AI in Business Systems: Smart Integration Vs. Dangerous Assumptions

Anthony Perl: 00:00:02

AI in business systems, smart integration

2

: 00:00:06

versus dangerous assumptions.

3

: 00:00:08

In this episode, John Tokin from Brain

In A Box navigates the promises and

4

: 00:00:12

pitfalls of artificial intelligence

in systems and development.

5

: 00:00:16

While AI offers incredible

capabilities for creating code

6

: 00:00:19

and filling software gaps.

7

: 00:00:21

John reveals why relying on it to

build from scratch can be dangerous.

8

: 00:00:26

You'll learn how AI is

being used effectively for

9

: 00:00:29

creating smart spreadsheets.

10

: 00:00:31

Discover why AI's generic

outputs look polished, but lack

11

: 00:00:35

crucial context and understand

the critical elements it misses.

12

: 00:00:39

Like tax treatments and cultural nuances.

13

: 00:00:42

Learn how to harness AI's genuine

strengths while avoiding the trap

14

: 00:00:46

of assuming it understands your

unique environment and discover

15

: 00:00:49

practical integration approaches.

16

: 00:00:52

I'm your co-host Anthony Pearl, and this

is the Systems From The Box Podcast.

17

: 00:00:56

Let's start unpacking, John.

18

: 00:00:59

This is an important conversation

that I think we need to have,

19

: 00:01:02

which is AI for systems.

20

: 00:01:05

Is it a good thing or not?

21

: 00:01:08

Because the rise of AI is absolutely here.

22

: 00:01:12

People are jumping at it and using it.

23

: 00:01:15

What's your take on how you

use it in within systems?

24

: 00:01:18

So

25

: 00:01:18

John Tonkin: we use

26

: 00:01:19

Anthony Perl: AI a lot.

27

: 00:01:21

John Tonkin: We use it for creating code.

28

: 00:01:24

So in addition to capturing systems,

you know, the process, flowcharts,

29

: 00:01:27

whatever you want to think of it as.

30

: 00:01:29

We also help our clients build props.

31

: 00:01:31

And props are the checklist, the

forms, and also the smart spreadsheets

32

: 00:01:35

that do this and capture that.

33

: 00:01:38

Change that data into this and so on.

34

: 00:01:41

And it, it's really good because it

can fill in the gaps between what their

35

: 00:01:45

software, their job management, or

whatever software they use does, and what

36

: 00:01:50

they need to be able to give to their

clients, or something along those lines.

37

: 00:01:54

And what happens is that we

use AI to give us that code.

38

: 00:01:58

We then check it and so on, because I

used to do a lot of coding myself, and

39

: 00:02:02

so I'll go through and check it because.

40

: 00:02:04

AI does not always

produce the right outcome.

41

: 00:02:07

So we know that it isn't

a thinking person yet.

42

: 00:02:12

Uh, but it's on the way towards thinking,

um, the moment it's just machine learning.

43

: 00:02:17

So it's done something so many

times that it'll pick up what

44

: 00:02:20

appears to be the right answer.

45

: 00:02:22

But I love that sort of little bit

down the bottom of the chat, GPT window

46

: 00:02:26

where it says, I'm reading it now.

47

: 00:02:29

Chat, GPT can make mistakes,

check important info.

48

: 00:02:33

Because I was using it earlier today.

49

: 00:02:35

So

50

: 00:02:37

AI is used in systems.

51

: 00:02:40

AI is a great thing to spark ideas.

52

: 00:02:43

So think about just about

every part of your business.

53

: 00:02:46

You'll see people talking

about how wonderful it is.

54

: 00:02:50

So we use it in marketing.

55

: 00:02:51

We don't use it in sales, but we will

be, and we're a Luddite in terms of use

56

: 00:02:56

of ai, really in terms of our systems.

57

: 00:02:58

Sadly, we use it very much for.

58

: 00:03:02

Tools we build for our clients, but

you know, we don't use it there.

59

: 00:03:05

So we are at that point where

we're looking for ways we can

60

: 00:03:09

implement it more effectively.

