Dynamics 365 Agents - Simply Explained
Key Takeaways
- Dynamics 365 Agents go beyond standard Copilot interactions by actively monitoring business events, following defined rules, and executing permitted workflows in the background without requiring direct user prompts.
- Unlike conversational chatbots, agents integrate deeply with connected business records across Microsoft Dataverse to handle end-to-end multi-step tasks across sales, customer service, finance, and supply chain operations.
- Robust guardrails and strict permissions are critical to ensure agents only access approved data, perform authorized actions, and successfully escalate complex or high-risk exceptions to human workers.
- Mirko Peters highlights how sales qualification and research agents can drastically reduce administrative friction, helping teams respond to incoming leads and prepare customer contexts more efficiently.
- Implementing AI agents successfully requires starting with a single, well-defined process, establishing clear governance boundaries, and ensuring underlying data quality is reliable before broader rollout.
Dynamics 365 Agents take AI beyond answering questions and generating drafts. Instead of waiting for someone to ask Copilot for help, agents can respond to defined business events, work with approved Dynamics 365 data, follow predefined rules, take permitted actions, and hand work to people when human judgment is required. In this episode of M365 FM, Mirko Peters explains how Dynamics 365 Agents work across sales, customer service, finance, supply chain, Business Central, Field Service, and Project Operations — and why permissions, guardrails, data quality, and human oversight are critical.
ㅤ
WHAT ARE DYNAMICS 365 AGENTS?
ㅤ
A Dynamics 365 Agent is AI configured to watch for a business event, understand approved information, follow defined rules, perform permitted actions, and report what happened.
The important difference is that an agent doesn't simply wait for a user to ask a question. A trigger such as a new lead, customer email, supplier response, order request, or financial mismatch can start the process automatically.
ㅤ
COPILOT VS DYNAMICS 365 AGENTS
ㅤ
Copilot typically works beside the user. You open a customer case and ask for a summary, request information, or generate a draft. The interaction begins because you asked Copilot to help.
An agent can continue working in the background according to its defined job. It can monitor an approved process, identify an event, collect relevant information, perform allowed actions, and escalate the work when necessary.
Think of Copilot as an assistant sitting beside you, while an agent is a specialized team member with a narrow job description, specific permissions, and clear rules.
ㅤ
HOW THE AI AGENT LOOP WORKS
ㅤ
The basic agent process can be understood as: notice, understand, act, check, and hand over.
The agent notices a defined event such as a new message or lead. It understands the situation using information and rules it has permission to access. It performs allowed actions, checks whether the result remains within its boundaries, and hands the work to a person when judgment or approval is required.
This creates supervised automation rather than unrestricted AI autonomy.
ㅤ
GUARDRAILS, PERMISSIONS AND HUMAN CONTROL
ㅤ
Guardrails define what an agent can see, change, create, or send. An organization might allow an agent to create a draft email but prevent it from sending that message automatically. An agent could categorize a customer case but be prevented from issuing a refund.
The same principle applies to supplier communication, financial transactions, customer promises, contracts, and other sensitive activities. The agent can prepare and route the work while people retain responsibility for higher-risk decisions.
ㅤ
DATAVERSE AND COPILOT STUDIO
ㅤ
Many Dynamics 365 applications use Microsoft Dataverse as the connected data foundation for customers, contacts, leads, cases, activities, products, and other business information.
For custom agents, Copilot Studio provides a place where organizations can define instructions, connect approved knowledge sources, configure actions, and determine how an agent should respond.
The quality of these agents depends heavily on the quality of the underlying data and processes.
ㅤ
SALES QUALIFICATION AGENTS
ㅤ
Sales agents can help close the gap between incoming interest and seller follow-up. A Sales Qualification Agent can research incoming leads, compare them against criteria established by the sales organization, prioritize potential opportunities, and prepare personalized outreach.
Instead of sellers spending significant time performing the first round of research for every new lead, the agent can prepare information while the salesperson decides whether the prospect deserves further attention.
ㅤ
SALES RESEARCH AND OPPORTUNITY AGENTS
ㅤ
Sales Research Agents can gather approved information before customer conversations, helping sellers understand companies, previous interactions, existing opportunities, and relevant account context.
Opportunity Agents operate later in the sales process. They can research active opportunities, surface deal context, identify potential risks, and highlight opportunities requiring attention.
Sales Close Agents can support the final stages through follow-ups, customer engagement, product suggestions, and handling common objections within defined business rules.
ㅤ
CUSTOMER SERVICE AGENTS
ㅤ
Dynamics 365 Customer Service Agents can reduce repetitive administrative work around customer cases. A Case Management Agent can interpret incoming customer communications, create cases, extract useful details, associate customers, and route cases appropriately.
Customer Intent Agents can identify why customers are contacting the organization and help direct requests toward appropriate queues, answers, or service representatives.
This can reduce the time agents spend manually creating and categorizing cases before they can begin solving the customer's actual problem.
ㅤ
KNOWLEDGE MANAGEMENT AND QUALITY AGENTS
ㅤ
A Customer Knowledge Management Agent can analyze completed cases to identify gaps in an organization's knowledge base and help prepare draft knowledge articles for human review.
Quality Evaluation Agents can evaluate larger numbers of customer interactions against standards established by service supervisors. This can help managers identify patterns and coaching opportunities without depending entirely on manually selected samples.
ㅤ
BUSINESS CENTRAL SALES ORDER AGENTS
ㅤ
In Dynamics 365 Business Central, a Sales Order Agent can process incoming customer order requests. It can read an order request received through email, identify the customer, extract requested products and quantities, and prepare order information for review.
People remain responsible for exceptions such as unusual pricing, missing products, incorrect customer information, or orders that don't match established patterns.
ㅤ
SUPPLY CHAIN AND SUPPLIER AGENTS
ㅤ
The Supplier Communications Agent in Dynamics 365 Supply Chain Management can support repetitive supplier follow-up activities.
It can request purchase order confirmations, ask suppliers for delivery updates, process incoming supplier communications, and associate responses with relevant orders. If a supplier reports a significant delay or another situation requiring a business decision, the agent can escalate the issue to a buyer or planner.
ㅤ
FINANCE AND RECONCILIATION AGENTS
ㅤ
Finance agents can support repetitive reconciliation work. An Account Reconciliation Agent can compare transactions and identify records that should match but don't, helping finance teams discover discrepancies earlier.
Financial Reconciliation Agents can support preparation and cleanup of information required for reconciliation and financial reporting. Accountants still determine what discrepancies mean and approve the appropriate resolution.
ㅤ
FIELD SERVICE AND PROJECT OPERATIONS AGENTS
ㅤ
Dynamics 365 Field Service can use scheduling-focused agents to help optimize technician schedules when appointments, availability, priorities, or other conditions change.
Project Operations can use agents to support administrative activities such as preparing time entries, processing expense information, and reviewing time, expense, and material records against organizational policies.
These capabilities target repetitive administrative work while keeping people responsible for exceptions and consequential decisions.
