1
00:00:00,040 --> 00:00:03,439
Welcome to the M365 FM podcast,
where we explore the people,
2
00:00:03,439 --> 00:00:06,280
ideas and technologies shaping
the future of work.
3
00:00:06,280 --> 00:00:08,800
What robotic process automation
actually does.
4
00:00:08,800 --> 00:00:10,480
Every Monday, someone in finance
5
00:00:10,480 --> 00:00:11,919
retypes the same numbers into a
6
00:00:11,919 --> 00:00:13,519
fifth screen and nobody checks
7
00:00:13,519 --> 00:00:14,480
the result.
8
00:00:14,480 --> 00:00:17,519
Software can do that job now one
click at a time, and it doesn't
9
00:00:17,519 --> 00:00:19,079
need a developer to set it up.
10
00:00:19,079 --> 00:00:22,399
Analysts who track this say most
automation programs stall before
11
00:00:22,399 --> 00:00:25,079
their tenth robot, and the
software is rarely the reason.
12
00:00:25,079 --> 00:00:29,280
Welcome to the M365 FM podcast,
where we explore the people,
13
00:00:29,280 --> 00:00:32,240
ideas and technologies shaping
the future of work.
14
00:00:32,240 --> 00:00:33,359
By the end, you'll know which of
15
00:00:33,359 --> 00:00:34,840
your weekly tasks a robot can
16
00:00:34,840 --> 00:00:36,020
take and which ones it will make
17
00:00:36,020 --> 00:00:36,560
worse.
18
00:00:36,560 --> 00:00:37,920
The job itself is smaller than
19
00:00:37,920 --> 00:00:39,560
the name suggests, so let's
20
00:00:39,560 --> 00:00:40,439
start there.
21
00:00:40,439 --> 00:00:41,960
Robotic process automation is
22
00:00:41,960 --> 00:00:43,280
software that does the copying a
23
00:00:43,280 --> 00:00:44,200
person would otherwise do,
24
00:00:44,200 --> 00:00:45,520
driving the same screens and the
25
00:00:45,520 --> 00:00:46,679
same keystrokes in the same
26
00:00:46,679 --> 00:00:46,960
order.
27
00:00:46,960 --> 00:00:48,719
Every time your colleague opens
28
00:00:48,719 --> 00:00:50,479
an email, reads a number, types
29
00:00:50,479 --> 00:00:51,840
it into a spreadsheet, and then
30
00:00:51,840 --> 00:00:52,799
types it again into an
31
00:00:52,799 --> 00:00:53,799
accounting system.
32
00:00:53,799 --> 00:00:57,159
A bot does exactly that all day
long without ever getting bored.
33
00:00:57,159 --> 00:00:58,560
Nothing underneath those systems
34
00:00:58,560 --> 00:00:59,840
had to change for this to work,
35
00:00:59,840 --> 00:01:01,115
and that detail is the single
36
00:01:01,115 --> 00:01:02,090
most important thing to
37
00:01:02,090 --> 00:01:03,149
understand about the whole
38
00:01:03,149 --> 00:01:04,150
topic.
39
00:01:04,150 --> 00:01:05,230
The bot walks in through the
40
00:01:05,230 --> 00:01:06,709
front door instead of rewiring
41
00:01:06,709 --> 00:01:07,189
the building.
42
00:01:07,189 --> 00:01:10,030
You might think a bot means a
machine on a factory floor or a
43
00:01:10,030 --> 00:01:12,750
chat window that answers
customer questions on a website.
44
00:01:12,750 --> 00:01:15,750
In this world, a bot is nothing
more than a script that opens
45
00:01:15,750 --> 00:01:19,349
windows, clicks, buttons, types,
text, and reads back whatever
46
00:01:19,349 --> 00:01:20,269
appears on the screen.
47
00:01:20,269 --> 00:01:21,670
It has no body, no personality,
48
00:01:21,670 --> 00:01:22,590
and no opinions about your
49
00:01:22,590 --> 00:01:22,989
business.
50
00:01:22,989 --> 00:01:26,069
The word robotic is doing a lot
of misleading work in that name.
51
00:01:26,069 --> 00:01:27,390
It was never about hardware.
52
00:01:27,390 --> 00:01:29,790
It described software that
copies a human routine so
53
00:01:29,790 --> 00:01:32,510
faithfully that the routine
survives the handover.
54
00:01:32,510 --> 00:01:35,109
The name stuck, even though the
thing it describes sits much
55
00:01:35,109 --> 00:01:37,750
closer to a very obedient macro
than to anything you would find
56
00:01:37,750 --> 00:01:38,989
on a production line.
57
00:01:38,989 --> 00:01:42,189
Robotic process automation isn't
artificial intelligence, and
58
00:01:42,189 --> 00:01:45,150
that gap matters more than the
shared word robot suggests.
59
00:01:45,150 --> 00:01:48,049
A bot follows the steps a person
wrote down for it, and it
60
00:01:48,049 --> 00:01:50,230
repeats those steps the same way
every single time.
61
00:01:50,230 --> 00:01:53,030
Even when the situation around
it has clearly changed, it
62
00:01:53,030 --> 00:01:54,150
cannot recognize a pattern.
63
00:01:54,150 --> 00:01:55,189
Nobody taught it.
64
00:01:55,189 --> 00:01:58,109
It cannot make a judgment call
either, and it cannot learn from
65
00:01:58,109 --> 00:01:59,590
its own mistakes.
66
00:01:59,590 --> 00:02:00,989
Artificial intelligence works
67
00:02:00,989 --> 00:02:02,230
the other way around because it
68
00:02:02,230 --> 00:02:03,750
studies a pile of examples,
69
00:02:03,750 --> 00:02:05,549
finds the patterns inside them,
70
00:02:05,549 --> 00:02:07,109
and improves as it sees more of
71
00:02:07,109 --> 00:02:07,549
them.
72
00:02:07,549 --> 00:02:09,870
That's why AI handles the mess
and the variation.
73
00:02:09,870 --> 00:02:10,750
A bot cannot touch.
74
00:02:10,750 --> 00:02:13,389
So a bot is small, literal, and
cheap to point at a screen.
75
00:02:13,389 --> 00:02:14,750
That raises an obvious question
76
00:02:14,750 --> 00:02:16,509
about why so many companies
77
00:02:16,509 --> 00:02:17,669
started buying this kind of
78
00:02:17,669 --> 00:02:19,189
software at the same moment, and
79
00:02:19,189 --> 00:02:20,349
it's worth answering before we
80
00:02:20,349 --> 00:02:21,949
go further.
81
00:02:21,949 --> 00:02:23,669
Why this suddenly matters right
now.
82
00:02:23,669 --> 00:02:25,270
Robotic process automation used
83
00:02:25,270 --> 00:02:26,349
to live quietly in the back
84
00:02:26,349 --> 00:02:27,909
office on a few laptops in
85
00:02:27,909 --> 00:02:28,750
finance.
86
00:02:28,750 --> 00:02:30,949
Now it shows up in budget
conversations, and the
87
00:02:30,949 --> 00:02:33,789
researchers who track spending
on this kind of software expect
88
00:02:33,789 --> 00:02:36,389
the market to keep growing for
the rest of the decade.
89
00:02:36,389 --> 00:02:38,990
Treat any single figure with
suspicion, because the firms
90
00:02:38,990 --> 00:02:41,870
that publish those numbers count
things differently, and the
91
00:02:41,870 --> 00:02:44,590
numbers disagree with each other
more than you would like.
92
00:02:44,590 --> 00:02:46,150
The direction is the useful
part.
93
00:02:46,150 --> 00:02:48,509
The reason companies keep buying
has very little to do with the
94
00:02:48,509 --> 00:02:49,909
technology being exciting.
95
00:02:49,909 --> 00:02:52,669
Most large organizations run on
a stack of systems that were
96
00:02:52,669 --> 00:02:55,389
never designed to talk to each
other, and a human being ends up
97
00:02:55,389 --> 00:02:56,750
as the wire between them.
98
00:02:56,750 --> 00:02:58,389
Somebody opens one screen, reads
99
00:02:58,389 --> 00:02:59,469
a value, and types it into
100
00:02:59,469 --> 00:02:59,990
another.
101
00:02:59,990 --> 00:03:01,205
Multiply that across a few
102
00:03:01,205 --> 00:03:02,400
hundred people and a few
103
00:03:02,400 --> 00:03:03,620
thousand repetitions every
104
00:03:03,620 --> 00:03:04,780
month, and you end up with a
105
00:03:04,780 --> 00:03:06,020
number that a finance director
106
00:03:06,020 --> 00:03:06,860
notices.
107
00:03:06,860 --> 00:03:08,020
The work being automated is
108
00:03:08,020 --> 00:03:09,300
boring work, and that's exactly
109
00:03:09,300 --> 00:03:09,699
the point.
110
00:03:09,699 --> 00:03:10,699
That work is frequent,
111
00:03:10,699 --> 00:03:11,840
predictable, and easy to
112
00:03:11,840 --> 00:03:12,259
measure.
113
00:03:12,259 --> 00:03:13,419
So the payoff shows up on a
114
00:03:13,419 --> 00:03:14,419
spreadsheet instead of in a
115
00:03:14,419 --> 00:03:14,939
press release.
116
00:03:14,939 --> 00:03:17,300
There's a second reason this
arrived when it did, and it has
117
00:03:17,300 --> 00:03:19,219
to do with who was allowed to
build the thing.
118
00:03:19,219 --> 00:03:20,900
The tool stopped requiring a
programmer.
119
00:03:20,900 --> 00:03:23,580
The builder became a drag and
drop canvas that anyone could
120
00:03:23,580 --> 00:03:24,900
learn in an afternoon.
121
00:03:24,900 --> 00:03:26,199
That meant the people closest to
122
00:03:26,199 --> 00:03:27,340
the work could automate it
123
00:03:27,340 --> 00:03:28,699
themselves without filing a
124
00:03:28,699 --> 00:03:29,740
ticket and waiting a full
125
00:03:29,740 --> 00:03:30,939
quarter for someone else to help
126
00:03:30,939 --> 00:03:31,340
out.
127
00:03:31,340 --> 00:03:32,259
Before we get to where this
128
00:03:32,259 --> 00:03:33,560
money actually goes wrong, it's
129
00:03:33,560 --> 00:03:35,500
worth watching a bot do its job
130
00:03:35,500 --> 00:03:36,979
one step at a time on a real
131
00:03:36,979 --> 00:03:37,340
screen.
132
00:03:37,340 --> 00:03:39,740
Once you see the mechanics, one
limit becomes obvious.
133
00:03:39,740 --> 00:03:42,340
That limit explains why most
automation programs never grow
134
00:03:42,340 --> 00:03:44,180
past a handful of bots.