61

: 00:03:11

And I think that's the approach that

every business owner has to take.

62

: 00:03:14

If you are not using it, you should

be thinking, why aren't I using it?

63

: 00:03:18

What could I use AI for that

I'm not doing right now?

64

: 00:03:22

And, uh, it's right there waiting to help.

65

: 00:03:26

You've just gotta ask the right questions,

give it the right prompts, and then be

66

: 00:03:29

critical of the answers they give it.

67

: 00:03:32

They give you that the AI

gives you, but there's so much

68

: 00:03:36

fun that can come out of it.

69

: 00:03:38

So much power that's in there.

70

: 00:03:40

So there are a few mistakes though,

of course there are some faults and

71

: 00:03:44

things that people have in the way.

72

: 00:03:46

They look at ai, and I think these are

just as interesting because there's this

73

: 00:03:51

assumption that AI can do everything.

74

: 00:03:53

AI will help you.

75

: 00:03:56

Let me go and kick sand in somebody's

face, just like the ads on the

76

: 00:03:59

back of the comic books used to do.

77

: 00:04:03

So really what we want is to

have that understanding that AI

78

: 00:04:06

is fantastic if used properly.

79

: 00:04:08

It's a horrible, uh, weapon if used badly.

80

: 00:04:12

So one of the mistakes we

have is that AI can create my

81

: 00:04:16

business systems from scratch.

82

: 00:04:18

And this is something I've actually,

I was talking to a business,

83

: 00:04:22

no hints here or clues, but.

84

: 00:04:25

I was talking to a business about

working with them on their systems

85

: 00:04:28

and straight away we had the comeback.

86

: 00:04:31

Well, what we're doing now, you'd be

really pleased with this, this, John,

87

: 00:04:35

what we're doing is to get AI to, to

tell us how we, how we do everything,

88

: 00:04:39

how we manage everything, how we do our

invoicing, and how we set up a new client

89

: 00:04:43

and how we do this, how we do that.

90

: 00:04:45

And it feared filled me with dread to hear

it, frankly, because I'm thinking, well.

91

: 00:04:50

Using AI to get ideas would be one thing,

but if we just use that example of the

92

: 00:04:55

invoice, there's so much in there that if

you just say, told me how to do an invoice

93

: 00:05:00

in zero or MYOB, or whatever, the AI will

tell you, but it's not necessarily gonna

94

: 00:05:06

say, but you've gotta check that you've

got the right tax treatment, you know,

95

: 00:05:10

GST included, GST free, whatever else.

96

: 00:05:13

It's not telling you necessarily to.

97

: 00:05:15

Make sure that you are invoicing

the correct entity because one

98

: 00:05:19

business may have multiple entities.

99

: 00:05:22

It's not telling you to check that

you've used the right pricing tier in

100

: 00:05:26

preparing that invoice, the data of the

invoice, and so there's so much there.

101

: 00:05:31

We assume that because it can tell

you, here's what you need to do

102

: 00:05:35

to create an invoice, but it knows

your business, whereas it doesn't

103

: 00:05:39

know your business, so it can't

know your unique operating context.

104

: 00:05:44

Your culture and so on.

105

: 00:05:45

It doesn't know the constraints

that you are in business against,

106

: 00:05:49

you know, that you've gotta

manage, doesn't know any of that.

107

: 00:05:52

So it'll give you a generic output.

108

: 00:05:55

It'll look polished, it'll look really

good, but it's a lipstick on a pig

109

: 00:05:58

for some things that it produces.

110

: 00:06:00

You know, it's, it makes it look good,

but it's not necessarily gonna be good.

111

: 00:06:04

It's gonna lack relevance.

112

: 00:06:06

It'll be accuracy questionable.

113

: 00:06:09

It may not be prac.

114

: 00:06:10

I'll give you a practical

implementation steps for you.

115

: 00:06:14

It may just be sort of, it works as a

generic thing, but not really good for

116

: 00:06:17

you, so you've gotta have that idea.