ㅤ
WHY DATA QUALITY MATTERS
ㅤ
An AI agent can only work with the information available to it. Duplicate customers, incomplete product records, outdated supplier contacts, and inconsistent business data can quickly reduce the reliability of automated processes.
Automation doesn't transform poor data into good decisions. Organizations need clean and consistent records before expecting agents to reliably connect incoming information with the correct customers, orders, cases, or suppliers.
ㅤ
HOW TO START WITH DYNAMICS 365 AGENTS
ㅤ
Start with one repetitive process that has clear steps, clear ownership, and a measurable result. Define what triggers the process, which records the agent can access, which actions it can perform, which actions require approval, and exactly when the agent must stop and involve a person.
Testing should include normal situations as well as difficult exceptions. Organizations should inspect what the agent reads, prepares, changes, and escalates before expanding its permissions or applying the approach to additional processes.
ㅤ
THE KEY TAKEAWAY
ㅤ
Dynamics 365 Agents turn repetitive business processes into supervised AI-powered actions. They can respond to events, understand approved business information, execute defined tasks, and escalate exceptions across sales, customer service, finance, operations, supply chain, and project workflows.
The objective isn't to give AI control of an entire department. Start with one clearly defined queue or process, establish permissions and guardrails, measure the result, and expand only after the process proves reliable.
An agent can only perform as well as the process, data, rules, and boundaries you give it.
Subscribe to M365 FM for more Microsoft Knowledge Nuggets and Simply Explained episodes covering Dynamics 365, Microsoft 365, Power Platform, Copilot, AI Agents, Azure, security, governance, and the Microsoft ecosystem.
Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support.
🚀 Want to be part of m365.fm?
Then stop just listening… and start showing up.
👉 Connect with me on LinkedIn and let’s make something happen:
- 🎙️ Be a podcast guest and share your story
- 🎧 Host your own episode (yes, seriously)
- 💡 Pitch topics the community actually wants to hear
- 🌍 Build your personal brand in the Microsoft 365 space
This isn’t just a podcast — it’s a platform for people who take action.
🔥 Most people wait. The best ones don’t.
👉 Connect with me on LinkedIn and send me a message:
"I want in"
Let’s build something awesome 👊
Frequently Asked Questions
What is a Dynamics 365 Agent?
A Dynamics 365 Agent is AI configured to watch for specific business events, review approved information, follow defined rules, take permitted actions, and escalate exceptions to human workers when judgment is required.
How do Dynamics 365 Agents differ from Copilot?
While Copilot acts as a real-time assistant sitting beside a user to answer questions or generate drafts on demand, agents operate autonomously in the background triggered by business events like new leads or customer emails.
What is the role of Dataverse in Dynamics 365 Agents?
Dataverse serves as the connected data foundation storing business information such as customers, contacts, leads, and cases, providing the essential knowledge base that agents rely on to understand context and execute tasks.
How are custom agents built for Dynamics 365?
Organizations use Copilot Studio to define custom agent instructions in plain English, connect approved knowledge sources, configure allowed actions, and establish boundaries for how the agent should respond.
00:00:00,000 --> 00:00:02,080
Copilot is great when you ask it a question,
2
00:00:02,080 --> 00:00:03,800
but what happens when nobody asks?
3
00:00:03,800 --> 00:00:06,800
Leads pile up, cases go unanswered, invoices sit,
4
00:00:06,800 --> 00:00:08,560
and supplier emails keep coming.
5
00:00:08,560 --> 00:00:11,160
That autumn is exactly where Dynamics 365 agents come in.
6
00:00:11,160 --> 00:00:13,720
I'm Mercopeter's from M365 FM.
7
00:00:13,720 --> 00:00:15,000
By the end of this knowledge nugget,
8
00:00:15,000 --> 00:00:16,560
you'll know what these agents actually do,
9
00:00:16,560 --> 00:00:18,840
where they work, and where people stay in control.
10
00:00:18,840 --> 00:00:20,400
So what makes an agent different?
11
00:00:20,400 --> 00:00:21,920
Think of it like an office building.
12
00:00:21,920 --> 00:00:24,320
Copilot sits right beside you at your desk,
13
00:00:24,320 --> 00:00:26,120
helping with whatever you're working on.
14
00:00:26,120 --> 00:00:28,600
An agent is trained staff working behind the scenes,
15
00:00:28,600 --> 00:00:30,520
only allowed into the rooms you approve.
16
00:00:30,520 --> 00:00:31,520
They don't wonder.
17
00:00:31,520 --> 00:00:35,600
The simple definition, what is a Dynamics 365 agent?
18
00:00:35,600 --> 00:00:38,360
We'll start by putting Dynamics 365 in plain English.
19
00:00:38,360 --> 00:00:41,000
Dynamics 365 is Microsoft's connected platform
20
00:00:41,000 --> 00:00:43,800
for running business work, oh, it holds customer records,
21
00:00:43,800 --> 00:00:45,840
sales leads, service cases, finance data,
22
00:00:45,840 --> 00:00:48,240
supplier details, orders, schedules, and much more.
23
00:00:48,240 --> 00:00:49,080
It isn't one app.
24
00:00:49,080 --> 00:00:50,680
A sales team works with leads,
25
00:00:50,680 --> 00:00:52,480
a service team works with cases,
26
00:00:52,480 --> 00:00:53,880
finance handles accounts,
27
00:00:53,880 --> 00:00:55,760
and supply chain manages orders.
28
00:00:55,760 --> 00:00:58,160
They use different apps, but the work all connects.
29
00:00:58,160 --> 00:01:00,800
A customer who buys something can start as a sales lead,
30
00:01:00,800 --> 00:01:02,680
become an account, then call support.
31
00:01:02,680 --> 00:01:03,920
They order history, payment history,
32
00:01:03,920 --> 00:01:07,160
and pass service tickets all matter when someone helps them.
33
00:01:07,160 --> 00:01:09,600
That connected data is where agents become useful.
34
00:01:09,600 --> 00:01:11,000
Here's the simplest definition.
35
00:01:11,000 --> 00:01:15,040
A Dynamics 365 agent is AI, set up to watch for a business event,
36
00:01:15,040 --> 00:01:17,800
look at approved information, follow approved rules,
37
00:01:17,800 --> 00:01:20,480
take allowed actions, and report what it did.
38
00:01:20,480 --> 00:01:22,200
Notice the word approved.
39
00:01:22,200 --> 00:01:24,400
An agent doesn't get free access to everything.
40
00:01:24,400 --> 00:01:26,680
People decide what it can read, what it can change,
41
00:01:26,680 --> 00:01:28,480
and when it must stop and ask for help.
42
00:01:28,480 --> 00:01:30,240
So is it just co-pilot with a new name?
43
00:01:30,240 --> 00:01:32,120
No, co-pilot helps you in the moment.
44
00:01:32,120 --> 00:01:33,880
You open a case, ask it to summarize,
45
00:01:33,880 --> 00:01:35,240
draft an email, find details,
46
00:01:35,240 --> 00:01:36,920
oh, it responds because you asked.
