Copilot Shows Wrong or Old Data: Fix Permissions and Trust Fast

1
00:00:00,000 --> 00:00:03,340
Your co-pilot rollout will fail for one reason, people not tech.

2
00:00:03,340 --> 00:00:06,740
You'll light up licenses, publish a heroic memo, and three months later?

3
00:00:06,740 --> 00:00:07,460
Nothing.

4
00:00:07,460 --> 00:00:10,540
Wasted budget, annoyed executives, zero behavior change.

5
00:00:10,540 --> 00:00:13,740
The truth, deployment isn't adoption, behavior change is the product.

6
00:00:13,740 --> 00:00:16,780
Here's the playbook practitioners actually use, targeted use cases,

7
00:00:16,780 --> 00:00:19,660
a leadership coalition that learns in public real telemetry,

8
00:00:19,660 --> 00:00:22,980
and yes, co-pilot studio builds that embed into workflows.

9
00:00:22,980 --> 00:00:25,900
I'll show you war stories, the week one decision that predicts MAU,

10
00:00:25,900 --> 00:00:27,660
and a 90-day plan you can steal.

11
00:00:27,660 --> 00:00:30,660
There's a single choice in your first week that tells me your outcome.

12
00:00:30,660 --> 00:00:31,660
Stay for that.

13
00:00:31,660 --> 00:00:33,660
Why tech first fails?

14
00:00:33,660 --> 00:00:35,940
The mismatch between tools and habits.

15
00:00:35,940 --> 00:00:38,260
Most teams treat co-pilot like a feature toggle.

16
00:00:38,260 --> 00:00:41,100
Flip it on, throw a town hall, and expect magic.

17
00:00:41,100 --> 00:00:46,180
The average user nods, opens word once, types, summarize, gets mush,

18
00:00:46,180 --> 00:00:48,140
and quietly never returns.

19
00:00:48,140 --> 00:00:51,420
The mismatch is brutal, you shift capability, they needed new habits.

20
00:00:51,420 --> 00:00:55,620
It's not a software project, it's a behavior project with technical dependencies.

21
00:00:55,620 --> 00:00:58,420
The thing most people miss is simple, deployment adoption.

22
00:00:58,420 --> 00:01:02,020
Adoption happens when a specific person does a specific task a new way

23
00:01:02,020 --> 00:01:04,220
because it's faster, safer, or clearer.

24
00:01:04,220 --> 00:01:07,940
If that sentence isn't engineered into your rollout, you're running a faith-based initiative.

25
00:01:07,940 --> 00:01:09,460
Now here's where most people mess up.

26
00:01:09,460 --> 00:01:13,100
They turn it on without roll targeting or use case design.

27
00:01:13,100 --> 00:01:16,700
No one wakes up thinking, "I hope to generally be more productive."

28
00:01:16,700 --> 00:01:20,460
They wake up thinking, "I need a first draft of the QBR email before 10."

29
00:01:20,460 --> 00:01:22,500
Generic prompts produce generic outcomes

30
00:01:22,500 --> 00:01:27,100
with your team correctly interprets as not worth the cognitive tax of changing how they work.

31
00:01:27,100 --> 00:01:30,700
Enter the emotional thermostat, leaders set it whether they mean to or not.

32
00:01:30,700 --> 00:01:34,700
If the executive narrative is AI anxiety, fear of errors, fear of optics,

33
00:01:34,700 --> 00:01:35,860
usage collapses.

34
00:01:35,860 --> 00:01:41,220
If the narrative is curiosity, permission to try to show ugly drafts to iterate usage climbs.

35
00:01:41,220 --> 00:01:42,980
The truth? People don't need cheerleading.

36
00:01:42,980 --> 00:01:45,580
They need visible permission and a reason to care this week.

37
00:01:45,580 --> 00:01:47,500
Leadership commitment isn't a name on a slide.

38
00:01:47,500 --> 00:01:50,500
It's a coalition that models the behavior in the open.

39
00:01:50,500 --> 00:01:53,900
Learn in public. Leaders run staff prep with co-pilot on screen.

40
00:01:53,900 --> 00:01:56,980
Narrate prompts accept imperfect output and critique it.

41
00:01:56,980 --> 00:01:59,500
That single act kills shame and creates permission.