117

: 00:06:22

Anthony Perl: Yeah, I mean, it

really is important, isn't it, to

118

: 00:06:26

understand the limitations, and

particularly when it comes to your

119

: 00:06:30

own systems, you have to understand

that it's not going to be perfect.

120

: 00:06:35

It's good as a guide, as

you say, in many respects.

121

: 00:06:38

Yeah.

122

: 00:06:39

But I know certainly you, if you talk to

someone in the legal profession, they'll

123

: 00:06:43

laugh at you if you use a document that's

generated by an AI because there's way

124

: 00:06:47

too many mistakes and things that are

in there that leave you vulnerable.

125

: 00:06:50

Yeah.

126

: 00:06:51

And you don't wanna do that

with your systems at all.

127

: 00:06:53

I think that's the

important lesson, isn't it?

128

: 00:06:54

That that to.

129

: 00:06:55

What an AI can guide you, but it

doesn't know what is the human element

130

: 00:07:00

and what is best for your business.

131

: 00:07:02

That's right.

132

: 00:07:02

It doesn't know the cultural

understandings that you might have

133

: 00:07:06

within a team, let alone where

you might physically be located.

134

: 00:07:09

There are so many variables that come

into play that completely relying on AI is

135

: 00:07:15

never gonna be something that you can do.

136

: 00:07:17

No,

137

: 00:07:18

John Tonkin: it's just a critical

thing, like it doesn't know my business.

138

: 00:07:22

'cause we rarely.

139

: 00:07:25

We train the ai, AI to do exactly

everything that our business does.

140

: 00:07:29

We don't say we do this, we do this

with and spend, you know, a half a

141

: 00:07:33

year telling it everything about it.

142

: 00:07:35

We can feed it document after document,

we can tell it to ask questions of

143

: 00:07:39

us so it understands the business,

and it'll do that very, very well.

144

: 00:07:44

And it'll produce a lot of good

output, but it's not gonna be something

145

: 00:07:47

that you can necessarily rely upon.

146

: 00:07:49

I'm sure that if we

went in there and said.

147

: 00:07:51

These are all the things

that make jet engines work.

148

: 00:07:55

Describe a and make a jet en

design a jet engine for me please.

149

: 00:08:00

It may and it may not, but it's, I would

imagine every bit is likely to create one

150

: 00:08:05

that doesn't work as one that does work

because it doesn't know all the things

151

: 00:08:09

to test all of critical parameters.

152

: 00:08:11

It may not be thinking of metal

fatigue or you know, how stressed or

153

: 00:08:14

how stressed the metals are or how

strong they have to be, and so might

154

: 00:08:18

design a beautiful one that could work.

155

: 00:08:21

Anyway, who knows?

156

: 00:08:23

But I think in systems, you know, we've

got that misunderstanding that, you

157

: 00:08:27

know, if we've got a, a flow chart there,

it must be good because I, I'm looking

158

: 00:08:32

at the flow chart and it's fantastic,

you know, and it looks really pretty

159

: 00:08:36

and I can't see any gaps in the line.

160

: 00:08:39

So it must be good, must be finished.

161

: 00:08:41

And to me, you know, it's only

as good as the tested outcome.

162

: 00:08:46

Did we actually tell the AI all the

risks that are in that process that

163

: 00:08:50

it has to manage when it's drafting

that flowchart or that system for you?

164

: 00:08:55

Did we tell it the best outcomes,

the critical things that the client

165

: 00:09:00

will expect when we give them the

outcome of that process so that

166

: 00:09:04

it can make sure that it tests the

process against those outcomes?

167

: 00:09:09

You know, there's so much that is

more than just, I want you to do

168

: 00:09:12

this, using this to create that.

169

: 00:09:16

You know, it's more than that.

170

: 00:09:17

There's gotta be so much more in there.

171

: 00:09:20

You know, a system isn't just a document,

it includes the actual behaviors,

172

: 00:09:24

responsibilities, the training, you

know, it's, it's got all of that.

173

: 00:09:27

The feedback loops, you know, every anyone

can draft a process, but implementing

174

: 00:09:32

it and embedding it into the team's

habits is a bit of, bit of a bigger job.