47
00:01:36,920 --> 00:01:38,920
But an agent keeps working after you step away.
48
00:01:38,920 --> 00:01:41,400
For example, an agent can watch an incoming queue,
49
00:01:41,400 --> 00:01:43,800
notice a new request, check related records,
50
00:01:43,800 --> 00:01:45,600
create a draft response, update a case,
51
00:01:45,600 --> 00:01:47,360
or send the work to the right person.
52
00:01:47,360 --> 00:01:48,680
That work happens from a trigger,
53
00:01:48,680 --> 00:01:50,440
not because someone clicked a button.
54
00:01:50,440 --> 00:01:52,480
Co-pilot is a helpful colleague beside you.
55
00:01:52,480 --> 00:01:55,160
An agent is a trained team member with a narrow job,
56
00:01:55,160 --> 00:01:58,040
a written rule book, and a manager checking the results.
57
00:01:58,040 --> 00:01:59,880
Now you might think, isn't that just a chatbot?
58
00:01:59,880 --> 00:02:02,680
A chatbot mainly talks, oh, ooh, you type a question,
59
00:02:02,680 --> 00:02:04,840
it gives an answer or points you to an article.
60
00:02:04,840 --> 00:02:07,680
That helps, but the conversation is usually the whole job.
61
00:02:07,680 --> 00:02:10,200
An agent uses that conversation as a starting point,
62
00:02:10,200 --> 00:02:12,160
then works with business records and processes
63
00:02:12,160 --> 00:02:13,160
behind the scenes.
64
00:02:13,160 --> 00:02:15,240
It might read an incoming customer email,
65
00:02:15,240 --> 00:02:18,160
identify the customer, connect the message to an open case,
66
00:02:18,160 --> 00:02:20,040
check and approved knowledge article,
67
00:02:20,040 --> 00:02:21,760
and hand the case to a specialist
68
00:02:21,760 --> 00:02:23,560
if the issue needs a person.
69
00:02:23,560 --> 00:02:25,960
That's a bigger job than answering a question.
70
00:02:25,960 --> 00:02:27,320
The agent loop stays simple.
71
00:02:27,320 --> 00:02:31,040
Notice, understand, act, check, then hand over when needed.
72
00:02:31,040 --> 00:02:33,240
Notice means it sees a defined event,
73
00:02:33,240 --> 00:02:34,560
like a new lead or message.
74
00:02:34,560 --> 00:02:37,320
Understand means it reads the data and rules it can use.
75
00:02:37,320 --> 00:02:38,880
Act means it takes allowed steps
76
00:02:38,880 --> 00:02:41,080
like creating a record or preparing an email,
77
00:02:41,080 --> 00:02:43,160
check confirms if the work meets the rules,
78
00:02:43,160 --> 00:02:45,600
and if the task crosses a limit, looks unusual,
79
00:02:45,600 --> 00:02:48,240
or needs judgment, it hands the work to a person.
80
00:02:48,240 --> 00:02:50,440
An agent isn't an independent employee, oh,
81
00:02:50,440 --> 00:02:52,320
it doesn't remove accountability.
82
00:02:52,320 --> 00:02:54,800
The sales rep still owns the customer relationship,
83
00:02:54,800 --> 00:02:56,560
the service manager owns service standards,
84
00:02:56,560 --> 00:02:58,920
the finance team owns financial decisions.
85
00:02:58,920 --> 00:03:01,320
People set the goal, define the boundaries,
86
00:03:01,320 --> 00:03:04,160
approve sensitive actions, and review the outcome.
87
00:03:04,160 --> 00:03:06,320
Now let's follow one incoming customer message
88
00:03:06,320 --> 00:03:08,320
and see how that loop works in practice.
89
00:03:08,320 --> 00:03:13,320
From message to action, how an agent works behind the scenes.
90
00:03:13,320 --> 00:03:14,360
Picture this.
91
00:03:14,360 --> 00:03:17,560
A customer sends an email at 8.15 in the morning.
92
00:03:17,560 --> 00:03:19,520
My delivery has now attempt arrived.
93
00:03:19,520 --> 00:03:21,360
Can you tell me where it is?
94
00:03:21,360 --> 00:03:24,480
Without an agent, that message sits in a shared inbox.
95
00:03:24,480 --> 00:03:26,720
Someone has to open it, read it, find the customer,
96
00:03:26,720 --> 00:03:28,520
check the order and create a case.
97
00:03:28,520 --> 00:03:31,720
With an agent, the incoming email starts the process right away.
98
00:03:31,720 --> 00:03:33,840
The trigger is the event that wakes the agent up.
99
00:03:33,840 --> 00:03:35,200
That could be a new email or web form,
100
00:03:35,200 --> 00:03:37,560
a new sales lead, a supplier missing a delivery date,
101
00:03:37,560 --> 00:03:40,200
or a finance transaction that does no TimT match.
102
00:03:40,200 --> 00:03:42,440
The agent does know, watch everything in the company.
103
00:03:42,440 --> 00:03:44,640
It only watches the queue, record, mailbox,
104
00:03:44,640 --> 00:03:46,760
or process you gave it permission to watch.
105
00:03:46,760 --> 00:03:48,960
Once the email arrives, it needs context.
106
00:03:48,960 --> 00:03:50,960
A message that says, where is my order?
107
00:03:50,960 --> 00:03:52,520
Means very little on its own.
108
00:03:52,520 --> 00:03:55,560
The agent needs to find the sender out to him as customer record.
109
00:03:55,560 --> 00:03:58,160
Look for recent orders, check delivery details,
110
00:03:58,160 --> 00:04:01,160
read earlier messages and search approved help articles.
111
00:04:01,160 --> 00:04:03,800
That context turns a vague message into a work item.
112
00:04:03,800 --> 00:04:05,360
The agent might find one open order
113
00:04:05,360 --> 00:04:07,600
that left the warehouse yesterday with a carrier delay,
114
00:04:07,600 --> 00:04:09,840
or it might find three orders and no clear match.
115
00:04:09,840 --> 00:04:11,160
Those are very different situations.
116
00:04:11,160 --> 00:04:14,120
Based on the rules you set, the agent takes the next step.
117
00:04:14,120 --> 00:04:16,600
For a simple case, it creates a service case,
118
00:04:16,600 --> 00:04:18,440
fills in the subject and description,
119
00:04:18,440 --> 00:04:20,480
connects it to the right customer and order,
120
00:04:20,480 --> 00:04:24,200
then prepares a reply using the approved delivery information.
121
00:04:24,200 --> 00:04:27,520
For a less clear case, it asks the customer for an order number.
122
00:04:27,520 --> 00:04:29,240
If the customer has an urgent contract,
123
00:04:29,240 --> 00:04:31,760
the agent roots the case to a priority queue,
124
00:04:31,760 --> 00:04:34,040
and when a delay breaks a customer promise,
125
00:04:34,040 --> 00:04:36,200
it hands the work to a person instead of trying
126
00:04:36,200 --> 00:04:38,560
to smooth it over with an automated reply.