42
00:01:59,500 --> 00:02:01,940
You want culture? Make the right thing visible.

43
00:02:01,940 --> 00:02:04,620
The game changer nobody talks about is celebration mechanics.

44
00:02:04,620 --> 00:02:08,700
Cultures scale what they celebrate if you clap for perfect and the people hide learning.

45
00:02:08,700 --> 00:02:10,820
If you clap for practice, people try.

46
00:02:10,820 --> 00:02:13,700
Create a weekly 10-minute ritual, one-roll spotlight,

47
00:02:13,700 --> 00:02:16,900
one prompt that save time to draft, one artifact you can reuse.

48
00:02:16,900 --> 00:02:19,700
Consistency beats flare, predictability beats hype.

49
00:02:20,700 --> 00:02:22,700
Let me show you exactly how this plays out.

50
00:02:22,700 --> 00:02:24,100
Field story.

51
00:02:24,100 --> 00:02:27,100
A company lit up licenses across 1,200 seats.

52
00:02:27,100 --> 00:02:29,500
30 days later, M-I-U was a rounding error.

53
00:02:29,500 --> 00:02:32,700
IT blamed training, business blamed accuracy.

54
00:02:32,700 --> 00:02:34,700
Then a VP ran a live demo.

55
00:02:34,700 --> 00:02:37,700
Her real deck real notes real time to draft in front of her team.

56
00:02:37,700 --> 00:02:43,100
She recorded it, shared the prompts, and required one AI assisted artifact in the next staff meeting.

57
00:02:43,100 --> 00:02:45,300
M-I-U climbed the following week and kept climbing.

58
00:02:45,300 --> 00:02:47,500
The deployment didn't change. The people did.

59
00:02:47,500 --> 00:02:50,300
Before we continue, you need to understand the physics of habits.

60
00:02:50,300 --> 00:02:54,100
New tools don't replace old ones unless they win the speed to first use race.

61
00:02:54,100 --> 00:02:56,600
That means your co-pilot program must surface.

62
00:02:56,600 --> 00:03:01,100
Tuesday tasks, not moon shots, draft the client recap, tighten the paragraph, extract action items,

63
00:03:01,100 --> 00:03:02,800
propose a meeting agenda.

64
00:03:02,800 --> 00:03:04,900
Microwinds create repeat behavior.

65
00:03:04,900 --> 00:03:06,800
Repeat behavior becomes the default.

66
00:03:06,800 --> 00:03:11,100
Once you nail that everything else clicks, use case clarity, anchors, trust conversations.

67
00:03:11,100 --> 00:03:13,300
When people know the task, they'll learn the limits.

68
00:03:13,300 --> 00:03:15,800
When they see leaders using it even imperfectly,

69
00:03:15,800 --> 00:03:17,600
they believe they're allowed to do the same.

70
00:03:17,600 --> 00:03:20,600
When you collect artifacts, prompts before after drafts, templates,

71
00:03:20,600 --> 00:03:23,000
you lower the activation energy for the next person.

72
00:03:23,000 --> 00:03:25,400
The reason this works is brutally practical.

73
00:03:25,400 --> 00:03:28,600
Behavior changes when the path of least resistance changes.

74
00:03:28,600 --> 00:03:32,200
If opening co-pilot means facing a blank box in judgment, they won't.

75
00:03:32,200 --> 00:03:36,900
If opening co-pilot means picking from a small library of role-specific starters

76
00:03:36,900 --> 00:03:39,600
that have already produced wins in their team, they will.

77
00:03:39,600 --> 00:03:40,800
Now a common objection.

78
00:03:40,800 --> 00:03:42,400
Can't we just train everyone?

79
00:03:42,400 --> 00:03:45,600
You can and you should, but training without use case design is theater.

80
00:03:45,600 --> 00:03:47,900
Adults learn to solve their own problems.

81
00:03:47,900 --> 00:03:51,400
Tie every session to a job task, show the delta in time to draft

82
00:03:51,400 --> 00:03:53,700
and hand them a reusable prompt pack.

83
00:03:53,700 --> 00:03:56,500
Then follow up with office hours, teach, then coach.