135
00:03:44,180 --> 00:03:45,979
How a bot actually does the
work.
136
00:03:45,979 --> 00:03:48,539
Building a bot starts with
teaching it a sequence.
137
00:03:48,539 --> 00:03:51,460
Most tools give you two ways to
do that, and the choice matters
138
00:03:51,460 --> 00:03:52,580
more than it looks.
139
00:03:52,580 --> 00:03:55,099
You can record yourself
performing the task once, or you
140
00:03:55,099 --> 00:03:57,740
can drag and drop each action
into a list by hand.
141
00:03:57,740 --> 00:03:59,340
Recording is faster to start and
142
00:03:59,340 --> 00:04:00,620
messier to maintain because the
143
00:04:00,620 --> 00:04:02,099
recorder captures everything you
144
00:04:02,099 --> 00:04:03,539
did, including the mis clicks
145
00:04:03,539 --> 00:04:04,580
and the detours you made by
146
00:04:04,580 --> 00:04:05,259
accident.
147
00:04:05,259 --> 00:04:07,099
Then the bot has to find things
on the screen.
148
00:04:07,099 --> 00:04:08,180
It does that in one of three
149
00:04:08,180 --> 00:04:09,860
ways, and the three are not
150
00:04:09,860 --> 00:04:10,860
equally sturdy.
151
00:04:10,860 --> 00:04:11,860
It can search for a named
152
00:04:11,860 --> 00:04:13,300
interface element, or a picture
153
00:04:13,300 --> 00:04:14,780
of a button or a fixed spot on
154
00:04:14,780 --> 00:04:15,620
the screen.
155
00:04:15,620 --> 00:04:16,860
Those three options get less
156
00:04:16,860 --> 00:04:18,420
reliable in that order, since a
157
00:04:18,420 --> 00:04:19,660
named element survives a full
158
00:04:19,660 --> 00:04:20,379
redesign.
159
00:04:20,379 --> 00:04:22,699
While a fixed pixel position
does not survive much of
160
00:04:22,699 --> 00:04:25,379
anything, the range of software
a bot can drive is wider than
161
00:04:25,379 --> 00:04:26,379
you might expect.
162
00:04:26,379 --> 00:04:29,540
A modern web app is easy, and so
is a desktop application, an
163
00:04:29,540 --> 00:04:32,699
Excel file, a shared folder, and
even a green screen terminal
164
00:04:32,699 --> 00:04:34,379
that predates the web.
165
00:04:34,379 --> 00:04:37,740
If a person can drive it with a
keyboard and a mouse, a bot can
166
00:04:37,740 --> 00:04:39,139
usually learn to drive it, too.
167
00:04:39,139 --> 00:04:41,579
Once the bot has the data, it
puts that data somewhere.
168
00:04:41,579 --> 00:04:44,259
It types into another
application, writes a row into a
169
00:04:44,259 --> 00:04:47,279
spreadsheet, drops a file into a
folder, or sends a message to
170
00:04:47,279 --> 00:04:48,459
whoever is waiting.
171
00:04:48,459 --> 00:04:51,379
The visible work is the
clicking, and the actual output
172
00:04:51,379 --> 00:04:54,300
is a record that moved from one
system to another without a
173
00:04:54,300 --> 00:04:55,379
person in the middle.
174
00:04:55,379 --> 00:04:57,060
There are two ways to let a bot
run.
175
00:04:57,060 --> 00:05:00,339
An attended bot sits next to a
person and fires when they ask
176
00:05:00,339 --> 00:05:03,329
for it, so a support agent
clicks one button and watches
177
00:05:03,329 --> 00:05:05,129
the details come back.
178
00:05:05,129 --> 00:05:06,610
An unattended bot runs on its
179
00:05:06,610 --> 00:05:08,089
own, on a schedule or on a
180
00:05:08,089 --> 00:05:09,649
trigger, with nobody in the room
181
00:05:09,649 --> 00:05:10,290
at all.
182
00:05:10,290 --> 00:05:12,730
The monthly report that used to
eat someone's Tuesday morning
183
00:05:12,730 --> 00:05:15,209
can now run at two in the
morning while everyone sleeps.
184
00:05:15,209 --> 00:05:17,610
Most bots in the world are the
unattended kind.
185
00:05:17,610 --> 00:05:19,009
An analyst at four fifty one
186
00:05:19,009 --> 00:05:20,209
research told TechTarget that
187
00:05:20,209 --> 00:05:21,649
the vast majority of bots built
188
00:05:21,649 --> 00:05:23,170
today are designed to run
189
00:05:23,170 --> 00:05:24,529
unattended because the goal is
190
00:05:24,529 --> 00:05:25,689
to take the person out of the
191
00:05:25,689 --> 00:05:26,850
loop entirely.
192
00:05:26,850 --> 00:05:28,050
So the mechanics are simple.
193
00:05:28,050 --> 00:05:30,769
The two operating modes are easy
to tell apart as well, which
194
00:05:30,769 --> 00:05:32,970
leaves the question that decides
whether any of this works.
195
00:05:32,970 --> 00:05:35,730
And that question is which work
deserves a bot?
196
00:05:35,730 --> 00:05:37,649
The jobs bots are genuinely good
at.
197
00:05:37,649 --> 00:05:38,370
Good automation.
198
00:05:38,370 --> 00:05:40,129
Work has a recognizable shape.
199
00:05:40,129 --> 00:05:40,930
It happens often.
200
00:05:40,930 --> 00:05:42,370
It follows the same rules every
time.
201
00:05:42,370 --> 00:05:43,649
It deals with clean, structured
202
00:05:43,649 --> 00:05:45,370
data, and it needs no judgment
203
00:05:45,370 --> 00:05:45,930
at all.
204
00:05:45,930 --> 00:05:48,449
If a task changes shape
depending on the customer or the
205
00:05:48,449 --> 00:05:51,449
wording of an email, a bot is
the wrong tool and you will find
206
00:05:51,449 --> 00:05:52,569
out the hard way.
207
00:05:52,569 --> 00:05:54,449
Invoices are the obvious
example.
208
00:05:54,449 --> 00:05:55,730
A supplier emails a document.
209
00:05:55,730 --> 00:05:57,889
Someone opens, it reads the
amount and the reference number
210
00:05:57,889 --> 00:06:00,889
and types both into a finance
system, one field after another.
211
00:06:00,889 --> 00:06:02,250
That's a perfect candidate.
212
00:06:02,250 --> 00:06:03,370
And it happens thousands of
213
00:06:03,370 --> 00:06:04,529
times a month in a mid-sized
214
00:06:04,529 --> 00:06:05,250
company.
215
00:06:05,250 --> 00:06:07,689
Onboarding and off boarding
follow the same pattern.
216
00:06:07,689 --> 00:06:10,250
When someone joins a person,
creates their account, adds them
217
00:06:10,250 --> 00:06:12,449
to the right groups, orders a
laptop, and sends out a welcome
218
00:06:12,449 --> 00:06:13,769
note to the whole team.
219
00:06:13,769 --> 00:06:16,529
The sequence barely changes from
one hire to the next.
220
00:06:16,529 --> 00:06:19,089
Offboarding is that job in
reverse, and it's often the one
221
00:06:19,089 --> 00:06:20,290
that finally gets funded.
222
00:06:20,290 --> 00:06:21,529
Reporting is the other big one.
223
00:06:21,529 --> 00:06:23,689
A weekly report that pulls
numbers out of four different
224
00:06:23,689 --> 00:06:26,689
places into one slide is the
kind of task that eats an
225
00:06:26,689 --> 00:06:29,569
afternoon and produces nothing
anyone enjoys making.
226
00:06:29,569 --> 00:06:32,009
Hand the collection to a bot,
and the human keeps the part
227
00:06:32,009 --> 00:06:34,889
that needs a brain, which is
deciding what the numbers mean.
228
00:06:34,889 --> 00:06:37,610
Banking and finance lead the
field by a wide margin.
229
00:06:37,610 --> 00:06:39,370
Analysts who study this put more
230
00:06:39,370 --> 00:06:40,689
than a third of all automation
231
00:06:40,689 --> 00:06:41,930
use cases in finance and
232
00:06:41,930 --> 00:06:42,970
accounting, followed by
233
00:06:42,970 --> 00:06:44,290
insurance, retail and
234
00:06:44,290 --> 00:06:45,089
healthcare.
235
00:06:45,089 --> 00:06:47,970
The patent isn't mysterious
because finance work was already
236
00:06:47,970 --> 00:06:50,610
rule based and documented long
before anyone called it
237
00:06:50,610 --> 00:06:53,949
automation, which made it
unusually ready to hand over
238
00:06:53,949 --> 00:06:55,889
everything up to this point is
the good news.
239
00:06:55,889 --> 00:06:57,370
The good news has a hard edge,
240
00:06:57,370 --> 00:06:58,589
because the qualities that make
241
00:06:58,589 --> 00:07:00,290
a task perfect for a bot are the
242
00:07:00,290 --> 00:07:01,329
same qualities that make it
243
00:07:01,329 --> 00:07:02,959
fragile in ways nobody plans
244
00:07:02,959 --> 00:07:04,000
for.
245
00:07:04,000 --> 00:07:05,199
Where RPA breaks.
246
00:07:05,199 --> 00:07:06,560
The limit I mentioned earlier is
247
00:07:06,560 --> 00:07:08,040
worth naming now, and it isn't
248
00:07:08,040 --> 00:07:09,160
really about the software being
249
00:07:09,160 --> 00:07:10,839
badly written, because every bot
250
00:07:10,839 --> 00:07:12,079
assumes the work always looks
251
00:07:12,079 --> 00:07:12,680
the same.
252
00:07:12,680 --> 00:07:14,040
Real work almost never does.
253
00:07:14,040 --> 00:07:16,259
Start with the data a bot reads
254
00:07:16,259 --> 00:07:18,199
a field, a cell, or a labeled
255
00:07:18,199 --> 00:07:19,839
box on a screen, and it needs
256
00:07:19,839 --> 00:07:20,920
that structure to actually
257
00:07:20,920 --> 00:07:21,680
exist.
258
00:07:21,680 --> 00:07:23,399
Hand it a scanned invoice, a PDF
259
00:07:23,399 --> 00:07:24,519
with the total in a slightly
260
00:07:24,519 --> 00:07:26,040
different place, or an email
261
00:07:26,040 --> 00:07:27,199
where someone buried the amount
262
00:07:27,199 --> 00:07:28,600
inside a paragraph and it has
263
00:07:28,600 --> 00:07:30,180
nothing to grab anything
264
00:07:30,180 --> 00:07:31,879
arriving as free text, an image
265
00:07:31,879 --> 00:07:33,319
or handwriting tends to stop a
266
00:07:33,319 --> 00:07:34,120
bot dead.