127
00:04:38,560 --> 00:04:39,880
That last part matters a lot.
128
00:04:39,880 --> 00:04:42,440
An agent should never guess its way through a situation
129
00:04:42,440 --> 00:04:45,240
where the wrong answer costs money, harms a relationship,
130
00:04:45,240 --> 00:04:48,000
or creates a promise your company cannot keep.
131
00:04:48,000 --> 00:04:49,440
So, guardrails come into play.
132
00:04:49,440 --> 00:04:51,240
Guardrails are the limits around the job.
133
00:04:51,240 --> 00:04:53,360
They include the data and agent can see,
134
00:04:53,360 --> 00:04:55,960
the records it can edit, the messages it can send,
135
00:04:55,960 --> 00:04:57,360
the rules it must follow,
136
00:04:57,360 --> 00:05:00,000
and the points where a person must approve its next move.
137
00:05:00,000 --> 00:05:02,640
For example, you might allow an agent to create a draft email
138
00:05:02,640 --> 00:05:03,440
but not send it.
139
00:05:03,440 --> 00:05:05,400
You might let it update a case category,
140
00:05:05,400 --> 00:05:06,960
but not issue a refund.
141
00:05:06,960 --> 00:05:09,520
You might let it chase a supplier for a delivery update
142
00:05:09,520 --> 00:05:13,080
while requiring a buyer to approve any change to a purchase order.
143
00:05:13,080 --> 00:05:15,720
The agent/autemus actions can also be recorded,
144
00:05:15,720 --> 00:05:18,040
so people can inspect what happened later.
145
00:05:18,040 --> 00:05:20,160
That gives managers a way to review results,
146
00:05:20,160 --> 00:05:22,840
fix poor rules, and answer a simple question.
147
00:05:22,840 --> 00:05:24,840
Why did this case go to that queue?
148
00:05:24,840 --> 00:05:26,920
Behind much of this work sits Dataverse.
149
00:05:26,920 --> 00:05:28,560
Dataverse is the shared data store
150
00:05:28,560 --> 00:05:32,240
used by many Dynamics 365 apps and Power Platform apps.
151
00:05:32,240 --> 00:05:34,000
Think of it as a digital filing cabinet
152
00:05:34,000 --> 00:05:36,280
where business records sit in connected drawers.
153
00:05:36,280 --> 00:05:38,080
Customers, contacts, leads, cases,
154
00:05:38,080 --> 00:05:40,040
activities, products, and more.
155
00:05:40,040 --> 00:05:43,000
An agent can only work well when it can find the right drawer
156
00:05:43,000 --> 00:05:44,640
and read the right file.
157
00:05:44,640 --> 00:05:47,040
If customer records contain duplicate names,
158
00:05:47,040 --> 00:05:49,600
missing email addresses or older details,
159
00:05:49,600 --> 00:05:51,040
the agent has less to work with.
160
00:05:51,040 --> 00:05:52,440
It may still spot a message,
161
00:05:52,440 --> 00:05:54,440
but it can now attempt to reliably connect it
162
00:05:54,440 --> 00:05:56,040
to the right customer or action.
163
00:05:56,040 --> 00:05:58,840
For Custom Agents, co-pilot studio is where teams build
164
00:05:58,840 --> 00:05:59,760
and manage the job.
165
00:05:59,760 --> 00:06:01,800
You define the instructions in plain English,
166
00:06:01,800 --> 00:06:04,560
connect approved knowledge sources, add actions,
167
00:06:04,560 --> 00:06:06,080
and decide how the agent responds
168
00:06:06,080 --> 00:06:07,520
when it can have them to continue.
169
00:06:07,520 --> 00:06:09,800
It totems where the job description becomes real.
170
00:06:09,800 --> 00:06:12,440
So the path for message to action is not metymagic.
171
00:06:12,440 --> 00:06:13,560
A trigger starts the work,
172
00:06:13,560 --> 00:06:16,400
a proof data gives it context, rules control its choices,
173
00:06:16,400 --> 00:06:18,200
and guardrails limit its reach.
174
00:06:18,200 --> 00:06:20,880
Then a person takes over whenever the work needs judgment.
175
00:06:20,880 --> 00:06:22,720
With those building blocks in place,
176
00:06:22,720 --> 00:06:24,480
we can look at the different jobs agents
177
00:06:24,480 --> 00:06:26,480
can handle across the business.
178
00:06:26,480 --> 00:06:29,400
Sales agents, turning interest into follow-up.
179
00:06:29,400 --> 00:06:31,240
Sales work often breaks down right between interest
180
00:06:31,240 --> 00:06:32,240
and follow-up.
181
00:06:32,240 --> 00:06:34,000
A person fills in a form, downloads a guide,
182
00:06:34,000 --> 00:06:35,200
or sends a message.
183
00:06:35,200 --> 00:06:37,560
The lead enters Dynamics 365 Sales,
184
00:06:37,560 --> 00:06:39,800
but the seller has meetings, existing customers,
185
00:06:39,800 --> 00:06:41,960
and a long list of records waiting for attention.
186
00:06:41,960 --> 00:06:44,920
That e-wyroms where sales agents take on a focused job.
187
00:06:44,920 --> 00:06:47,960
The sales qualification agent works at the start of the process.
188
00:06:47,960 --> 00:06:49,600
It researches incoming leads,
189
00:06:49,600 --> 00:06:52,080
compares them with the fit rules your sales team sets,
190
00:06:52,080 --> 00:06:54,440
and helps rank which leads to the service attention first.
191
00:06:54,440 --> 00:06:56,080
It also prepares personal outreach.
192
00:06:56,080 --> 00:06:58,720
Imagine a lead arrives from a manufacturing company.
193
00:06:58,720 --> 00:07:01,680
The agent checks the details already held in Dynamics 365,
194
00:07:01,680 --> 00:07:03,240
looks at the information it how,
195
00:07:03,240 --> 00:07:04,560
team cars allowed to use,
196
00:07:04,560 --> 00:07:06,360
and prepares an email based on the lead-o,
197
00:07:06,360 --> 00:07:08,240
TMS company and interest.
198
00:07:08,240 --> 00:07:10,880
The seller can review that message, change it, send it,
199
00:07:10,880 --> 00:07:12,400
or take over the conversation.
200
00:07:12,400 --> 00:07:15,280
The purpose is now to remove the seller.
201
00:07:15,280 --> 00:07:17,600
It attempts to stop good leads from sitting untouched
202
00:07:17,600 --> 00:07:19,240
because someone has noticed it had time
203
00:07:19,240 --> 00:07:21,080
to do the first round of research.
204
00:07:21,080 --> 00:07:23,680
Microsoft shared one early example from Sandvik Coromant
205
00:07:23,680 --> 00:07:25,440
in its first three weeks of a pilot.
206
00:07:25,440 --> 00:07:28,120
The company reported more than 120 hours saved
207
00:07:28,120 --> 00:07:29,840
and $19,000 saved.
208
00:07:29,840 --> 00:07:33,040
It also forecast a 5% revenue increase after a full rollout.