84
00:03:56,500 --> 00:03:58,200
Lecture alone is a souvenir.

85
00:03:58,200 --> 00:04:01,100
Another objection will wait until governance is perfect.

86
00:04:01,100 --> 00:04:04,200
Governance matters and will map it, but paralysis is not policies.

87
00:04:04,200 --> 00:04:06,500
Start with a safe sandbox and role-limited pilots

88
00:04:06,500 --> 00:04:08,500
while you clean labels and DLP.

89
00:04:08,500 --> 00:04:11,200
You don't suspend driving school because the highway exists,

90
00:04:11,200 --> 00:04:14,400
you start in a parking lot with cones in a coach.

91
00:04:14,400 --> 00:04:16,500
If you remember nothing else, remember this.

92
00:04:16,500 --> 00:04:18,000
Behavior change is the product.

93
00:04:18,000 --> 00:04:21,700
Your tooling enables it, your leaders authorize it, your rituals sustain it

94
00:04:21,700 --> 00:04:23,200
and your artifacts scale it.

95
00:04:23,200 --> 00:04:27,500
Once you accept it's a people problem, you finally get to use technology correctly.

96
00:04:27,500 --> 00:04:29,300
As the lever, not the load.

97
00:04:29,300 --> 00:04:30,200
And that's our pivot.

98
00:04:30,200 --> 00:04:34,800
Now that you see the mismatch, you need a map to avoid the common failure modes coming next.

99
00:04:34,800 --> 00:04:36,000
Failure mode 1.

100
00:04:36,000 --> 00:04:38,300
Vague use cases and weak problem framing.

101
00:04:38,300 --> 00:04:40,400
This is the one that quietly kills momentum.

102
00:04:40,400 --> 00:04:42,900
Your announce co-pilot makes everyone more productive,

103
00:04:42,900 --> 00:04:45,800
which is the corporate equivalent of eat healthier.

104
00:04:45,800 --> 00:04:48,500
No one knows what to do next, so they do nothing.

105
00:04:48,500 --> 00:04:51,800
The thing most people miss, generic prompts produce generic outcomes

106
00:04:51,800 --> 00:04:55,700
and generic outcomes don't beat current habits, so there's no pull from teams.

107
00:04:55,700 --> 00:05:02,000
What grade looks like is painfully specific, role specific, task-level scenarios with them before after.

108
00:05:02,000 --> 00:05:04,300
Not sales rights better proposals.

109
00:05:04,300 --> 00:05:09,000
It's account executives reduce time to draft for first proposals from 90 minutes to 25

110
00:05:09,000 --> 00:05:11,400
using this three prompt sequence and this template.

111
00:05:11,400 --> 00:05:13,100
Not finance saves time.

112
00:05:13,100 --> 00:05:17,000
It's controller's reconcil variance notes in 15 minutes by summarizing journal entries

113
00:05:17,000 --> 00:05:19,500
then generating hypotheses to investigate.

114
00:05:19,500 --> 00:05:21,400
Clarity creates demand.

115
00:05:21,400 --> 00:05:23,800
Use the triad that actually moves needles.

116
00:05:23,800 --> 00:05:26,600
Time to draft decision clarity and meeting compression.

117
00:05:26,600 --> 00:05:29,800
If a scenario doesn't shrink time to draft, sharpen a decision

118
00:05:29,800 --> 00:05:32,700
or remove 10 minutes from a meeting, deprioritize it.

119
00:05:32,700 --> 00:05:34,500
That constraint focuses everyone.

120
00:05:34,500 --> 00:05:39,200
The truth, you'll get faster wins by shaving meetings and drafts than by chasing moonshots.

121
00:05:39,200 --> 00:05:41,100
Here's the shortcut nobody teaches.

122
00:05:41,100 --> 00:05:43,400
The 10/30/60 model for pilots.

123
00:05:43,400 --> 00:05:46,000
Design Tuesday tasks, not science projects.

124
00:05:46,000 --> 00:05:47,500
10 minutes.

125
00:05:47,500 --> 00:05:53,300
Micro-asks like tighten this paragraph or extract action items.

126
00:05:53,300 --> 00:05:58,500
30 minutes short artifacts like a client recap, a status email or a meeting agenda.