267
00:07:34,120 --> 00:07:35,480
Then there are the exceptions.
268
00:07:35,480 --> 00:07:37,000
Every real process has edge
269
00:07:37,000 --> 00:07:38,839
cases, and a bot handles the
270
00:07:38,839 --> 00:07:39,959
standard path and only the
271
00:07:39,959 --> 00:07:40,720
standard path.
272
00:07:40,720 --> 00:07:43,439
No matter how many exceptions,
show up every week when an order
273
00:07:43,439 --> 00:07:45,079
doesn't match the usual pattern.
274
00:07:45,079 --> 00:07:46,519
The work doesn't slow down, it
275
00:07:46,519 --> 00:07:48,240
just stops and it usually stops
276
00:07:48,240 --> 00:07:49,160
silently.
277
00:07:49,160 --> 00:07:50,160
Brittleness is the third
278
00:07:50,160 --> 00:07:51,600
problem, and this one catches
279
00:07:51,600 --> 00:07:52,860
people off guard because a bot
280
00:07:52,860 --> 00:07:53,879
drives the surface of an
281
00:07:53,879 --> 00:07:54,600
application.
282
00:07:54,600 --> 00:07:56,759
A small change to that surface
can break it.
283
00:07:56,759 --> 00:08:00,480
A button moves a field, gets
renamed, a login screen adds a
284
00:08:00,480 --> 00:08:03,759
step, and the bot that ran
perfectly for six months fails
285
00:08:03,759 --> 00:08:04,759
on a Monday morning.
286
00:08:04,759 --> 00:08:07,560
Scale is the fourth problem, and
it's what ends most programs.
287
00:08:07,560 --> 00:08:09,000
The research on this is blunt
288
00:08:09,000 --> 00:08:09,920
and more than half of the
289
00:08:09,920 --> 00:08:11,120
customers surveyed say they
290
00:08:11,120 --> 00:08:12,079
struggle to grow their
291
00:08:12,079 --> 00:08:13,519
automation beyond the first few
292
00:08:13,519 --> 00:08:14,639
processes.
293
00:08:14,639 --> 00:08:17,060
Their yardstick for an advanced
program is one hundred live
294
00:08:17,060 --> 00:08:19,639
bots, and most companies never
get past ten.
295
00:08:19,639 --> 00:08:22,600
This is also where the confusion
with AI agents creeps in.
296
00:08:22,600 --> 00:08:24,319
When a bot hits something it
wasn't taught.
297
00:08:24,319 --> 00:08:26,720
Someone has to decide what
happens next, and that decision
298
00:08:26,720 --> 00:08:29,439
is the part people keep hoping a
smarter model will handle.
299
00:08:29,439 --> 00:08:30,959
Sometimes a model really can.
300
00:08:30,959 --> 00:08:34,000
Often the honest answer is that
the process itself was never
301
00:08:34,000 --> 00:08:36,240
defined well enough for any tool
to follow it.
302
00:08:36,240 --> 00:08:38,159
Put those four limits together
and.
303
00:08:38,159 --> 00:08:39,679
The conclusion is uncomfortable,
304
00:08:39,679 --> 00:08:41,159
but useful for anyone planning
305
00:08:41,159 --> 00:08:41,960
this work.
306
00:08:41,960 --> 00:08:44,039
This is not a way to fix a messy
process.
307
00:08:44,039 --> 00:08:45,679
It's a way to copy a tidy one,
308
00:08:45,679 --> 00:08:47,159
and it stays reliable only for
309
00:08:47,159 --> 00:08:48,419
as long as the tidy version of
310
00:08:48,419 --> 00:08:49,639
that process holds up in
311
00:08:49,639 --> 00:08:50,320
practice.
312
00:08:50,320 --> 00:08:52,720
If the limits are this well
documented, you would expect
313
00:08:52,720 --> 00:08:54,279
companies to steer around them.
314
00:08:54,279 --> 00:08:55,679
And yet plenty of companies walk
315
00:08:55,679 --> 00:08:57,080
straight into all four of them
316
00:08:57,080 --> 00:08:57,960
anyway.
317
00:08:57,960 --> 00:08:59,240
The reasons have almost nothing
318
00:08:59,240 --> 00:09:01,440
to do with technology the
319
00:09:01,440 --> 00:09:02,950
mistakes that sink automation
320
00:09:02,950 --> 00:09:03,750
projects.
321
00:09:03,750 --> 00:09:05,149
The first mistake is automating
322
00:09:05,149 --> 00:09:06,350
something that should have been
323
00:09:06,350 --> 00:09:07,029
deleted.
324
00:09:07,029 --> 00:09:08,610
A bad process that runs by hand
325
00:09:08,610 --> 00:09:10,029
is annoying and visible, while
326
00:09:10,029 --> 00:09:11,350
the same process behind a bot
327
00:09:11,350 --> 00:09:13,429
runs faster, fails more quietly,
328
00:09:13,429 --> 00:09:14,850
and becomes much harder to argue
329
00:09:14,850 --> 00:09:15,429
with.
330
00:09:15,429 --> 00:09:18,350
Now it looks like a system
instead of a bad habit.
331
00:09:18,350 --> 00:09:20,470
The second mistake is launching
a bot and walking away.
332
00:09:20,470 --> 00:09:23,389
Someone has to own it because
when a screen changes or the
333
00:09:23,389 --> 00:09:26,389
volume doubles, a person needs
to notice and fix it.
334
00:09:26,389 --> 00:09:28,870
Programs that skip this step
find out their bots are broken
335
00:09:28,870 --> 00:09:31,629
at month end, when the numbers
no longer add up and nobody can
336
00:09:31,629 --> 00:09:32,389
explain the gap.
337
00:09:32,389 --> 00:09:33,750
The third mistake is leaving no
338
00:09:33,750 --> 00:09:35,230
path for the cases a bot can't
339
00:09:35,230 --> 00:09:35,789
handle.
340
00:09:35,789 --> 00:09:37,110
If the only options are a
341
00:09:37,110 --> 00:09:38,990
perfect run or a silent stop,
342
00:09:38,990 --> 00:09:40,269
then every unusual document
343
00:09:40,269 --> 00:09:41,909
becomes a lost invoice and the
344
00:09:41,909 --> 00:09:43,470
humans around the bot start
345
00:09:43,470 --> 00:09:44,750
working around it instead of
346
00:09:44,750 --> 00:09:45,350
with it.
347
00:09:45,350 --> 00:09:46,590
A five minute conversation about
348
00:09:46,590 --> 00:09:48,029
who handles the weird cases
349
00:09:48,029 --> 00:09:49,110
would have prevented the whole
350
00:09:49,110 --> 00:09:49,669
mess.
351
00:09:49,669 --> 00:09:51,990
The fourth mistake is measuring
the wrong thing.
352
00:09:51,990 --> 00:09:54,789
Hours saved is a lovely number
for a slide, and it says nothing
353
00:09:54,789 --> 00:09:57,149
about whether the work actually
left somebody's desk at the end
354
00:09:57,149 --> 00:09:57,549
of the week.
355
00:09:57,549 --> 00:09:58,909
If the same person still spends
356
00:09:58,909 --> 00:10:00,110
Friday afternoon checking the
357
00:10:00,110 --> 00:10:02,029
bots output, then no time was
358
00:10:02,029 --> 00:10:02,669
saved.
359
00:10:02,669 --> 00:10:04,190
the time just moved somewhere
else.
360
00:10:04,190 --> 00:10:07,070
The fifth mistake is treating
access as an afterthought.
361
00:10:07,070 --> 00:10:09,950
A bot usually logs in as a real
person or a shared account, so
362
00:10:09,950 --> 00:10:12,389
it inherits everything that
account can see and do.
363
00:10:12,389 --> 00:10:13,649
When nobody tracks which bot
364
00:10:13,649 --> 00:10:15,029
holds which access, you end up
365
00:10:15,029 --> 00:10:16,190
with automation that reaches
366
00:10:16,190 --> 00:10:17,750
further than anyone intended and
367
00:10:17,750 --> 00:10:19,070
no clean way to explain who
368
00:10:19,070 --> 00:10:19,750
approved it.
369
00:10:19,750 --> 00:10:21,350
Notice what's missing from that
list?
370
00:10:21,350 --> 00:10:23,990
Not one of those failures comes
from a badly built bot.
371
00:10:23,990 --> 00:10:25,429
They are process failures.
372
00:10:25,429 --> 00:10:27,830
Ownership failures and
measurement failures.
373
00:10:27,830 --> 00:10:29,389
Wearing a technology costume.
374
00:10:29,389 --> 00:10:31,230
And that is why buying newer
375
00:10:31,230 --> 00:10:33,070
software almost never fixes any
376
00:10:33,070 --> 00:10:33,669
of them.
377
00:10:33,669 --> 00:10:36,509
All of this matters more once
you realize something simple.
378
00:10:36,509 --> 00:10:37,669
The tool is probably already
379
00:10:37,669 --> 00:10:38,950
sitting inside your Microsoft
380
00:10:38,950 --> 00:10:40,429
license, and it is waiting for
381
00:10:40,429 --> 00:10:41,870
someone in your team to point it
382
00:10:41,870 --> 00:10:43,190
at a task that deserves the
383
00:10:43,190 --> 00:10:44,230
effort.
384
00:10:44,230 --> 00:10:45,629
Where RPA fits in Microsoft
385
00:10:45,629 --> 00:10:46,629
three hundred sixty five and
386
00:10:46,629 --> 00:10:47,149
Azure.
387
00:10:47,149 --> 00:10:49,149
If your company runs on
Microsoft three hundred sixty
388
00:10:49,149 --> 00:10:51,470
five, the automation tool is
already in the building, and
389
00:10:51,470 --> 00:10:53,190
it's called Power Automate.
390
00:10:53,190 --> 00:10:56,149
The RPA part has its own name,
which is desktop flows.
391
00:10:56,149 --> 00:10:57,330
You build them by dragging
392
00:10:57,330 --> 00:10:58,750
actions into a sequence or by
393
00:10:58,750 --> 00:10:59,830
recording yourself doing the
394
00:10:59,830 --> 00:11:01,389
task once, which is the same
395
00:11:01,389 --> 00:11:03,059
approach every other RPA product
396
00:11:03,059 --> 00:11:04,059
takes power.
397
00:11:04,059 --> 00:11:05,259
Automate holds two kinds of
398
00:11:05,259 --> 00:11:06,139
flow, and the difference
399
00:11:06,139 --> 00:11:06,580
matters.