209
00:07:33,040 --> 00:07:35,960
Those numbers belong to one pilot, not every sales team.
210
00:07:35,960 --> 00:07:39,120
Still, they point to a simple problem agents can address.
211
00:07:39,120 --> 00:07:41,600
Sellers waste a lot of time looking for background details
212
00:07:41,600 --> 00:07:43,680
before they can start a useful conversation.
213
00:07:43,680 --> 00:07:46,080
The sales research agent focuses on that preparation work.
214
00:07:46,080 --> 00:07:47,880
Before a call, a seller wants answers.
215
00:07:47,880 --> 00:07:48,720
Who is this company?
216
00:07:48,720 --> 00:07:50,080
Which products do they use?
217
00:07:50,080 --> 00:07:52,200
What happened in earlier emails and meetings?
218
00:07:52,200 --> 00:07:54,120
Is there an open opportunity already?
219
00:07:54,120 --> 00:07:56,520
Finding that information can mean opening several records
220
00:07:56,520 --> 00:07:58,360
and searching through old activity notes.
221
00:07:58,360 --> 00:08:01,080
A research agent brings together the approved context,
222
00:08:01,080 --> 00:08:03,000
so the seller starts the call better prepared.
223
00:08:03,000 --> 00:08:04,840
Then you have the sales opportunity agent.
224
00:08:04,840 --> 00:08:07,360
When a lead becomes an active deal, the work changes.
225
00:08:07,360 --> 00:08:10,760
The question shifts from, should we contact this person?
226
00:08:10,760 --> 00:08:13,560
To what needs to happen next and where could this deal stall?
227
00:08:13,560 --> 00:08:15,880
The opportunity agent researches the opportunity,
228
00:08:15,880 --> 00:08:18,800
brings forward deal context, points out emerging risks,
229
00:08:18,800 --> 00:08:20,960
and helps the seller see promising opportunities
230
00:08:20,960 --> 00:08:22,160
that need attention.
231
00:08:22,160 --> 00:08:24,560
It does not admit decide whether to give a discount.
232
00:08:24,560 --> 00:08:26,560
Instead, it gives the seller a clearer picture
233
00:08:26,560 --> 00:08:28,760
before that decision.
234
00:08:28,760 --> 00:08:31,240
The sales-close agent sits later in the journey.
235
00:08:31,240 --> 00:08:33,920
It supports moving a deal to what a finished transaction,
236
00:08:33,920 --> 00:08:36,360
including customer engagement, follow-ups, product
237
00:08:36,360 --> 00:08:38,080
suggestions, and handling common objections
238
00:08:38,080 --> 00:08:39,720
within the rules the business sets.
239
00:08:39,720 --> 00:08:41,200
Think about a normal flow.
240
00:08:41,200 --> 00:08:42,880
A new lead arrives on Monday morning.
241
00:08:42,880 --> 00:08:44,520
The qualification agent checks the lead
242
00:08:44,520 --> 00:08:47,320
against your fit criteria and prepares outreach.
243
00:08:47,320 --> 00:08:49,800
The seller reviews the result, starts a real conversation
244
00:08:49,800 --> 00:08:51,720
and learns what the customer needs.
245
00:08:51,720 --> 00:08:53,920
As the deal develops, the opportunity agent helps
246
00:08:53,920 --> 00:08:56,800
keep the record current and brings attention to next steps.
247
00:08:56,800 --> 00:08:59,760
Near the end, the close agent helps reduce the small delays
248
00:08:59,760 --> 00:09:01,760
that often leave a deal sitting still.
249
00:09:01,760 --> 00:09:04,280
But the seller remains responsible for the relationship.
250
00:09:04,280 --> 00:09:06,480
People read the room, understand a customer out,
251
00:09:06,480 --> 00:09:10,080
tempts history, negotiate pricing, make promises,
252
00:09:10,080 --> 00:09:12,280
and decide when an exception makes sense.
253
00:09:12,280 --> 00:09:15,280
An agent can prepare work and keep follow-ups from slipping,
254
00:09:15,280 --> 00:09:17,800
but it can autehim own the outcome.
255
00:09:17,800 --> 00:09:20,920
Sales agents deal with people who may become customers.
256
00:09:20,920 --> 00:09:22,920
Service agents handle a different moment
257
00:09:22,920 --> 00:09:25,480
when a customer already needs help.
258
00:09:25,480 --> 00:09:27,640
Customer service agents, keeping cases
259
00:09:27,640 --> 00:09:29,680
from falling through the cracks.
260
00:09:29,680 --> 00:09:31,760
Customer service brings a different kind of pressure
261
00:09:31,760 --> 00:09:32,680
than sales.
262
00:09:32,680 --> 00:09:34,520
A sales lead can wait a while, but a customer
263
00:09:34,520 --> 00:09:37,760
with a broken order or a billing problem already expects an answer,
264
00:09:37,760 --> 00:09:39,920
which means your service team spends a lot of time
265
00:09:39,920 --> 00:09:43,080
reading messages, creating cases, sorting cues,
266
00:09:43,080 --> 00:09:46,360
and checking whether the customer actually got help.
267
00:09:46,360 --> 00:09:48,720
The case management agent handles some of that early admin
268
00:09:48,720 --> 00:09:51,600
work when a customer sends an email, starts a chat,
269
00:09:51,600 --> 00:09:53,760
or reaches out through another digital channel,
270
00:09:53,760 --> 00:09:56,000
this agent steps in to create a case
271
00:09:56,000 --> 00:09:59,080
and pull the useful details from the message A.
272
00:09:59,080 --> 00:10:01,680
It identifies the subject, adds a description,
273
00:10:01,680 --> 00:10:03,680
connects the customer record, and places
274
00:10:03,680 --> 00:10:06,320
the case in the right category.
275
00:10:06,320 --> 00:10:08,720
A service rep then starts with a usable case
276
00:10:08,720 --> 00:10:11,240
instead of a blank form, picture a customer writing,
277
00:10:11,240 --> 00:10:14,080
how I was charged twice for my subscription, A.O.
278
00:10:14,080 --> 00:10:16,960
The agent recognizes this as a billing issue,
279
00:10:16,960 --> 00:10:18,760
attaches the message to a new case,
280
00:10:18,760 --> 00:10:20,560
and roots it to the payments team.
281
00:10:20,560 --> 00:10:23,240
The rep still checks the account and decides what to do,
282
00:10:23,240 --> 00:10:25,040
but they don't need to copy the same details
283
00:10:25,040 --> 00:10:26,880
into five fields first.
284
00:10:26,880 --> 00:10:28,920
The customer intent agent looks for patterns
285
00:10:28,920 --> 00:10:30,680
across your customer conversations.
286
00:10:30,680 --> 00:10:33,320
Intent simply means the reason someone contacted you
287
00:10:33,320 --> 00:10:36,000
are they asking about a delivery, reporting a fault,
288
00:10:36,000 --> 00:10:38,280
trying to cancel or looking for an invoice.