127
00:05:58,500 --> 00:06:02,400
60 minutes heavier drafts like a proposal outline or a policy summary.

128
00:06:02,400 --> 00:06:05,600
Force each role to pick two tasks in each band.

129
00:06:05,600 --> 00:06:08,900
Now you have six concrete scenarios per role that recur weekly.

130
00:06:08,900 --> 00:06:11,400
Let me show you exactly how to frame one.

131
00:06:11,400 --> 00:06:12,200
Why?

132
00:06:12,200 --> 00:06:15,000
Proposal drafting eats your mornings and delays follow ups.

133
00:06:15,000 --> 00:06:17,500
Faster first drafts mean more at bats.

134
00:06:17,500 --> 00:06:18,800
What?

135
00:06:18,800 --> 00:06:20,500
A repeatable prompt chain.

136
00:06:20,500 --> 00:06:22,100
One, capture context.

137
00:06:22,100 --> 00:06:24,300
Two, outline with constraints.

138
00:06:24,300 --> 00:06:26,900
Three, expand sections with examples.

139
00:06:26,900 --> 00:06:29,200
Four, critique for tone and compliance.

140
00:06:29,200 --> 00:06:30,200
How?

141
00:06:30,200 --> 00:06:34,500
Start with last quarter's winning proposal as grounding, swap client details, run the chain

142
00:06:34,500 --> 00:06:37,100
and save the best outputs to an artifact library.

143
00:06:37,100 --> 00:06:41,800
Common mistake starting from a blank co-pilot pane and vibes, a quick field story.

144
00:06:41,800 --> 00:06:45,700
A sales org thought they were doing AI, adoption flatlined.

145
00:06:45,700 --> 00:06:51,200
We replotted baselines actual stopwatch time from brief to first draft, average 82 minutes.

146
00:06:51,200 --> 00:06:56,300
We introduced a 30 minute proposal starter, context block, audience, constraints,

147
00:06:56,300 --> 00:07:00,200
and a don't hallucinate pricing insert placeholders instruction.

148
00:07:00,200 --> 00:07:03,200
Two weeks later, median time to draft set at 27 minutes.

149
00:07:03,200 --> 00:07:08,500
The difference wasn't magic, it was framing constraints and an artifact library that killed the blank page tags.

150
00:07:08,500 --> 00:07:14,200
About that artifact library, if you don't capture prompts before after examples and approve templates,

151
00:07:14,200 --> 00:07:17,000
you force every user to rediscover the same path.

152
00:07:17,000 --> 00:07:18,100
That's malpractice.

153
00:07:18,100 --> 00:07:23,700
Build it where they work, share point for docs, team steps for quick access, pin its snippets in word outlook,

154
00:07:23,700 --> 00:07:25,300
tag by role and task.

155
00:07:25,300 --> 00:07:26,600
The library is your on ramp.

156
00:07:26,600 --> 00:07:30,000
Without it, you're asking people to merge onto the highway from a gravel road.

157
00:07:30,000 --> 00:07:31,700
Common mistakes I keep seeing.

158
00:07:31,700 --> 00:07:33,400
Be more productive campaigns.

159
00:07:33,400 --> 00:07:36,500
That's a slogan, not a scenario, training without artifacts.

160
00:07:36,500 --> 00:07:39,000
People leave with theory and return to a blank box.

161
00:07:39,000 --> 00:07:39,900
Moonshot selection.

162
00:07:39,900 --> 00:07:43,300
If the first demo is a complex legal brief, you've guaranteed disappointment.

163
00:07:43,300 --> 00:07:44,000
No baselines.

164
00:07:44,000 --> 00:07:47,600
If you don't measure current time to draft or meeting length, you can't prove improvement.

165
00:07:47,600 --> 00:07:48,400
No owner.

166
00:07:48,400 --> 00:07:53,100
A use case without a single accountable role lead will drift into ambiguity.

167
00:07:53,100 --> 00:07:58,200
Now the practical template you'll reuse, context, constraint, critique, continue.

168
00:07:58,200 --> 00:07:59,500
See for prompting.

169
00:07:59,500 --> 00:08:00,700
Context gives ground truth.

170
00:08:00,700 --> 00:08:04,000
Constraints set tone, audience, length and exclusions.