400
00:11:06,580 --> 00:11:09,500
Cloud flows run in Microsoft's
cloud and connect services to
401
00:11:09,500 --> 00:11:12,220
each other, so a new form
response can create a record,
402
00:11:12,220 --> 00:11:15,580
send a message, and update a
list on its own desktop flows,
403
00:11:15,580 --> 00:11:18,299
run on a machine and drive the
screen, which is what lets them
404
00:11:18,299 --> 00:11:20,460
handle software with no modern
connection at all.
405
00:11:20,460 --> 00:11:23,620
Real automation usually uses
both with a cloud flow starting
406
00:11:23,620 --> 00:11:26,340
the work and a desktop flow
finishing it on an older system.
407
00:11:26,340 --> 00:11:28,139
Here's what that looks like end
to end.
408
00:11:28,139 --> 00:11:31,139
Someone submits an expense form,
a cloud flow catches the
409
00:11:31,139 --> 00:11:33,940
response, and it checks the
amount against the policy.
410
00:11:33,940 --> 00:11:35,659
If the amount looks normal, the
411
00:11:35,659 --> 00:11:37,419
cloud flow hands the record to a
412
00:11:37,419 --> 00:11:39,460
desktop flow, which logs into
413
00:11:39,460 --> 00:11:41,220
the finance system that has no
414
00:11:41,220 --> 00:11:42,659
modern connection and enters the
415
00:11:42,659 --> 00:11:43,299
claim.
416
00:11:43,299 --> 00:11:46,259
If anything looks off the flow,
routes it to a person instead.
417
00:11:46,259 --> 00:11:48,659
Nobody watches the whole thing
happen in real time, and the
418
00:11:48,659 --> 00:11:50,940
trail of what occurred stays in
the run history.
419
00:11:50,940 --> 00:11:54,299
For anyone who wants to check
under the hood, Power Automate
420
00:11:54,299 --> 00:11:57,059
is built on a service called
Azure Logic Apps.
421
00:11:57,059 --> 00:11:59,220
The two look similar on the
surface, and the difference is
422
00:11:59,220 --> 00:12:00,860
mostly about who they're for.
423
00:12:00,860 --> 00:12:03,740
Power automate is aimed at
makers and everyday users inside
424
00:12:03,740 --> 00:12:06,740
Microsoft three hundred sixty
five with per user licensing.
425
00:12:06,740 --> 00:12:08,019
While Logic Apps is aimed at
426
00:12:08,019 --> 00:12:09,500
developers and IT teams working
427
00:12:09,500 --> 00:12:10,379
in Azure.
428
00:12:10,379 --> 00:12:12,659
If you're a beginner, you'll
live in Power Automate and you
429
00:12:12,659 --> 00:12:15,259
can grow into logic apps later
without starting over.
430
00:12:15,259 --> 00:12:16,980
The part everyone asks about is
what it costs.
431
00:12:16,980 --> 00:12:19,659
Many Microsoft three hundred
sixty five plans already include
432
00:12:19,659 --> 00:12:22,659
standard connectors, so basic
cloud flows are often covered.
433
00:12:22,659 --> 00:12:25,740
Attended desktop flows sit in
the premium plan, which lists at
434
00:12:25,740 --> 00:12:28,700
fifteen dollars per user per
month when you pay yearly.
435
00:12:28,700 --> 00:12:31,620
Unattended bots are licensed per
bot instead of per person, at
436
00:12:31,620 --> 00:12:34,039
one hundred and fifty dollars a
month, and that price rises to
437
00:12:34,039 --> 00:12:36,379
two hundred and fifteen dollars
if Microsoft hosts the machine
438
00:12:36,379 --> 00:12:37,740
for you on Azure.
439
00:12:37,740 --> 00:12:39,179
That per bot licensing is worth
440
00:12:39,179 --> 00:12:40,259
sitting with for a moment
441
00:12:40,259 --> 00:12:41,740
because it explains a decision
442
00:12:41,740 --> 00:12:42,899
you will see inside real
443
00:12:42,899 --> 00:12:43,299
companies.
444
00:12:43,299 --> 00:12:46,419
One bot can run one unattended
task at a time, which means
445
00:12:46,419 --> 00:12:49,059
running five things at the same
time turns into an invoice for
446
00:12:49,059 --> 00:12:50,419
five separate bots.
447
00:12:50,419 --> 00:12:53,139
Starting is cheap, and scaling
is where the invoice grows.
448
00:12:53,139 --> 00:12:54,659
Around all of that sits the
449
00:12:54,659 --> 00:12:56,019
wider story Microsoft keeps
450
00:12:56,019 --> 00:12:56,620
telling.
451
00:12:56,620 --> 00:12:59,419
Process mining finds the work
worth automating by reading the
452
00:12:59,419 --> 00:13:02,340
trails your systems already
leave behind, and copilot and
453
00:13:02,340 --> 00:13:04,570
agents add judgment where the
rules run out.
454
00:13:04,570 --> 00:13:07,450
That's the direction of travel,
and it deserves its own episode.
455
00:13:07,450 --> 00:13:09,250
Knowing the tool exists is easy.
456
00:13:09,250 --> 00:13:11,490
Knowing what to do with it on a
Monday morning when nobody has
457
00:13:11,490 --> 00:13:13,970
told you which tasks are worth
automating, is where most
458
00:13:13,970 --> 00:13:16,250
beginners stall out and stop.
459
00:13:16,250 --> 00:13:17,850
What this means for your career.
460
00:13:17,850 --> 00:13:18,970
The skill worth building here
461
00:13:18,970 --> 00:13:20,210
has nothing to do with building
462
00:13:20,210 --> 00:13:20,929
bots.
463
00:13:20,929 --> 00:13:22,409
It's the ability to spot bot
464
00:13:22,409 --> 00:13:23,929
shaped work and describe it
465
00:13:23,929 --> 00:13:25,210
clearly enough that someone else
466
00:13:25,210 --> 00:13:26,490
can take it and act on it
467
00:13:26,490 --> 00:13:27,809
without asking you a dozen
468
00:13:27,809 --> 00:13:28,570
questions.
469
00:13:28,570 --> 00:13:30,529
Here's a simple exercise you can
run this week.
470
00:13:30,529 --> 00:13:31,970
Pick one task you do every week
471
00:13:31,970 --> 00:13:33,129
that runs through the same steps
472
00:13:33,129 --> 00:13:34,570
in the same order, and write
473
00:13:34,570 --> 00:13:35,730
those steps down as if you were
474
00:13:35,730 --> 00:13:36,730
explaining them to a new
475
00:13:36,730 --> 00:13:37,129
colleague.
476
00:13:37,129 --> 00:13:39,610
If the list comes out clean,
that task is a real candidate
477
00:13:39,610 --> 00:13:40,330
for a bot.
478
00:13:40,330 --> 00:13:42,289
If you keep stopping to write
that, it depends.
479
00:13:42,289 --> 00:13:44,529
You found the part that would
break it, and that's useful
480
00:13:44,529 --> 00:13:45,450
information to that.
481
00:13:45,450 --> 00:13:47,289
Exercise is also why this skill
travels.
482
00:13:47,289 --> 00:13:50,289
The people who get real value
from automation are the ones who
483
00:13:50,289 --> 00:13:53,129
can explain a process clearly
enough that a machine and a
484
00:13:53,129 --> 00:13:54,929
colleague both follow it.
485
00:13:54,929 --> 00:13:57,370
That ability comes up in job
interviews far more often than
486
00:13:57,370 --> 00:13:59,090
any specific certification.
487
00:13:59,090 --> 00:14:00,450
One honest note before you
488
00:14:00,450 --> 00:14:01,950
automate everything in sight,
489
00:14:01,950 --> 00:14:04,110
this is a useful edge and not a
490
00:14:04,110 --> 00:14:05,610
guaranteed promotion, and it
491
00:14:05,610 --> 00:14:06,649
doesn't replace knowing the
492
00:14:06,649 --> 00:14:07,330
business.
493
00:14:07,330 --> 00:14:08,570
A bot that saves four hours a
494
00:14:08,570 --> 00:14:10,110
week on a process nobody needed
495
00:14:10,110 --> 00:14:11,809
is still a wasted project, and
496
00:14:11,809 --> 00:14:12,809
the person who understands the
497
00:14:12,809 --> 00:14:14,370
business usually catches that
498
00:14:14,370 --> 00:14:15,649
first.
499
00:14:15,649 --> 00:14:17,889
The short version and one
question for you.
500
00:14:17,889 --> 00:14:19,250
The short version is simple.
501
00:14:19,250 --> 00:14:20,450
The software clicks and types
502
00:14:20,450 --> 00:14:21,649
the way a person does on the
503
00:14:21,649 --> 00:14:23,009
same screens with no AI
504
00:14:23,009 --> 00:14:23,450
involved.
505
00:14:23,450 --> 00:14:26,690
It's genuinely good at high
volume, rule based work with
506
00:14:26,690 --> 00:14:29,690
tidy data, and it breaks the
moment the work stops looking
507
00:14:29,690 --> 00:14:30,929
the same way twice.
508
00:14:30,929 --> 00:14:32,330
Getting started is cheap.
509
00:14:32,330 --> 00:14:35,529
The bot licensing only kicks in
once you go unattended and the
510
00:14:35,529 --> 00:14:36,809
whole thing lives inside.
511
00:14:36,809 --> 00:14:39,409
Power Automate If your company
already runs on Microsoft three
512
00:14:39,409 --> 00:14:40,490
hundred sixty five.
513
00:14:40,490 --> 00:14:42,009
Most programs stall for reasons
514
00:14:42,009 --> 00:14:43,210
that sit outside the bots
515
00:14:43,210 --> 00:14:43,809
entirely.
516
00:14:43,809 --> 00:14:46,450
Nobody owned the process around
them, so nothing got fixed when
517
00:14:46,450 --> 00:14:49,129
the work changed and the bots
took the blame for a problem
518
00:14:49,129 --> 00:14:50,129
they never caused.
519
00:14:50,129 --> 00:14:52,169
So here is the question I'll
leave you with.
520
00:14:52,169 --> 00:14:53,450
Which single task in your own
521
00:14:53,450 --> 00:14:54,570
week would you hand to a bot
522
00:14:54,570 --> 00:14:54,970
tomorrow?
523
00:14:54,970 --> 00:14:55,970
And which part of it would you
524
00:14:55,970 --> 00:14:57,370
still want a human to check if
525
00:14:57,370 --> 00:14:58,529
this made the topic feel less
526
00:14:58,529 --> 00:14:59,490
intimidating?
527
00:14:59,490 --> 00:15:01,289
Subscribe to M365 FM.
528
00:15:01,289 --> 00:15:02,289
To master the tools and the
529
00:15:02,289 --> 00:15:03,809
mindset shaping the future of
530
00:15:03,809 --> 00:15:04,450
work.