289
00:10:38,280 --> 00:10:39,680
When the system spots that reason,
290
00:10:39,680 --> 00:10:41,520
it directs the request toward the right queue,
291
00:10:41,520 --> 00:10:43,160
answer path, or service rep.
292
00:10:43,160 --> 00:10:46,240
It also finds new reasons customers keep contacting support.
293
00:10:46,240 --> 00:10:47,560
Maybe after a product update,
294
00:10:47,560 --> 00:10:49,880
customers suddenly ask the same question.
295
00:10:49,880 --> 00:10:52,160
Your help center has no article for yet,
296
00:10:52,160 --> 00:10:54,480
which tells the service team something needs attention
297
00:10:54,480 --> 00:10:56,520
before the inbox fills up.
298
00:10:56,520 --> 00:10:58,920
Next up is the customer knowledge management agent.
299
00:10:58,920 --> 00:11:01,400
Support teams often know the answer to a common question,
300
00:11:01,400 --> 00:11:04,560
but that answer lives in old case notes, chat transcripts,
301
00:11:04,560 --> 00:11:06,680
or inside one experienced person's head.
302
00:11:06,680 --> 00:11:08,360
This agent looks at completed cases
303
00:11:08,360 --> 00:11:10,000
to find gaps in your knowledge base
304
00:11:10,000 --> 00:11:12,360
and helps create draft articles for review.
305
00:11:12,360 --> 00:11:14,840
The word draft matters who a person should check the article
306
00:11:14,840 --> 00:11:16,840
before it becomes customer facing content,
307
00:11:16,840 --> 00:11:19,160
especially when it covers refunds, contracts,
308
00:11:19,160 --> 00:11:21,120
safety, or product instructions.
309
00:11:21,120 --> 00:11:22,800
But the agent spots the repeated question
310
00:11:22,800 --> 00:11:24,640
and brings the raw material together,
311
00:11:24,640 --> 00:11:27,120
rather than leaving someone to find it months later,
312
00:11:27,120 --> 00:11:29,560
the quality evaluation agent works at another level.
313
00:11:29,560 --> 00:11:32,400
Most service managers only review a small sample of calls,
314
00:11:32,400 --> 00:11:35,120
chats, and emails because manual reviews take time,
315
00:11:35,120 --> 00:11:37,680
but this agent assesses more customer interactions
316
00:11:37,680 --> 00:11:39,880
against the standard supervisors define.
317
00:11:39,880 --> 00:11:42,360
Maybe the team checks whether the rep confirmed the issue,
318
00:11:42,360 --> 00:11:44,840
used approved wording, recorded the right details,
319
00:11:44,840 --> 00:11:46,520
or gave a clear next step.
320
00:11:46,520 --> 00:11:48,400
The agent helps supervisors see patterns
321
00:11:48,400 --> 00:11:50,240
across a larger set of conversations,
322
00:11:50,240 --> 00:11:53,840
so coaching focuses on real gaps instead of random samples.
323
00:11:53,840 --> 00:11:56,360
Copilot and agents work side-by-side in service.
324
00:11:56,360 --> 00:11:59,000
A service rep might use copilot during a live case
325
00:11:59,000 --> 00:12:01,640
to summarize a long conversation, search approved knowledge,
326
00:12:01,640 --> 00:12:03,080
or draft a reply.
327
00:12:03,080 --> 00:12:05,280
Meanwhile, service agents keep background processes
328
00:12:05,280 --> 00:12:08,400
moving out creating cases, finding intent, spotting knowledge
329
00:12:08,400 --> 00:12:10,360
gaps, and reviewing quality.
330
00:12:10,360 --> 00:12:12,160
Imagine a customer asks a question that
331
00:12:12,160 --> 00:12:14,680
keeps appearing after a new product release.
332
00:12:14,680 --> 00:12:16,800
The intent agent recognizes the type of question
333
00:12:16,800 --> 00:12:18,360
and the system either points the customer
334
00:12:18,360 --> 00:12:20,120
toward the right approved answer
335
00:12:20,120 --> 00:12:24,120
or roots the case to the right team when self-service won't solve it.
336
00:12:24,120 --> 00:12:25,600
Later, the knowledge agent notices
337
00:12:25,600 --> 00:12:27,240
the same question across many cases
338
00:12:27,240 --> 00:12:29,520
and prepares material for a better help article.
339
00:12:29,520 --> 00:12:32,440
Not every customer problem belongs in an automated path
340
00:12:32,440 --> 00:12:35,280
because of frustrated customer and unusual fault,
341
00:12:35,280 --> 00:12:36,880
or a complaint that needs empathy
342
00:12:36,880 --> 00:12:38,720
should reach a person quickly.
343
00:12:38,720 --> 00:12:41,080
Agents help clear routine work from the path,
344
00:12:41,080 --> 00:12:42,960
so people can spend more time on the cases
345
00:12:42,960 --> 00:12:44,720
where their judgment changes the outcome.
346
00:12:44,720 --> 00:12:47,200
That same pattern reaches beyond the customer inbox
347
00:12:47,200 --> 00:12:50,160
into order supplier follow-ups, schedules, and finance
348
00:12:50,160 --> 00:12:52,920
work that teams often spend their day chasing.
349
00:12:52,920 --> 00:12:55,920
Finance, operations, and supply chain agents,
350
00:12:55,920 --> 00:12:58,040
the work nobody wants to chase.
351
00:12:58,040 --> 00:13:00,720
Some business work doesn't arrive as a customer case or a sales
352
00:13:00,720 --> 00:13:01,240
lead.
353
00:13:01,240 --> 00:13:02,960
It arrives as an order email that needs
354
00:13:02,960 --> 00:13:05,160
typing into a system, a supplier who hasn't
355
00:13:05,160 --> 00:13:08,320
confirmed a delivery, a time sheet sitting unapproved,
356
00:13:08,320 --> 00:13:11,560
or numbers that don't match near the end of a finance period.
357
00:13:11,560 --> 00:13:14,600
These jobs are often repetitive, but they still affect cash,
358
00:13:14,600 --> 00:13:17,040
customer promises, and daily operations.
359
00:13:17,040 --> 00:13:18,800
In Dynamics 365 Business Central,
360
00:13:18,800 --> 00:13:21,640
the sales order agent helps with incoming customer orders.
361
00:13:21,640 --> 00:13:23,760
It reads an order request sent by email,
362
00:13:23,760 --> 00:13:25,680
finds the customer in Business Central,
363
00:13:25,680 --> 00:13:28,040
pulls out the requested items and quantities,
364
00:13:28,040 --> 00:13:31,000
and prepares the order information for review.
365
00:13:31,000 --> 00:13:34,440
Before lunch, your team might have 20 order emails to process.
366
00:13:34,440 --> 00:13:37,240
Without help, somebody opens each message, checks who sent it,
367
00:13:37,240 --> 00:13:39,720
searches for products, copies the details into an order,
368
00:13:39,720 --> 00:13:41,800
and checks for missing information.