171
00:08:04,000 --> 00:08:07,600
Critique asks Copilot to evaluate against criteria you care about.

172
00:08:07,600 --> 00:08:09,900
Continue iterates on the weakest section.

173
00:08:09,900 --> 00:08:14,600
Teach this pattern and embed ready to run snippets inside the apps where people live.

174
00:08:14,600 --> 00:08:16,300
Do this for two roles this month.

175
00:08:16,300 --> 00:08:21,100
Six Tuesday tasks each baseline run C4, save artifacts, report the triad.

176
00:08:21,100 --> 00:08:27,200
Once the winds are visible, the pull begins in failure mode 2, data access, governance theater and trust gaps.

177
00:08:27,200 --> 00:08:29,700
This is where otherwise competent teams go off the rails.

178
00:08:29,700 --> 00:08:33,100
They either lock everything down so hard, Copilot can't find a grocery list.

179
00:08:33,100 --> 00:08:38,100
Or they fling open the doors and pray no one notices finance in the break room with HR salary folder.

180
00:08:38,100 --> 00:08:43,600
Both are lazy, overlocking produces useless answers, reckless exposure detonates trust.

181
00:08:43,600 --> 00:08:44,900
Adoption dies in either case.

182
00:08:44,900 --> 00:08:57,900
Do the adult thing map your data surfaces, not vibes, surfaces, sharepoint sites and libraries, one drive scopes, exchange mailboxes and calendars, teams, channels and chats, plus whatever CRM or line of business systems you actually want in play.

183
00:08:57,900 --> 00:09:04,300
If you can't name the top 10 content sources your target roles use every week, you are not ready to argue about governance.

184
00:09:04,300 --> 00:09:06,700
You're guessing.

185
00:09:06,700 --> 00:09:08,500
Now guardrails.

186
00:09:08,500 --> 00:09:11,100
Sensitivity labels are not decorative stickers.

187
00:09:11,100 --> 00:09:11,900
Use them.

188
00:09:11,900 --> 00:09:16,300
Apply or fix labels on the high traffic libraries and make sure they actually drive policy.

189
00:09:16,300 --> 00:09:20,500
Who can access, where content can travel and what Copilot can surface.

190
00:09:20,500 --> 00:09:25,300
Add DLP policies that stop obvious exfiltration and don't break Tuesday work.

191
00:09:25,300 --> 00:09:31,100
Yes, that means you pilot the policies with live humans and adjust because Microsoft is not performing magic tricks.

192
00:09:31,100 --> 00:09:34,000
Policies do what you told them to do, painfully literally.

193
00:09:34,000 --> 00:09:36,100
Before we continue, you need to understand grounding.

194
00:09:36,100 --> 00:09:37,600
Retrieval is not training.

195
00:09:37,600 --> 00:09:41,500
Copilot isn't hoovering your crown jewels into a public model.

196
00:09:41,500 --> 00:09:47,200
It's retrieving content your user already has permission to see and using it as context for generation.

197
00:09:47,200 --> 00:09:52,700
Kill the it reads everything myth in week one or enjoy permanent ghost stories in every hallway.

198
00:09:52,700 --> 00:09:57,500
The physics matter if Alice can't open a file in SharePoint, Copilot won't use it for Alice either.

199
00:09:57,500 --> 00:09:59,900
Same ACLs, same consequences.

200
00:09:59,900 --> 00:10:03,000
The game changer nobody talks about is the safe sandbox.

201
00:10:03,000 --> 00:10:05,900
Create a curated content space per pilot role.

202
00:10:05,900 --> 00:10:07,300
Approved examples.

203
00:10:07,300 --> 00:10:12,400
Sanitized templates, prior wins and intentionally label documents that demonstrate boundaries.

204
00:10:12,400 --> 00:10:13,900
It's the parking lot with cones.

205
00:10:13,900 --> 00:10:17,500
People learn faster when they're not afraid of steering into legal exposure.

206
00:10:17,500 --> 00:10:19,500
Put the sandbox one click from where they work.

207
00:10:19,500 --> 00:10:23,800
Teams tab, Pint SharePoint and a link inside the Copilot learn more panel.

208
00:10:23,800 --> 00:10:25,600
Reduce friction or they won't use it.