00:00:00,040 --> 00:00:03,439
Welcome to the M365 FM podcast,
where we explore the people,
2
00:00:03,439 --> 00:00:06,280
ideas and technologies shaping
the future of work.
3
00:00:06,280 --> 00:00:08,800
What robotic process automation
actually does.
4
00:00:08,800 --> 00:00:10,480
Every Monday, someone in finance
5
00:00:10,480 --> 00:00:11,919
retypes the same numbers into a
6
00:00:11,919 --> 00:00:13,519
fifth screen and nobody checks
7
00:00:13,519 --> 00:00:14,480
the result.
8
00:00:14,480 --> 00:00:17,519
Software can do that job now one
click at a time, and it doesn't
9
00:00:17,519 --> 00:00:19,079
need a developer to set it up.
10
00:00:19,079 --> 00:00:22,399
Analysts who track this say most
automation programs stall before
11
00:00:22,399 --> 00:00:25,079
their tenth robot, and the
software is rarely the reason.
12
00:00:25,079 --> 00:00:29,280
Welcome to the M365 FM podcast,
where we explore the people,
13
00:00:29,280 --> 00:00:32,240
ideas and technologies shaping
the future of work.
14
00:00:32,240 --> 00:00:33,359
By the end, you'll know which of
15
00:00:33,359 --> 00:00:34,840
your weekly tasks a robot can
16
00:00:34,840 --> 00:00:36,020
take and which ones it will make
17
00:00:36,020 --> 00:00:36,560
worse.
18
00:00:36,560 --> 00:00:37,920
The job itself is smaller than
19
00:00:37,920 --> 00:00:39,560
the name suggests, so let's
20
00:00:39,560 --> 00:00:40,439
start there.
21
00:00:40,439 --> 00:00:41,960
Robotic process automation is
22
00:00:41,960 --> 00:00:43,280
software that does the copying a
23
00:00:43,280 --> 00:00:44,200
person would otherwise do,
24
00:00:44,200 --> 00:00:45,520
driving the same screens and the
25
00:00:45,520 --> 00:00:46,679
same keystrokes in the same
26
00:00:46,679 --> 00:00:46,960
order.
27
00:00:46,960 --> 00:00:48,719
Every time your colleague opens
28
00:00:48,719 --> 00:00:50,479
an email, reads a number, types
29
00:00:50,479 --> 00:00:51,840
it into a spreadsheet, and then
30
00:00:51,840 --> 00:00:52,799
types it again into an
31
00:00:52,799 --> 00:00:53,799
accounting system.
32
00:00:53,799 --> 00:00:57,159
A bot does exactly that all day
long without ever getting bored.
33
00:00:57,159 --> 00:00:58,560
Nothing underneath those systems
34
00:00:58,560 --> 00:00:59,840
had to change for this to work,
35
00:00:59,840 --> 00:01:01,115
and that detail is the single
36
00:01:01,115 --> 00:01:02,090
most important thing to
37
00:01:02,090 --> 00:01:03,149
understand about the whole
38
00:01:03,149 --> 00:01:04,150
topic.
39
00:01:04,150 --> 00:01:05,230
The bot walks in through the
40
00:01:05,230 --> 00:01:06,709
front door instead of rewiring
41
00:01:06,709 --> 00:01:07,189
the building.
42
00:01:07,189 --> 00:01:10,030
You might think a bot means a
machine on a factory floor or a
43
00:01:10,030 --> 00:01:12,750
chat window that answers
customer questions on a website.
44
00:01:12,750 --> 00:01:15,750
In this world, a bot is nothing
more than a script that opens
45
00:01:15,750 --> 00:01:19,349
windows, clicks, buttons, types,
text, and reads back whatever
46
00:01:19,349 --> 00:01:20,269
appears on the screen.
47
00:01:20,269 --> 00:01:21,670
It has no body, no personality,
48
00:01:21,670 --> 00:01:22,590
and no opinions about your
49
00:01:22,590 --> 00:01:22,989
business.
50
00:01:22,989 --> 00:01:26,069
The word robotic is doing a lot
of misleading work in that name.
51
00:01:26,069 --> 00:01:27,390
It was never about hardware.
52
00:01:27,390 --> 00:01:29,790
It described software that
copies a human routine so
53
00:01:29,790 --> 00:01:32,510
faithfully that the routine
survives the handover.
54
00:01:32,510 --> 00:01:35,109
The name stuck, even though the
thing it describes sits much
55
00:01:35,109 --> 00:01:37,750
closer to a very obedient macro
than to anything you would find
56
00:01:37,750 --> 00:01:38,989
on a production line.
57
00:01:38,989 --> 00:01:42,189
Robotic process automation isn't
artificial intelligence, and
58
00:01:42,189 --> 00:01:45,150
that gap matters more than the
shared word robot suggests.
59
00:01:45,150 --> 00:01:48,049
A bot follows the steps a person
wrote down for it, and it
60
00:01:48,049 --> 00:01:50,230
repeats those steps the same way
every single time.
61
00:01:50,230 --> 00:01:53,030
Even when the situation around
it has clearly changed, it
62
00:01:53,030 --> 00:01:54,150
cannot recognize a pattern.
63
00:01:54,150 --> 00:01:55,189
Nobody taught it.
64
00:01:55,189 --> 00:01:58,109
It cannot make a judgment call
either, and it cannot learn from
65
00:01:58,109 --> 00:01:59,590
its own mistakes.
66
00:01:59,590 --> 00:02:00,989
Artificial intelligence works
67
00:02:00,989 --> 00:02:02,230
the other way around because it
68
00:02:02,230 --> 00:02:03,750
studies a pile of examples,
69
00:02:03,750 --> 00:02:05,549
finds the patterns inside them,
70
00:02:05,549 --> 00:02:07,109
and improves as it sees more of
71
00:02:07,109 --> 00:02:07,549
them.
72
00:02:07,549 --> 00:02:09,870
That's why AI handles the mess
and the variation.
73
00:02:09,870 --> 00:02:10,750
A bot cannot touch.
74
00:02:10,750 --> 00:02:13,389
So a bot is small, literal, and
cheap to point at a screen.
75
00:02:13,389 --> 00:02:14,750
That raises an obvious question
76
00:02:14,750 --> 00:02:16,509
about why so many companies
77
00:02:16,509 --> 00:02:17,669
started buying this kind of
78
00:02:17,669 --> 00:02:19,189
software at the same moment, and
79
00:02:19,189 --> 00:02:20,349
it's worth answering before we
80
00:02:20,349 --> 00:02:21,949
go further.
81
00:02:21,949 --> 00:02:23,669
Why this suddenly matters right
now.
82
00:02:23,669 --> 00:02:25,270
Robotic process automation used
83
00:02:25,270 --> 00:02:26,349
to live quietly in the back
84
00:02:26,349 --> 00:02:27,909
office on a few laptops in
85
00:02:27,909 --> 00:02:28,750
finance.
86
00:02:28,750 --> 00:02:30,949
Now it shows up in budget
conversations, and the
87
00:02:30,949 --> 00:02:33,789
researchers who track spending
on this kind of software expect
88
00:02:33,789 --> 00:02:36,389
the market to keep growing for
the rest of the decade.
89
00:02:36,389 --> 00:02:38,990
Treat any single figure with
suspicion, because the firms
90
00:02:38,990 --> 00:02:41,870
that publish those numbers count
things differently, and the
91
00:02:41,870 --> 00:02:44,590
numbers disagree with each other
more than you would like.
92
00:02:44,590 --> 00:02:46,150
The direction is the useful
part.
93
00:02:46,150 --> 00:02:48,509
The reason companies keep buying
has very little to do with the
94
00:02:48,509 --> 00:02:49,909
technology being exciting.
95
00:02:49,909 --> 00:02:52,669
Most large organizations run on
a stack of systems that were
96
00:02:52,669 --> 00:02:55,389
never designed to talk to each
other, and a human being ends up
97
00:02:55,389 --> 00:02:56,750
as the wire between them.
98
00:02:56,750 --> 00:02:58,389
Somebody opens one screen, reads
99
00:02:58,389 --> 00:02:59,469
a value, and types it into
100
00:02:59,469 --> 00:02:59,990
another.
101
00:02:59,990 --> 00:03:01,205
Multiply that across a few
102
00:03:01,205 --> 00:03:02,400
hundred people and a few
103
00:03:02,400 --> 00:03:03,620
thousand repetitions every
104
00:03:03,620 --> 00:03:04,780
month, and you end up with a
105
00:03:04,780 --> 00:03:06,020
number that a finance director
106
00:03:06,020 --> 00:03:06,860
notices.
107
00:03:06,860 --> 00:03:08,020
The work being automated is
108
00:03:08,020 --> 00:03:09,300
boring work, and that's exactly
109
00:03:09,300 --> 00:03:09,699
the point.
110
00:03:09,699 --> 00:03:10,699
That work is frequent,
111
00:03:10,699 --> 00:03:11,840
predictable, and easy to
112
00:03:11,840 --> 00:03:12,259
measure.
113
00:03:12,259 --> 00:03:13,419
So the payoff shows up on a
114
00:03:13,419 --> 00:03:14,419
spreadsheet instead of in a
115
00:03:14,419 --> 00:03:14,939
press release.
116
00:03:14,939 --> 00:03:17,300
There's a second reason this
arrived when it did, and it has
117
00:03:17,300 --> 00:03:19,219
to do with who was allowed to
build the thing.
118
00:03:19,219 --> 00:03:20,900
The tool stopped requiring a
programmer.
119
00:03:20,900 --> 00:03:23,580
The builder became a drag and
drop canvas that anyone could
120
00:03:23,580 --> 00:03:24,900
learn in an afternoon.
121
00:03:24,900 --> 00:03:26,199
That meant the people closest to
122
00:03:26,199 --> 00:03:27,340
the work could automate it
123
00:03:27,340 --> 00:03:28,699
themselves without filing a
124
00:03:28,699 --> 00:03:29,740
ticket and waiting a full
125
00:03:29,740 --> 00:03:30,939
quarter for someone else to help
126
00:03:30,939 --> 00:03:31,340
out.
127
00:03:31,340 --> 00:03:32,259
Before we get to where this
128
00:03:32,259 --> 00:03:33,560
money actually goes wrong, it's
129
00:03:33,560 --> 00:03:35,500
worth watching a bot do its job
130
00:03:35,500 --> 00:03:36,979
one step at a time on a real
131
00:03:36,979 --> 00:03:37,340
screen.
132
00:03:37,340 --> 00:03:39,740
Once you see the mechanics, one
limit becomes obvious.
133
00:03:39,740 --> 00:03:42,340
That limit explains why most
automation programs never grow
134
00:03:42,340 --> 00:03:44,180
past a handful of bots.