369
00:13:41,800 --> 00:13:44,520
The agent handles the first pass, while a person reviews
370
00:13:44,520 --> 00:13:47,080
exceptions, unusual pricing, missing products,
371
00:13:47,080 --> 00:13:49,760
or any order that doesn't match the customer's normal pattern.
372
00:13:49,760 --> 00:13:52,520
That means your team spends less time retiping information
373
00:13:52,520 --> 00:13:54,600
and more time dealing with the order that actually
374
00:13:54,600 --> 00:13:55,560
needs attention.
375
00:13:55,560 --> 00:13:58,120
Supply chain teams face a similar problem with suppliers.
376
00:13:58,120 --> 00:14:01,120
A buyer sends a purchase order, waits for confirmation,
377
00:14:01,120 --> 00:14:03,600
then chases the supplier if no reply comes,
378
00:14:03,600 --> 00:14:05,400
and when a response finally arrives,
379
00:14:05,400 --> 00:14:08,680
somebody needs to read it, connected to the right purchase order,
380
00:14:08,680 --> 00:14:10,360
and decide what comes next.
381
00:14:10,360 --> 00:14:13,160
The supplier communications agent helps with that follow-up work
382
00:14:13,160 --> 00:14:15,880
inside Dynamics 365 Supply Chain Management.
383
00:14:15,880 --> 00:14:18,800
It sends reminders asking suppliers to confirm purchase orders
384
00:14:18,800 --> 00:14:20,280
or provide delivery updates,
385
00:14:20,280 --> 00:14:23,720
sorts incoming supplier emails, matches them with the right order,
386
00:14:23,720 --> 00:14:26,640
and responds within the rules the purchasing team sets.
387
00:14:26,640 --> 00:14:28,880
A buyer can allow the agent to send a polite reminder
388
00:14:28,880 --> 00:14:31,960
after a set number of days, but if a supplier reports a delay
389
00:14:31,960 --> 00:14:33,800
that could stop production, the agent
390
00:14:33,800 --> 00:14:36,200
should send that issue to a buyer or planner,
391
00:14:36,200 --> 00:14:39,120
who can then decide whether to contact another supplier,
392
00:14:39,120 --> 00:14:42,520
change a schedule, or speak directly with the customer.
393
00:14:42,520 --> 00:14:44,560
Finance teams have their own pile of work,
394
00:14:44,560 --> 00:14:47,040
and much of it appears around reconciliation,
395
00:14:47,040 --> 00:14:49,480
an account reconciliation agent in Dynamics.
396
00:14:49,480 --> 00:14:51,880
365 Finance helps compare transactions
397
00:14:51,880 --> 00:14:55,360
between sub-ledgers and the general ledger, in plain English.
398
00:14:55,360 --> 00:14:58,000
It looks for records that should match, but don't.
399
00:14:58,000 --> 00:15:00,600
Instead of discovering a mismatch only at period end,
400
00:15:00,600 --> 00:15:02,720
the agent brings the difference forward earlier,
401
00:15:02,720 --> 00:15:04,360
suggests ways to resolve it,
402
00:15:04,360 --> 00:15:07,880
and supports the work of matching and clearing transactions.
403
00:15:07,880 --> 00:15:10,240
Accountants still decide what a difference means.
404
00:15:10,240 --> 00:15:13,000
A mismatch might come from timing, from a missing entry,
405
00:15:13,000 --> 00:15:16,200
or it might point to a real problem that needs investigation.
406
00:15:16,200 --> 00:15:18,480
The agent helps find and sort those differences,
407
00:15:18,480 --> 00:15:20,520
but it doesn't replace the financial judgment
408
00:15:20,520 --> 00:15:21,920
behind the final decision.
409
00:15:21,920 --> 00:15:24,400
The financial reconciliation agent supports a related part
410
00:15:24,400 --> 00:15:26,800
of the process by helping teams prepare and clean data
411
00:15:26,800 --> 00:15:29,840
for the reconciliation work that leads into financial reporting.
412
00:15:29,840 --> 00:15:32,320
Anyone who has spent hours trying to make spreadsheet data
413
00:15:32,320 --> 00:15:34,560
consistent will understand the appeal.
414
00:15:34,560 --> 00:15:35,840
The work still needs review,
415
00:15:35,840 --> 00:15:37,520
but the agent reduces the manual sorting
416
00:15:37,520 --> 00:15:39,960
before the finance team begins its deeper checks.
417
00:15:39,960 --> 00:15:42,280
Field service has a different kind of moving target.
418
00:15:42,280 --> 00:15:44,400
Dispatchers need to schedule technicians, equipment,
419
00:15:44,400 --> 00:15:46,120
and customer visits while dealing with traffic,
420
00:15:46,120 --> 00:15:48,320
cancellations, double bookings, and urgent requests
421
00:15:48,320 --> 00:15:49,480
that appear during the day.
422
00:15:49,480 --> 00:15:52,400
The scheduling operations agent helps improve those schedules
423
00:15:52,400 --> 00:15:54,680
based on the goals and rules of business sets,
424
00:15:54,680 --> 00:15:56,480
and it supports schedule changes
425
00:15:56,480 --> 00:15:59,200
when plans no longer fit the real day.
426
00:15:59,200 --> 00:16:01,720
A dispatcher still knows which customer needs extra care,
427
00:16:01,720 --> 00:16:03,280
which technician has the right experience,
428
00:16:03,280 --> 00:16:05,840
and when a promise matters more than the shortest travel route.
429
00:16:05,840 --> 00:16:07,120
The agent helps with the reshuffling,
430
00:16:07,120 --> 00:16:09,080
but the dispatcher owns the decision.
431
00:16:09,080 --> 00:16:11,480
Project operations also includes agents for admin work
432
00:16:11,480 --> 00:16:12,800
that often waits in queues.
433
00:16:12,800 --> 00:16:15,120
Time entry agents prepare draft time entries
434
00:16:15,120 --> 00:16:17,280
from project assignments, bookings, and past patterns
435
00:16:17,280 --> 00:16:19,480
while expense agents pull details from receipts
436
00:16:19,480 --> 00:16:21,000
and prepare expense entries,
437
00:16:21,000 --> 00:16:22,680
and approvals agents review time,
438
00:16:22,680 --> 00:16:25,600
expense, and material entries against policy documents
439
00:16:25,600 --> 00:16:27,880
before they reach a project manager.
440
00:16:27,880 --> 00:16:29,040
The result is simple.
441
00:16:29,040 --> 00:16:31,480
Fewer details copied from one place to another,
442
00:16:31,480 --> 00:16:34,200
fewer follow-up emails that someone forgets to send,
443
00:16:34,200 --> 00:16:36,320
and more time for the exceptions that need a person
444
00:16:36,320 --> 00:16:38,280
to stop, think, and decide.
445
00:16:38,280 --> 00:16:40,600
Still, none of these agents can fix a process
446
00:16:40,600 --> 00:16:43,480
that nobody understands, or data that nobody trusts.
447
00:16:43,480 --> 00:16:46,440
The parts you must put in place first.