175

: 00:09:37

So it requires human involvement.

176

: 00:09:39

We need that in

177

: 00:09:40

Anthony Perl: there.

178

: 00:09:40

Hey John, I just wanna take a break

from the main part of the podcast

179

: 00:09:43

for a moment so you can tell us a

little bit more about Brain In A Box.

180

: 00:09:46

John Tonkin: So Brain In A Box started.

181

: 00:09:48

The idea started when I

was working in corporate.

182

: 00:09:50

Uh, I was working in a couple of

major companies, internationals, and

183

: 00:09:54

I decided that when I went out and

started my first business in: 2000

184

: 00:09:58

That it was going to use the same

principles as I had done in corporate

185

: 00:10:01

world to capture systems and to

make them accessible to the team.

186

: 00:10:05

But I thought, I've gotta be able

to do this for small business,

187

: 00:10:08

small to medium business.

188

: 00:10:09

And that's what we focus on.

189

: 00:10:11

So we work with the business

owner, with the team.

190

: 00:10:14

To capture what the business does and

then to minimize the risk by going

191

: 00:10:19

through and looking at all the risks

that we have to manage, make sure we

192

: 00:10:22

capture those and manage them effectively.

193

: 00:10:25

And also to achieve the benefit.

194

: 00:10:27

That's just effectively what it is.

195

: 00:10:29

Anthony Perl: Yeah, and, and I

think there's a keenness to use ai.

196

: 00:10:32

I mean, I think that's the, that's the

battle that you have here, isn't it?

197

: 00:10:35

And there's a difference between

using AI to produce your systems.

198

: 00:10:41

Using a and using systems to

understand where you might be able

199

: 00:10:45

to bring AI into your business.

200

: 00:10:47

That's a different, a

different slant on things.

201

: 00:10:50

And I think it's an

important one, isn't it?

202

: 00:10:51

Because that's the positivity of AI is

that you can, if you've got the systems

203

: 00:10:56

mapped out, as we've something we've

talked about in a previous episode,

204

: 00:10:59

you map out your systems properly.

205

: 00:11:01

Enables you to understand the gaps.

206

: 00:11:03

Yeah.

207

: 00:11:03

And then understand where AI might be

able to add some value, and That's right.

208

: 00:11:07

AI in a broader sense of automation

and, and other bits of technology.

209

: 00:11:11

John Tonkin: Yeah.

210

: 00:11:12

And that's, that's where AI can

come into its own like AI can, we've

211

: 00:11:16

mapped something, we've got it there.

212

: 00:11:18

We've got three adjacent processes,

configured processes, and he comes

213

: 00:11:22

in and says, you now say, could you

please look at these three processes

214

: 00:11:27

and tell me if there are any gaps?

215

: 00:11:30

The typical risks or the typical

errors we find are these 1, 2, 3,

216

: 00:11:33

4, 5, uh, the outcome must be tested

against this 1, 2, 3, 4, 5, and so on.

217

: 00:11:39

Can you see any issues or potential risks

in the processes as they are mapped at

218

: 00:11:44

the moment that could be overcome, that

we could overcome, you know, manually

219

: 00:11:49

or whatever it might be, and then

you're going to possibly get something

220

: 00:11:53

that would be really useful there?

221

: 00:11:55

You know, AI can fast track

systemization, there's no question of it.

222

: 00:11:59

But it's not going to manage it

to the point where you can hand

223

: 00:12:03

it over and say, here, do that.

224

: 00:12:05

You're still going to have to review it.

225

: 00:12:06

You're still going to have to

make sure that everything you

226

: 00:12:09

wanted to be in there is in there.

227

: 00:12:16

Anthony Perl: Yeah.

228

: 00:12:16

One of the keys is really that when you

are looking at a business is that again,

229

: 00:12:24

the AI is enabled, is a good way of.

230

: 00:12:28

Guiding you to see if there are

things that you've missed, but

231

: 00:12:31

then being thorough yourself in

going through them all and then

232

: 00:12:34

testing them and doing those things.