135
00:03:44,180 --> 00:03:45,979
How a bot actually does the
work.
136
00:03:45,979 --> 00:03:48,539
Building a bot starts with
teaching it a sequence.
137
00:03:48,539 --> 00:03:51,460
Most tools give you two ways to
do that, and the choice matters
138
00:03:51,460 --> 00:03:52,580
more than it looks.
139
00:03:52,580 --> 00:03:55,099
You can record yourself
performing the task once, or you
140
00:03:55,099 --> 00:03:57,740
can drag and drop each action
into a list by hand.
141
00:03:57,740 --> 00:03:59,340
Recording is faster to start and
142
00:03:59,340 --> 00:04:00,620
messier to maintain because the
143
00:04:00,620 --> 00:04:02,099
recorder captures everything you
144
00:04:02,099 --> 00:04:03,539
did, including the mis clicks
145
00:04:03,539 --> 00:04:04,580
and the detours you made by
146
00:04:04,580 --> 00:04:05,259
accident.
147
00:04:05,259 --> 00:04:07,099
Then the bot has to find things
on the screen.
148
00:04:07,099 --> 00:04:08,180
It does that in one of three
149
00:04:08,180 --> 00:04:09,860
ways, and the three are not
150
00:04:09,860 --> 00:04:10,860
equally sturdy.
151
00:04:10,860 --> 00:04:11,860
It can search for a named
152
00:04:11,860 --> 00:04:13,300
interface element, or a picture
153
00:04:13,300 --> 00:04:14,780
of a button or a fixed spot on
154
00:04:14,780 --> 00:04:15,620
the screen.
155
00:04:15,620 --> 00:04:16,860
Those three options get less
156
00:04:16,860 --> 00:04:18,420
reliable in that order, since a
157
00:04:18,420 --> 00:04:19,660
named element survives a full
158
00:04:19,660 --> 00:04:20,379
redesign.
159
00:04:20,379 --> 00:04:22,699
While a fixed pixel position
does not survive much of
160
00:04:22,699 --> 00:04:25,379
anything, the range of software
a bot can drive is wider than
161
00:04:25,379 --> 00:04:26,379
you might expect.
162
00:04:26,379 --> 00:04:29,540
A modern web app is easy, and so
is a desktop application, an
163
00:04:29,540 --> 00:04:32,699
Excel file, a shared folder, and
even a green screen terminal
164
00:04:32,699 --> 00:04:34,379
that predates the web.
165
00:04:34,379 --> 00:04:37,740
If a person can drive it with a
keyboard and a mouse, a bot can
166
00:04:37,740 --> 00:04:39,139
usually learn to drive it, too.
167
00:04:39,139 --> 00:04:41,579
Once the bot has the data, it
puts that data somewhere.
168
00:04:41,579 --> 00:04:44,259
It types into another
application, writes a row into a
169
00:04:44,259 --> 00:04:47,279
spreadsheet, drops a file into a
folder, or sends a message to
170
00:04:47,279 --> 00:04:48,459
whoever is waiting.
171
00:04:48,459 --> 00:04:51,379
The visible work is the
clicking, and the actual output
172
00:04:51,379 --> 00:04:54,300
is a record that moved from one
system to another without a
173
00:04:54,300 --> 00:04:55,379
person in the middle.
174
00:04:55,379 --> 00:04:57,060
There are two ways to let a bot
run.
175
00:04:57,060 --> 00:05:00,339
An attended bot sits next to a
person and fires when they ask
176
00:05:00,339 --> 00:05:03,329
for it, so a support agent
clicks one button and watches
177
00:05:03,329 --> 00:05:05,129
the details come back.
178
00:05:05,129 --> 00:05:06,610
An unattended bot runs on its
179
00:05:06,610 --> 00:05:08,089
own, on a schedule or on a
180
00:05:08,089 --> 00:05:09,649
trigger, with nobody in the room
181
00:05:09,649 --> 00:05:10,290
at all.
182
00:05:10,290 --> 00:05:12,730
The monthly report that used to
eat someone's Tuesday morning
183
00:05:12,730 --> 00:05:15,209
can now run at two in the
morning while everyone sleeps.
184
00:05:15,209 --> 00:05:17,610
Most bots in the world are the
unattended kind.
185
00:05:17,610 --> 00:05:19,009
An analyst at four fifty one
186
00:05:19,009 --> 00:05:20,209
research told TechTarget that
187
00:05:20,209 --> 00:05:21,649
the vast majority of bots built
188
00:05:21,649 --> 00:05:23,170
today are designed to run
189
00:05:23,170 --> 00:05:24,529
unattended because the goal is
190
00:05:24,529 --> 00:05:25,689
to take the person out of the
191
00:05:25,689 --> 00:05:26,850
loop entirely.
192
00:05:26,850 --> 00:05:28,050
So the mechanics are simple.
193
00:05:28,050 --> 00:05:30,769
The two operating modes are easy
to tell apart as well, which
194
00:05:30,769 --> 00:05:32,970
leaves the question that decides
whether any of this works.
195
00:05:32,970 --> 00:05:35,730
And that question is which work
deserves a bot?
196
00:05:35,730 --> 00:05:37,649
The jobs bots are genuinely good
at.
197
00:05:37,649 --> 00:05:38,370
Good automation.
198
00:05:38,370 --> 00:05:40,129
Work has a recognizable shape.
199
00:05:40,129 --> 00:05:40,930
It happens often.
200
00:05:40,930 --> 00:05:42,370
It follows the same rules every
time.
201
00:05:42,370 --> 00:05:43,649
It deals with clean, structured
202
00:05:43,649 --> 00:05:45,370
data, and it needs no judgment
203
00:05:45,370 --> 00:05:45,930
at all.
204
00:05:45,930 --> 00:05:48,449
If a task changes shape
depending on the customer or the
205
00:05:48,449 --> 00:05:51,449
wording of an email, a bot is
the wrong tool and you will find
206
00:05:51,449 --> 00:05:52,569
out the hard way.
207
00:05:52,569 --> 00:05:54,449
Invoices are the obvious
example.
208
00:05:54,449 --> 00:05:55,730
A supplier emails a document.
209
00:05:55,730 --> 00:05:57,889
Someone opens, it reads the
amount and the reference number
210
00:05:57,889 --> 00:06:00,889
and types both into a finance
system, one field after another.
211
00:06:00,889 --> 00:06:02,250
That's a perfect candidate.
212
00:06:02,250 --> 00:06:03,370
And it happens thousands of
213
00:06:03,370 --> 00:06:04,529
times a month in a mid-sized
214
00:06:04,529 --> 00:06:05,250
company.
215
00:06:05,250 --> 00:06:07,689
Onboarding and off boarding
follow the same pattern.
216
00:06:07,689 --> 00:06:10,250
When someone joins a person,
creates their account, adds them
217
00:06:10,250 --> 00:06:12,449
to the right groups, orders a
laptop, and sends out a welcome
218
00:06:12,449 --> 00:06:13,769
note to the whole team.
219
00:06:13,769 --> 00:06:16,529
The sequence barely changes from
one hire to the next.
220
00:06:16,529 --> 00:06:19,089
Offboarding is that job in
reverse, and it's often the one
221
00:06:19,089 --> 00:06:20,290
that finally gets funded.
222
00:06:20,290 --> 00:06:21,529
Reporting is the other big one.
223
00:06:21,529 --> 00:06:23,689
A weekly report that pulls
numbers out of four different
224
00:06:23,689 --> 00:06:26,689
places into one slide is the
kind of task that eats an
225
00:06:26,689 --> 00:06:29,569
afternoon and produces nothing
anyone enjoys making.
226
00:06:29,569 --> 00:06:32,009
Hand the collection to a bot,
and the human keeps the part
227
00:06:32,009 --> 00:06:34,889
that needs a brain, which is
deciding what the numbers mean.
228
00:06:34,889 --> 00:06:37,610
Banking and finance lead the
field by a wide margin.
229
00:06:37,610 --> 00:06:39,370
Analysts who study this put more
230
00:06:39,370 --> 00:06:40,689
than a third of all automation
231
00:06:40,689 --> 00:06:41,930
use cases in finance and
232
00:06:41,930 --> 00:06:42,970
accounting, followed by
233
00:06:42,970 --> 00:06:44,290
insurance, retail and
234
00:06:44,290 --> 00:06:45,089
healthcare.
235
00:06:45,089 --> 00:06:47,970
The patent isn't mysterious
because finance work was already
236
00:06:47,970 --> 00:06:50,610
rule based and documented long
before anyone called it
237
00:06:50,610 --> 00:06:53,949
automation, which made it
unusually ready to hand over
238
00:06:53,949 --> 00:06:55,889
everything up to this point is
the good news.
239
00:06:55,889 --> 00:06:57,370
The good news has a hard edge,
240
00:06:57,370 --> 00:06:58,589
because the qualities that make
241
00:06:58,589 --> 00:07:00,290
a task perfect for a bot are the
242
00:07:00,290 --> 00:07:01,329
same qualities that make it
243
00:07:01,329 --> 00:07:02,959
fragile in ways nobody plans
244
00:07:02,959 --> 00:07:04,000
for.
245
00:07:04,000 --> 00:07:05,199
Where RPA breaks.
246
00:07:05,199 --> 00:07:06,560
The limit I mentioned earlier is
247
00:07:06,560 --> 00:07:08,040
worth naming now, and it isn't
248
00:07:08,040 --> 00:07:09,160
really about the software being
249
00:07:09,160 --> 00:07:10,839
badly written, because every bot
250
00:07:10,839 --> 00:07:12,079
assumes the work always looks
251
00:07:12,079 --> 00:07:12,680
the same.
252
00:07:12,680 --> 00:07:14,040
Real work almost never does.
253
00:07:14,040 --> 00:07:16,259
Start with the data a bot reads
254
00:07:16,259 --> 00:07:18,199
a field, a cell, or a labeled
255
00:07:18,199 --> 00:07:19,839
box on a screen, and it needs
256
00:07:19,839 --> 00:07:20,920
that structure to actually
257
00:07:20,920 --> 00:07:21,680
exist.
258
00:07:21,680 --> 00:07:23,399
Hand it a scanned invoice, a PDF
259
00:07:23,399 --> 00:07:24,519
with the total in a slightly
260
00:07:24,519 --> 00:07:26,040
different place, or an email
261
00:07:26,040 --> 00:07:27,199
where someone buried the amount
262
00:07:27,199 --> 00:07:28,600
inside a paragraph and it has
263
00:07:28,600 --> 00:07:30,180
nothing to grab anything
264
00:07:30,180 --> 00:07:31,879
arriving as free text, an image
265
00:07:31,879 --> 00:07:33,319
or handwriting tends to stop a
266
00:07:33,319 --> 00:07:34,120
bot dead.
