14:35,509 --> 00:14:38,029 events with instructions that tell the agent what to do. 526 00:14:38,029 --> 00:14:40,350 Get those right and your agent feels alive. 527 00:14:40,350 --> 00:14:41,669 Get them vague and it either 528 00:14:41,669 --> 00:14:42,850 does nothing or does the wrong 529 00:14:42,850 --> 00:14:43,429 thing. 530 00:14:43,429 --> 00:14:45,789 One question for you if you could switch on one trigger 531 00:14:45,789 --> 00:14:48,590 tomorrow, what would you have your agent watch for? 532 00:14:48,590 --> 00:14:51,789 A new file, a new hire, or a schedule that never sleeps? 533 00:14:51,789 --> 00:14:52,950 Tell me in the comments, because 534 00:14:52,950 --> 00:14:53,970 that answer usually tells you 535 00:14:53,970 --> 00:14:54,909 which kind of trigger you 536 00:14:54,909 --> 00:14:55,669 actually need. 537 00:14:55,669 --> 00:14:57,029 If this made the topic feel less 538 00:14:57,029 --> 00:14:59,490 intimidating, subscribe to M365 539 00:14:59,490 --> 00:15:01,350 FM to master the tools and the 540 00:15:01,350 --> 00:15:02,669 mindset shaping the future of 541 00:15:02,669 --> 00:15:03,269 work.