209
00:10:25,600 --> 00:10:31,100
Field reality, a finance team panicked when a draft summary referenced old board materials.

210
00:10:31,100 --> 00:10:35,400
Governance theater ensued, freeze everything, unplugged the internet, lighter candle.

211
00:10:35,400 --> 00:10:36,400
What actually fixed it?

212
00:10:36,400 --> 00:10:41,600
A two day label cleanup sprint on the board site, removing stale everyone links from a legacy library

213
00:10:41,600 --> 00:10:46,500
and adding an exclusion rule so Copilot couldn't ground from that site until the cleanup passed a spot check.

214
00:10:46,500 --> 00:10:51,500
No breach, no scandal just permissions doing exactly what they were misconfigured to do.

215
00:10:51,500 --> 00:10:53,300
Common mistakes you'll avoid now.

216
00:10:53,300 --> 00:10:57,300
Announcing Copilot sees only what you can see without showing it live.

217
00:10:57,300 --> 00:11:01,300
Show an end user who can't access a file then show Copilot failing to use it.

218
00:11:01,300 --> 00:11:03,000
Demonstration beats rumour.

219
00:11:03,000 --> 00:11:04,900
Writing policies in an ivory tower.

220
00:11:04,900 --> 00:11:09,500
DLP that blocks normal paste operations will teach your users to hate you faster than any memo.

221
00:11:09,500 --> 00:11:13,100
Ignoring audit and telemetry turn on the logs and you want to explain not speculate.

222
00:11:13,100 --> 00:11:13,700
Quick win.

223
00:11:13,700 --> 00:11:22,400
Publish a plain English one-pager what Copilot can access what it can't what the labels mean and who to call if something looks off then paired with a five minute mid-busting video.

224
00:11:22,400 --> 00:11:29,400
Retrieval verse training scope boundaries and how to report a misscoped file adults trust what they can verify give them receipts.

225
00:11:29,400 --> 00:11:42,500
Once you map surfaces fix labels and stage the sandbox you'll notice something confidence returns and when confidence returns usage returns which brings us to the machine that keeps usage from spiking and dying your change engine.

226
00:11:42,500 --> 00:12:03,300
Failure mode 3 no change management engine or comms cadence most rollouts die on calendar not code week one launch party CEO quote shiny deck week two attraining webinar week three silence without a drum beat the program flatlines the truth tactics without cadence are cardio without a heartbeat you need an engine that converts novelty into habit.

227
00:12:03,300 --> 00:12:12,100
Build a cadence stack top layer the executive narrative clear repeated statements that Copilot is for velocity and clarity not for replacing judgment.

228
00:12:12,100 --> 00:12:29,800
Middle layer weekly wins short role specific examples with screenshots or 30 second clips bottom layer office hours recurring sessions where people bring real tasks get live help and leave with artifacts you don't need fireworks you need rhythm champions are your transmission but the thing most people miss is that not all champions are the same.

229
00:12:29,800 --> 00:12:54,300
Here to peer help us need empathy and patterns spotting they translate c4 prompting into the slang of their team champions who mentor leaders need stage craft and coaching they script live demos set expectations and rescue flops gracefully to tracks to playbooks one shared library learn in public or expect your people to hide leaders must show their work wins and misses narrate prompts call out where copilot guest wrong and show the critique step.

230
00:12:54,300 --> 00:13:18,800
When a VP says I use this to get from blank page to outline in eight minutes then I rewrote section three the org here's permission granted when that same VP hides behind perfect outputs the org here's don't get caught trying a field story to prove it a division launched copilot with decent training but an enemy comes plan am I you sag by week four we dropped in an eight week comes calendar every Monday a 90 second win video from a different role.

231
00:13:18,800 --> 00:13:22,300
Wednesdays office hours focused on one Tuesday task.

232
00:13:22,300 --> 00:13:51,300
Friday's a tiny prompt pack plus two artifacts added to the library we measured adoption and time to draft weekly by week eight M.I.U. doubled in that lagging division no new tech just a metronome practical rule choose two channels and one ritual for example a team's announcements channel and an email roundup ritual equals 10 minute weekly copilot in the wild at the start of staff meetings consistency beets flare the average user needs to see the same pattern in the same place at the same time to believe it's real.