267
00:07:34,120 --> 00:07:35,480
Then there are the exceptions.
268
00:07:35,480 --> 00:07:37,000
Every real process has edge
269
00:07:37,000 --> 00:07:38,839
cases, and a bot handles the
270
00:07:38,839 --> 00:07:39,959
standard path and only the
271
00:07:39,959 --> 00:07:40,720
standard path.
272
00:07:40,720 --> 00:07:43,439
No matter how many exceptions,
show up every week when an order
273
00:07:43,439 --> 00:07:45,079
doesn't match the usual pattern.
274
00:07:45,079 --> 00:07:46,519
The work doesn't slow down, it
275
00:07:46,519 --> 00:07:48,240
just stops and it usually stops
276
00:07:48,240 --> 00:07:49,160
silently.
277
00:07:49,160 --> 00:07:50,160
Brittleness is the third
278
00:07:50,160 --> 00:07:51,600
problem, and this one catches
279
00:07:51,600 --> 00:07:52,860
people off guard because a bot
280
00:07:52,860 --> 00:07:53,879
drives the surface of an
281
00:07:53,879 --> 00:07:54,600
application.
282
00:07:54,600 --> 00:07:56,759
A small change to that surface
can break it.
283
00:07:56,759 --> 00:08:00,480
A button moves a field, gets
renamed, a login screen adds a
284
00:08:00,480 --> 00:08:03,759
step, and the bot that ran
perfectly for six months fails
285
00:08:03,759 --> 00:08:04,759
on a Monday morning.
286
00:08:04,759 --> 00:08:07,560
Scale is the fourth problem, and
it's what ends most programs.
287
00:08:07,560 --> 00:08:09,000
The research on this is blunt
288
00:08:09,000 --> 00:08:09,920
and more than half of the
289
00:08:09,920 --> 00:08:11,120
customers surveyed say they
290
00:08:11,120 --> 00:08:12,079
struggle to grow their
291
00:08:12,079 --> 00:08:13,519
automation beyond the first few
292
00:08:13,519 --> 00:08:14,639
processes.
293
00:08:14,639 --> 00:08:17,060
Their yardstick for an advanced
program is one hundred live
294
00:08:17,060 --> 00:08:19,639
bots, and most companies never
get past ten.
295
00:08:19,639 --> 00:08:22,600
This is also where the confusion
with AI agents creeps in.
296
00:08:22,600 --> 00:08:24,319
When a bot hits something it
wasn't taught.
297
00:08:24,319 --> 00:08:26,720
Someone has to decide what
happens next, and that decision
298
00:08:26,720 --> 00:08:29,439
is the part people keep hoping a
smarter model will handle.
299
00:08:29,439 --> 00:08:30,959
Sometimes a model really can.
300
00:08:30,959 --> 00:08:34,000
Often the honest answer is that
the process itself was never
301
00:08:34,000 --> 00:08:36,240
defined well enough for any tool
to follow it.
302
00:08:36,240 --> 00:08:38,159
Put those four limits together
and.
303
00:08:38,159 --> 00:08:39,679
The conclusion is uncomfortable,
304
00:08:39,679 --> 00:08:41,159
but useful for anyone planning
305
00:08:41,159 --> 00:08:41,960
this work.
306
00:08:41,960 --> 00:08:44,039
This is not a way to fix a messy
process.
307
00:08:44,039 --> 00:08:45,679
It's a way to copy a tidy one,
308
00:08:45,679 --> 00:08:47,159
and it stays reliable only for
309
00:08:47,159 --> 00:08:48,419
as long as the tidy version of
310
00:08:48,419 --> 00:08:49,639
that process holds up in
311
00:08:49,639 --> 00:08:50,320
practice.
312
00:08:50,320 --> 00:08:52,720
If the limits are this well
documented, you would expect
313
00:08:52,720 --> 00:08:54,279
companies to steer around them.
314
00:08:54,279 --> 00:08:55,679
And yet plenty of companies walk
315
00:08:55,679 --> 00:08:57,080
straight into all four of them
316
00:08:57,080 --> 00:08:57,960
anyway.
317
00:08:57,960 --> 00:08:59,240
The reasons have almost nothing
318
00:08:59,240 --> 00:09:01,440
to do with technology the
319
00:09:01,440 --> 00:09:02,950
mistakes that sink automation
320
00:09:02,950 --> 00:09:03,750
projects.
321
00:09:03,750 --> 00:09:05,149
The first mistake is automating
322
00:09:05,149 --> 00:09:06,350
something that should have been
323
00:09:06,350 --> 00:09:07,029
deleted.
324
00:09:07,029 --> 00:09:08,610
A bad process that runs by hand
325
00:09:08,610 --> 00:09:10,029
is annoying and visible, while
326
00:09:10,029 --> 00:09:11,350
the same process behind a bot
327
00:09:11,350 --> 00:09:13,429
runs faster, fails more quietly,
328
00:09:13,429 --> 00:09:14,850
and becomes much harder to argue
329
00:09:14,850 --> 00:09:15,429
with.
330
00:09:15,429 --> 00:09:18,350
Now it looks like a system
instead of a bad habit.
331
00:09:18,350 --> 00:09:20,470
The second mistake is launching
a bot and walking away.
332
00:09:20,470 --> 00:09:23,389
Someone has to own it because
when a screen changes or the
333
00:09:23,389 --> 00:09:26,389
volume doubles, a person needs
to notice and fix it.
334
00:09:26,389 --> 00:09:28,870
Programs that skip this step
find out their bots are broken
335
00:09:28,870 --> 00:09:31,629
at month end, when the numbers
no longer add up and nobody can
336
00:09:31,629 --> 00:09:32,389
explain the gap.
337
00:09:32,389 --> 00:09:33,750
The third mistake is leaving no
338
00:09:33,750 --> 00:09:35,230
path for the cases a bot can't
339
00:09:35,230 --> 00:09:35,789
handle.
340
00:09:35,789 --> 00:09:37,110
If the only options are a
341
00:09:37,110 --> 00:09:38,990
perfect run or a silent stop,
342
00:09:38,990 --> 00:09:40,269
then every unusual document
343
00:09:40,269 --> 00:09:41,909
becomes a lost invoice and the
344
00:09:41,909 --> 00:09:43,470
humans around the bot start
345
00:09:43,470 --> 00:09:44,750
working around it instead of
346
00:09:44,750 --> 00:09:45,350
with it.
347
00:09:45,350 --> 00:09:46,590
A five minute conversation about
348
00:09:46,590 --> 00:09:48,029
who handles the weird cases
349
00:09:48,029 --> 00:09:49,110
would have prevented the whole
350
00:09:49,110 --> 00:09:49,669
mess.
351
00:09:49,669 --> 00:09:51,990
The fourth mistake is measuring
the wrong thing.
352
00:09:51,990 --> 00:09:54,789
Hours saved is a lovely number
for a slide, and it says nothing
353
00:09:54,789 --> 00:09:57,149
about whether the work actually
left somebody's desk at the end
354
00:09:57,149 --> 00:09:57,549
of the week.
355
00:09:57,549 --> 00:09:58,909
If the same person still spends
356
00:09:58,909 --> 00:10:00,110
Friday afternoon checking the
357
00:10:00,110 --> 00:10:02,029
bots output, then no time was
358
00:10:02,029 --> 00:10:02,669
saved.
359
00:10:02,669 --> 00:10:04,190
the time just moved somewhere
else.
360
00:10:04,190 --> 00:10:07,070
The fifth mistake is treating
access as an afterthought.
361
00:10:07,070 --> 00:10:09,950
A bot usually logs in as a real
person or a shared account, so
362
00:10:09,950 --> 00:10:12,389
it inherits everything that
account can see and do.
363
00:10:12,389 --> 00:10:13,649
When nobody tracks which bot
364
00:10:13,649 --> 00:10:15,029
holds which access, you end up
365
00:10:15,029 --> 00:10:16,190
with automation that reaches
366
00:10:16,190 --> 00:10:17,750
further than anyone intended and
367
00:10:17,750 --> 00:10:19,070
no clean way to explain who
368
00:10:19,070 --> 00:10:19,750
approved it.
369
00:10:19,750 --> 00:10:21,350
Notice what's missing from that
list?
370
00:10:21,350 --> 00:10:23,990
Not one of those failures comes
from a badly built bot.
371
00:10:23,990 --> 00:10:25,429
They are process failures.
372
00:10:25,429 --> 00:10:27,830
Ownership failures and
measurement failures.
373
00:10:27,830 --> 00:10:29,389
Wearing a technology costume.
374
00:10:29,389 --> 00:10:31,230
And that is why buying newer
375
00:10:31,230 --> 00:10:33,070
software almost never fixes any
376
00:10:33,070 --> 00:10:33,669
of them.
377
00:10:33,669 --> 00:10:36,509
All of this matters more once
you realize something simple.
378
00:10:36,509 --> 00:10:37,669
The tool is probably already
379
00:10:37,669 --> 00:10:38,950
sitting inside your Microsoft
380
00:10:38,950 --> 00:10:40,429
license, and it is waiting for
381
00:10:40,429 --> 00:10:41,870
someone in your team to point it
382
00:10:41,870 --> 00:10:43,190
at a task that deserves the
383
00:10:43,190 --> 00:10:44,230
effort.
384
00:10:44,230 --> 00:10:45,629
Where RPA fits in Microsoft
385
00:10:45,629 --> 00:10:46,629
three hundred sixty five and
386
00:10:46,629 --> 00:10:47,149
Azure.
387
00:10:47,149 --> 00:10:49,149
If your company runs on
Microsoft three hundred sixty
388
00:10:49,149 --> 00:10:51,470
five, the automation tool is
already in the building, and
389
00:10:51,470 --> 00:10:53,190
it's called Power Automate.
390
00:10:53,190 --> 00:10:56,149
The RPA part has its own name,
which is desktop flows.
391
00:10:56,149 --> 00:10:57,330
You build them by dragging
392
00:10:57,330 --> 00:10:58,750
actions into a sequence or by
393
00:10:58,750 --> 00:10:59,830
recording yourself doing the
394
00:10:59,830 --> 00:11:01,389
task once, which is the same
395
00:11:01,389 --> 00:11:03,059
approach every other RPA product
396
00:11:03,059 --> 00:11:04,059
takes power.
397
00:11:04,059 --> 00:11:05,259
Automate holds two kinds of
398
00:11:05,259 --> 00:11:06,139
flow, and the difference
399
00:11:06,139 --> 00:11:06,580
matters.