448
00:16:46,440 --> 00:16:47,520
So here's the thing.
449
00:16:47,520 --> 00:16:50,120
Before you turn on an agent, pick one repeated job.
450
00:16:50,120 --> 00:16:53,080
Choose work with clear steps, a clear owner,
451
00:16:53,080 --> 00:16:54,760
and an outcome you can measure.
452
00:16:54,760 --> 00:16:57,480
Oh, like how long supplier confirmations take,
453
00:16:57,480 --> 00:16:59,760
how many sales leads get a first reply,
454
00:16:59,760 --> 00:17:03,240
or how many cases land in the right queue on the first try.
455
00:17:03,240 --> 00:17:05,600
Don't start with the messiest process in the company.
456
00:17:05,600 --> 00:17:08,320
Instead, start with a queue where people already agree
457
00:17:08,320 --> 00:17:09,920
on what good work looks like.
458
00:17:09,920 --> 00:17:11,720
Your business data needs attention first.
459
00:17:11,720 --> 00:17:14,160
If one customer shows up under three different names,
460
00:17:14,160 --> 00:17:15,840
product details are incomplete,
461
00:17:15,840 --> 00:17:17,600
or supplier contacts are outdated,
462
00:17:17,600 --> 00:17:19,560
the agent can only work with what it finds.
463
00:17:19,560 --> 00:17:21,840
It might act fast, but speed won't turn bad data
464
00:17:21,840 --> 00:17:23,120
into a good result.
465
00:17:23,120 --> 00:17:26,280
Clean, consistent records give the agent a solid foundation.
466
00:17:26,280 --> 00:17:28,000
Now, write the job in plain English.
467
00:17:28,000 --> 00:17:29,000
What starts the work?
468
00:17:29,000 --> 00:17:30,640
What records can the agent read?
469
00:17:30,640 --> 00:17:31,720
Which actions can it take?
470
00:17:31,720 --> 00:17:33,280
What action needs approval?
471
00:17:33,280 --> 00:17:35,480
And when must it stop and send the work to a person?
472
00:17:35,480 --> 00:17:38,520
These questions sound basic, but they prevent confusion later.
473
00:17:38,520 --> 00:17:40,920
Imagine an agent that follows up with suppliers.
474
00:17:40,920 --> 00:17:42,880
You might allow it to request a delivery date,
475
00:17:42,880 --> 00:17:44,840
but not agree to a changed price.
476
00:17:44,840 --> 00:17:47,360
Or you might let it create a draft customer reply,
477
00:17:47,360 --> 00:17:49,920
but not send an apology that includes a refund promise.
478
00:17:49,920 --> 00:17:51,600
Clear boundaries protect both the business
479
00:17:51,600 --> 00:17:53,160
and the person using the agent.
480
00:17:53,160 --> 00:17:54,400
Access also matters.
481
00:17:54,400 --> 00:17:56,480
Entra ID handles identity and permissions
482
00:17:56,480 --> 00:17:58,200
across Microsoft services.
483
00:17:58,200 --> 00:18:00,680
Think of it as the reception desk that checks who or what
484
00:18:00,680 --> 00:18:03,920
who can enter a room and what they're allowed to do once inside.
485
00:18:03,920 --> 00:18:05,480
An agent should only reach the records
486
00:18:05,480 --> 00:18:07,640
and actions needed for its specific job.
487
00:18:07,640 --> 00:18:09,320
A sales agent doesn't need permission
488
00:18:09,320 --> 00:18:11,720
to change finance records, and a supplier agent
489
00:18:11,720 --> 00:18:14,000
doesn't need access to every customer case.
490
00:18:14,000 --> 00:18:15,840
Limiting access keeps the job focused
491
00:18:15,840 --> 00:18:18,120
and reduces risk if something goes wrong.
492
00:18:18,120 --> 00:18:20,600
Make sure people stay involved when the stakes rise.
493
00:18:20,600 --> 00:18:23,040
Money movement, contract terms, customer promises,
494
00:18:23,040 --> 00:18:25,960
unusual cases, legal questions, and sensitive data
495
00:18:25,960 --> 00:18:28,080
all need clear human review points.
496
00:18:28,080 --> 00:18:30,000
The agent can prepare the work, collect the facts,
497
00:18:30,000 --> 00:18:32,360
and root it quickly, but a person should approve the action
498
00:18:32,360 --> 00:18:34,400
when the result carries real business risk.
499
00:18:34,400 --> 00:18:36,480
Testing comes before wider use.
500
00:18:36,480 --> 00:18:39,280
Run the agent against real examples, including the awkward ones.
501
00:18:39,280 --> 00:18:41,480
Check what it read, what it prepared, what it changed,
502
00:18:41,480 --> 00:18:42,680
and where it stopped.
503
00:18:42,680 --> 00:18:45,040
Then review its activity records, listen to the people
504
00:18:45,040 --> 00:18:47,040
handling the work and improve the instructions
505
00:18:47,040 --> 00:18:49,200
before you give it access to more processes.
506
00:18:49,200 --> 00:18:51,160
You'll also need to watch usage.
507
00:18:51,160 --> 00:18:53,560
Dynamics 365 agents use something called
508
00:18:53,560 --> 00:18:56,520
co-pilot credits, which follow a consumption model.
509
00:18:56,520 --> 00:18:59,920
Dynamics 365 premium plans include monthly credits per user,
510
00:18:59,920 --> 00:19:02,360
but the exact allowance and how they're used depends
511
00:19:02,360 --> 00:19:03,880
on the current licensing terms.
512
00:19:03,880 --> 00:19:05,920
Always review the official Microsoft documentation
513
00:19:05,920 --> 00:19:07,560
before planning a large rollout.
514
00:19:07,560 --> 00:19:10,080
Starts more, inspect the work, and prove
515
00:19:10,080 --> 00:19:14,120
that one queue improved before expanding further.
516
00:19:14,120 --> 00:19:16,160
Conclusion.
517
00:19:16,160 --> 00:19:17,720
Pick one queue dynamics.
518
00:19:17,720 --> 00:19:22,480
365 agents turn repeated business tasks into supervised actions.
519
00:19:22,480 --> 00:19:25,240
They use the records and rules your teams already rely on.
520
00:19:25,240 --> 00:19:28,000
Don't start by asking AI to run an entire department.
521
00:19:28,000 --> 00:19:30,840
Instead, pick one queue this week, maybe unqualified leads,
522
00:19:30,840 --> 00:19:33,200
or open service cases, or supplier follow-ups,
523
00:19:33,200 --> 00:19:35,480
then write down every step and every approval.
524
00:19:35,480 --> 00:19:38,280
Subscribe to Microsoft Knowledge Nuggets on M365.
525
00:19:38,280 --> 00:19:40,640
FM share this with someone starting their journey
526
00:19:40,640 --> 00:19:42,920
and watch the next episode on co-pilot studio
527
00:19:42,920 --> 00:19:45,000
or Dynamics 365 Data Foundations.
528
00:19:45,000 --> 00:19:48,120
An agent can only act as well as the process you give it.