233
00:13:51,300 --> 00:14:20,300
What goes in the coms skip the adjectives show the delta before 42 minutes to draft a client recap after 14 minutes using this context block and this crit prompt here's the artifact link at a one line governance reminder when relevant sprinkle in micro stories just been shaved eight minutes by extracting action items live in the meeting we cut the recap time to zero repetition isn't boring it's how memory works office hours are where adoption compounds structure them start with a two minute mid bust then three five minute hot seats real time.

234
00:14:20,300 --> 00:14:49,300
Hot seats real tasks live prompting safety outputs to the library and with a one minute assignment try X prompt on Y task before Friday and post your artifact adults practice what they commit to publicly its group fitness for knowledge work common mistakes you'll avoid now changing the message every week pick your triad time to draft decision clarity meeting compression and hammer it measuring vanity metrics weapon our registrations aren't behavior track M.A.U. artifacts submissions and ticket volume trend if ticket spike after a coms change

235
00:14:49,300 --> 00:15:17,300
you broke something or finally got attention both are useful signals over delegating to it business leaders must be visible or the culture reverts to it tool not my tool is the reason this engine works is simple culture is a schedule plus stories the schedule makes it predictable the stories make it desirable run the stack for eight weeks and the program survives the hype cycle run it for 12 and the habit stick and yes will tie those habits to the 90 day plan soon

236
00:15:17,300 --> 00:15:46,300
for license sprawl and no adoption telemetry this is where budgets go to die you buy a heroic pile of SKU spray them across departments like confetti and then stare at a dashboard that might as well be a ransom note wrong roads wrong timing zero visibility the truth licenses are accelerants without targeting and telemetry you're just burning money faster segment by job to be done not org chart compare that to the average user distribution everyone gets one no stage entitlements by readiness and recurrence of Tuesday tasks give early access to the

237
00:15:46,300 --> 00:16:02,300
rolls with repeatable knowledge artifacts sales customer success pms HR ops put tourist rolls into a wait list fed by real wins scarcity creates focus and data you can act on now measure like an adult adoption may you tell you if humans even show up

238
00:16:02,300 --> 00:16:14,300
time to draft tells you if anything got faster ticket volume trend tells you if you broke workflows or finally earned attention if m.i.u. is flat your use cases are vague or your leaders are hiding if time to draft doesn't move your

239
00:16:14,300 --> 00:16:25,300
templates are weak if tickets spike investigate pattern clusters not anecdotes here's the game changer entitlement waves with checkpoints wave one gets 200 licenses plus a weekly readout on m.a.u.

240
00:16:25,300 --> 00:16:36,300
triad metrics and artifact submissions if usage clears the bar say sustained m.a.u. growth and at least 10 new artifacts per week expand to wave two if not pause fix then proceed this isn't

241
00:16:36,300 --> 00:17:04,300
punishment its control theory you don't open the throttle when the engines miss firing field reality one or three claimed 18% of licenses simply by correlating m.a.u. with artifact contributions and Tuesday task recurrence they re issued to power users and roll clusters with clear scenarios m.a.u. spike the next month without purchasing a single additional seat the software didn't change targeting did stop asking who wants a license and start asking who creates artifacts weekly measured by our triad that question pays for itself

242
00:17:04,300 --> 00:17:32,300
failure mode five skills gap and no workflow embedding even with licenses and cadence adoption stalls if people don't know how to talk to the machine or if the machine isn't where the work lives prompting isn't a vibe it's a skill and skills don't survive the copy paste gauntlet unless you embed them into the workflow like guardrails on a bridge teach the four moves context constraint critique continue c4 prompting yes you've heard me say it know your team hasn't mastered it context anchors to

243
00:17:32,300 --> 00:17:48,300
ground truth constraints said tone audience format and exclusions critique forces evaluation against your criteria continue targets the weakest section for iteration the reason this works is it mirrors how experts write fast drafts tight constraints

244
00:17:48,300 --> 00:18:03,300
ruthless edits now embed or it evaporates templates in word and loop with pre-filled context blocks outlook quick parts for meeting recaps teams message extensions that insert prompt starters share point pages with copy prompt buttons next to artifact libraries