400
00:11:06,580 --> 00:11:09,500
Cloud flows run in Microsoft's
cloud and connect services to
401
00:11:09,500 --> 00:11:12,220
each other, so a new form
response can create a record,
402
00:11:12,220 --> 00:11:15,580
send a message, and update a
list on its own desktop flows,
403
00:11:15,580 --> 00:11:18,299
run on a machine and drive the
screen, which is what lets them
404
00:11:18,299 --> 00:11:20,460
handle software with no modern
connection at all.
405
00:11:20,460 --> 00:11:23,620
Real automation usually uses
both with a cloud flow starting
406
00:11:23,620 --> 00:11:26,340
the work and a desktop flow
finishing it on an older system.
407
00:11:26,340 --> 00:11:28,139
Here's what that looks like end
to end.
408
00:11:28,139 --> 00:11:31,139
Someone submits an expense form,
a cloud flow catches the
409
00:11:31,139 --> 00:11:33,940
response, and it checks the
amount against the policy.
410
00:11:33,940 --> 00:11:35,659
If the amount looks normal, the
411
00:11:35,659 --> 00:11:37,419
cloud flow hands the record to a
412
00:11:37,419 --> 00:11:39,460
desktop flow, which logs into
413
00:11:39,460 --> 00:11:41,220
the finance system that has no
414
00:11:41,220 --> 00:11:42,659
modern connection and enters the
415
00:11:42,659 --> 00:11:43,299
claim.
416
00:11:43,299 --> 00:11:46,259
If anything looks off the flow,
routes it to a person instead.
417
00:11:46,259 --> 00:11:48,659
Nobody watches the whole thing
happen in real time, and the
418
00:11:48,659 --> 00:11:50,940
trail of what occurred stays in
the run history.
419
00:11:50,940 --> 00:11:54,299
For anyone who wants to check
under the hood, Power Automate
420
00:11:54,299 --> 00:11:57,059
is built on a service called
Azure Logic Apps.
421
00:11:57,059 --> 00:11:59,220
The two look similar on the
surface, and the difference is
422
00:11:59,220 --> 00:12:00,860
mostly about who they're for.
423
00:12:00,860 --> 00:12:03,740
Power automate is aimed at
makers and everyday users inside
424
00:12:03,740 --> 00:12:06,740
Microsoft three hundred sixty
five with per user licensing.
425
00:12:06,740 --> 00:12:08,019
While Logic Apps is aimed at
426
00:12:08,019 --> 00:12:09,500
developers and IT teams working
427
00:12:09,500 --> 00:12:10,379
in Azure.
428
00:12:10,379 --> 00:12:12,659
If you're a beginner, you'll
live in Power Automate and you
429
00:12:12,659 --> 00:12:15,259
can grow into logic apps later
without starting over.
430
00:12:15,259 --> 00:12:16,980
The part everyone asks about is
what it costs.
431
00:12:16,980 --> 00:12:19,659
Many Microsoft three hundred
sixty five plans already include
432
00:12:19,659 --> 00:12:22,659
standard connectors, so basic
cloud flows are often covered.
433
00:12:22,659 --> 00:12:25,740
Attended desktop flows sit in
the premium plan, which lists at
434
00:12:25,740 --> 00:12:28,700
fifteen dollars per user per
month when you pay yearly.
435
00:12:28,700 --> 00:12:31,620
Unattended bots are licensed per
bot instead of per person, at
436
00:12:31,620 --> 00:12:34,039
one hundred and fifty dollars a
month, and that price rises to
437
00:12:34,039 --> 00:12:36,379
two hundred and fifteen dollars
if Microsoft hosts the machine
438
00:12:36,379 --> 00:12:37,740
for you on Azure.
439
00:12:37,740 --> 00:12:39,179
That per bot licensing is worth
440
00:12:39,179 --> 00:12:40,259
sitting with for a moment
441
00:12:40,259 --> 00:12:41,740
because it explains a decision
442
00:12:41,740 --> 00:12:42,899
you will see inside real
443
00:12:42,899 --> 00:12:43,299
companies.
444
00:12:43,299 --> 00:12:46,419
One bot can run one unattended
task at a time, which means
445
00:12:46,419 --> 00:12:49,059
running five things at the same
time turns into an invoice for
446
00:12:49,059 --> 00:12:50,419
five separate bots.
447
00:12:50,419 --> 00:12:53,139
Starting is cheap, and scaling
is where the invoice grows.
448
00:12:53,139 --> 00:12:54,659
Around all of that sits the
449
00:12:54,659 --> 00:12:56,019
wider story Microsoft keeps
450
00:12:56,019 --> 00:12:56,620
telling.
451
00:12:56,620 --> 00:12:59,419
Process mining finds the work
worth automating by reading the
452
00:12:59,419 --> 00:13:02,340
trails your systems already
leave behind, and copilot and
453
00:13:02,340 --> 00:13:04,570
agents add judgment where the
rules run out.
454
00:13:04,570 --> 00:13:07,450
That's the direction of travel,
and it deserves its own episode.
455
00:13:07,450 --> 00:13:09,250
Knowing the tool exists is easy.
456
00:13:09,250 --> 00:13:11,490
Knowing what to do with it on a
Monday morning when nobody has
457
00:13:11,490 --> 00:13:13,970
told you which tasks are worth
automating, is where most
458
00:13:13,970 --> 00:13:16,250
beginners stall out and stop.
459
00:13:16,250 --> 00:13:17,850
What this means for your career.
460
00:13:17,850 --> 00:13:18,970
The skill worth building here
461
00:13:18,970 --> 00:13:20,210
has nothing to do with building
462
00:13:20,210 --> 00:13:20,929
bots.
463
00:13:20,929 --> 00:13:22,409
It's the ability to spot bot
464
00:13:22,409 --> 00:13:23,929
shaped work and describe it
465
00:13:23,929 --> 00:13:25,210
clearly enough that someone else
466
00:13:25,210 --> 00:13:26,490
can take it and act on it
467
00:13:26,490 --> 00:13:27,809
without asking you a dozen
468
00:13:27,809 --> 00:13:28,570
questions.
469
00:13:28,570 --> 00:13:30,529
Here's a simple exercise you can
run this week.
470
00:13:30,529 --> 00:13:31,970
Pick one task you do every week
471
00:13:31,970 --> 00:13:33,129
that runs through the same steps
472
00:13:33,129 --> 00:13:34,570
in the same order, and write
473
00:13:34,570 --> 00:13:35,730
those steps down as if you were
474
00:13:35,730 --> 00:13:36,730
explaining them to a new
475
00:13:36,730 --> 00:13:37,129
colleague.
476
00:13:37,129 --> 00:13:39,610
If the list comes out clean,
that task is a real candidate
477
00:13:39,610 --> 00:13:40,330
for a bot.
478
00:13:40,330 --> 00:13:42,289
If you keep stopping to write
that, it depends.
479
00:13:42,289 --> 00:13:44,529
You found the part that would
break it, and that's useful
480
00:13:44,529 --> 00:13:45,450
information to that.
481
00:13:45,450 --> 00:13:47,289
Exercise is also why this skill
travels.
482
00:13:47,289 --> 00:13:50,289
The people who get real value
from automation are the ones who
483
00:13:50,289 --> 00:13:53,129
can explain a process clearly
enough that a machine and a
484
00:13:53,129 --> 00:13:54,929
colleague both follow it.
485
00:13:54,929 --> 00:13:57,370
That ability comes up in job
interviews far more often than
486
00:13:57,370 --> 00:13:59,090
any specific certification.
487
00:13:59,090 --> 00:14:00,450
One honest note before you
488
00:14:00,450 --> 00:14:01,950
automate everything in sight,
489
00:14:01,950 --> 00:14:04,110
this is a useful edge and not a
490
00:14:04,110 --> 00:14:05,610
guaranteed promotion, and it
491
00:14:05,610 --> 00:14:06,649
doesn't replace knowing the
492
00:14:06,649 --> 00:14:07,330
business.
493
00:14:07,330 --> 00:14:08,570
A bot that saves four hours a
494
00:14:08,570 --> 00:14:10,110
week on a process nobody needed
495
00:14:10,110 --> 00:14:11,809
is still a wasted project, and
496
00:14:11,809 --> 00:14:12,809
the person who understands the
497
00:14:12,809 --> 00:14:14,370
business usually catches that
498
00:14:14,370 --> 00:14:15,649
first.
499
00:14:15,649 --> 00:14:17,889
The short version and one
question for you.
500
00:14:17,889 --> 00:14:19,250
The short version is simple.
501
00:14:19,250 --> 00:14:20,450
The software clicks and types
502
00:14:20,450 --> 00:14:21,649
the way a person does on the
503
00:14:21,649 --> 00:14:23,009
same screens with no AI
504
00:14:23,009 --> 00:14:23,450
involved.
505
00:14:23,450 --> 00:14:26,690
It's genuinely good at high
volume, rule based work with
506
00:14:26,690 --> 00:14:29,690
tidy data, and it breaks the
moment the work stops looking
507
00:14:29,690 --> 00:14:30,929
the same way twice.
508
00:14:30,929 --> 00:14:32,330
Getting started is cheap.
509
00:14:32,330 --> 00:14:35,529
The bot licensing only kicks in
once you go unattended and the
510
00:14:35,529 --> 00:14:36,809
whole thing lives inside.
511
00:14:36,809 --> 00:14:39,409
Power Automate If your company
already runs on Microsoft three
512
00:14:39,409 --> 00:14:40,490
hundred sixty five.
513
00:14:40,490 --> 00:14:42,009
Most programs stall for reasons
514
00:14:42,009 --> 00:14:43,210
that sit outside the bots
515
00:14:43,210 --> 00:14:43,809
entirely.
516
00:14:43,809 --> 00:14:46,450
Nobody owned the process around
them, so nothing got fixed when
517
00:14:46,450 --> 00:14:49,129
the work changed and the bots
took the blame for a problem
518
00:14:49,129 --> 00:14:50,129
they never caused.
519
00:14:50,129 --> 00:14:52,169
So here is the question I'll
leave you with.
520
00:14:52,169 --> 00:14:53,450
Which single task in your own
521
00:14:53,450 --> 00:14:54,570
week would you hand to a bot
522
00:14:54,570 --> 00:14:54,970
tomorrow?
523
00:14:54,970 --> 00:14:55,970
And which part of it would you
524
00:14:55,970 --> 00:14:57,370
still want a human to check if
525
00:14:57,370 --> 00:14:58,529
this made the topic feel less
526
00:14:58,529 --> 00:14:59,490
intimidating?
527
00:14:59,490 --> 00:15:01,289
Subscribe to M365 FM.
528
00:15:01,289 --> 00:15:02,289
To master the tools and the
529
00:15:02,289 --> 00:15:03,809
mindset shaping the future of
530
00:15:03,809 --> 00:15:04,450
work.