245
00:18:03,300 --> 00:18:33,100
pint snippets and co-pilot saved prompts were fingers already click if they have to go hunting they want if it's one tap they will enter co-pilot studio for the repeatables build micro co pilots that codify winning chains for specific roles HR intake triage sales proposal outline pm risk register samarizer keep them narrow opinionated and grounded with curated content or connectors you actually govern then iterate based on usage not committee poetry field story HR was drowning in intake emails we stood up a studio bought that classifies

246
00:18:33,100 --> 00:18:55,100
requests draft first responses with policy safe language and assembles manager ready summaries ticket volume dropped and time to draft for common replies collapsed no it didn't replace judgment it replaced road typing and context gathering so humans could decide faster common mistakes to skip teaching prompt engineering as trivia teach c4 on Tuesday tasks parking libraries in random share point

247
00:18:55,100 --> 00:19:12,100
cul-de-sac put them where work happens building frankenbots that try to do everything microbeats mega that skipping feedback loops at a this saved me x minutes button and track it do this next published to prompt packs per role wire three templates into the apps ship one micro co pilot per pilot role and measure the triad

248
00:19:12,100 --> 00:19:25,100
one skills live in the workflow adoption stops depending on memory and starts depending on muscle 90 day co pilot adoption plan practitioner's playbook day one seven form the a i council and name your exact

249
00:19:25,100 --> 00:19:38,100
one business owner one operations leader one tech lead picked two roles for each role defined six Tuesday tasks using the 10 30 60 model and baseline time to draft publish the c4 prompt pattern and a plain English governance one

250
00:19:38,100 --> 00:19:57,100
schedule office hours and a weekly win slot now not later day eight or twenty one data readiness sprint map the top content surfaces fix labels on high traffic libraries enable DLP that catches obvious nonsense but doesn't block Tuesday work build a safe sandbox per role with sanitize examples approve templates and an artifact

251
00:19:57,100 --> 00:20:12,100
itself turn on telemetry and audit record of five minute mid-busting clip retrieval training day twenty two thirty train c4 prompting tied to the six tasks each attendee leaves with a prompt pack and two artifacts submitted launch office hours

252
00:20:12,100 --> 00:20:25,100
leaders do a five minute learn in public demo using their real work start the cadence Monday win Wednesday office hours Friday prompt pack day thirty one to forty five pilot life measure MAU

253
00:20:25,100 --> 00:20:45,100
draft meeting compression and ticket trend weekly capture before after artifacts with constraints and critique notes prune week prompts promote proven ones keep comes boringly consistent day forty six to sixty co pilot studio build one micro co pilot parole for the highest recurrence 30 60 task keeps go narrow ground with curated

254
00:20:45,100 --> 00:21:00,100
content add a saved me X minutes button iterate based on usage not taste day sixty one to seventy five license right sizing expand champions peer track and leader mentor track wave to entitlements only if MAU and artifact submissions are rising publish a

255
00:21:00,100 --> 00:21:14,100
change lock for governance tweaks and a short risk review day seventy six to ninety scale to two new roles using the same kid publish the internal playbook use cases prompt packs templates studio bots governance and metrics exec

256
00:21:14,100 --> 00:21:29,100
showcase real demos real artifacts real delta metrics to report weekly adoption MAU time to draft delta ticket volume delta now you try checklist who's your exact trio which two roles six Tuesday tasks

257
00:21:29,100 --> 00:21:39,100
where's the sandbox how you'll measure by Friday the game changer the week one decision that predicts MAU choose leaders who will demo their actual work live with co pilot record

258
00:21:39,100 --> 00:21:53,100
transcribe and publish their prompts and templates tie incentives every staff meeting must include one AI assisted artifact if leaders want model learning in public adoption stalls if they do MAU climbs because permission becomes visible weekly

259
00:21:53,100 --> 00:22:08,100
the one sentence take away and next move co pilot succeeds when culture engineers behavior into workflows measured by time to draft meeting compression and sustained MAU adopt the 90 day plan then subscribe for the advanced co pilot studio builds and real rollout war stories that make this stick.

260
00:22:08,100 --> 00:22:13,100
Next up, the playbook for scaling from two rolls to 10 without breaking trust.