233

: 00:12:36

Because I think the interesting

thing about I, AI and technology

234

: 00:12:39

in a broader sense, which leads to

change in your business, is when

235

: 00:12:44

is the right time to do it as well?

236

: 00:12:46

Mm-hmm.

237

: 00:12:46

Because it's not just about jumping

on the bandwagon immediately.

238

: 00:12:51

But it's also not wanting to be the last

one to jump on the bandwagon either.

239

: 00:12:54

You've gotta find that sweet spot,

don't you, and work out where

240

: 00:12:57

that's going to come into play in

your business and add real value.

241

: 00:13:00

John Tonkin: Yeah, I think you've

gotta have a reason to do it.

242

: 00:13:02

You, I mean, anyone can wake up and

say, today I'm gonna go and throw all

243

: 00:13:07

my systems out the window and tell AI

to write them for me and do it new.

244

: 00:13:12

You know?

245

: 00:13:12

It wouldn't be a very clever thing.

246

: 00:13:15

Nothing stops us from sort of using AI to

test what we've done to give us feedback.

247

: 00:13:19

It'll give us insights.

248

: 00:13:20

We haven't.

249

: 00:13:21

Come across or haven't decided

for ourselves, but we really

250

: 00:13:25

do have to make sure that we

don't just hand it over to ai.

251

: 00:13:28

We can't hand over anything without

having it chest checked, tested.

252

: 00:13:33

Uh, we've gotta go right back to the

fundamental principles in it to make

253

: 00:13:36

sure that we actually do have a process

that's gonna be robust and going to meet

254

: 00:13:41

our needs and the needs of our clients.

255

: 00:13:44

So to me, it's great to use AI as an

adjunct to doing the systems work.

256

: 00:13:50

But not as a replacement for

anything that somebody's gonna do.

257

: 00:13:55

Anthony Perl: Yeah, I mean, I, and I

think what we've looked at as well is

258

: 00:13:58

that we've talked around, you know, not

using, not relying on AI to write your

259

: 00:14:01

systems, how you bring AI in to do things.

260

: 00:14:04

But there's also, you know, AI in

terms of adding some value in how

261

: 00:14:09

you actually produce things or help

and, and helping you and guiding

262

: 00:14:12

you along the way with your systems.

263

: 00:14:14

I think there's a difference between

press a button, let's hope that AI

264

: 00:14:17

tells us what, how we should do it.

265

: 00:14:19

Versus using some of the technology

to help record systems and to be able

266

: 00:14:24

to, you know, guide you along that

way to even point out as you're going

267

: 00:14:29

through it, you know, maybe there's

some efficiencies that we could grab

268

: 00:14:33

here, because that's one some of the

wonderful things about technology can

269

: 00:14:36

see where you are losing efficiencies.

270

: 00:14:40

I know I spoke to someone recently

where they were looking at, at

271

: 00:14:44

something that just sort of looked at

in a particular manufacturing place.

272

: 00:14:48

How much time was being

spent on toilet breaks.

273

: 00:14:52

Yeah.

274

: 00:14:53

And that it was clearly a case

of people abusing some of that

275

: 00:14:58

so-called toilet break time.

276

: 00:15:00

Yeah.

277

: 00:15:00

To go and take an actual break and

therefore, you know, accumulated over

278

: 00:15:04

a number of people over the week was

suddenly a significant amount of time.

279

: 00:15:09

And being able to use technology to be

able to see that and see where there are

280

: 00:15:14

breaks in your system is a huge advantage.

281

: 00:15:16

John Tonkin: Yeah.

282

: 00:15:17

Yep, it is.

283

: 00:15:18

And you can take those things

too far as one of our leading

284

: 00:15:22

shopping chains discovered.

285

: 00:15:24

But there are so many things where

there's an AI tool or an ai or a tool

286

: 00:15:32

that isn't ai, but which AI can point

to, to tell us how to get benefits

287

: 00:15:36

that we aren't seeing normally.

288

: 00:15:39

And I think there are, you know, it's.

289

: 00:15:42

We've gotta manage the risks in business.

290

: 00:15:44

The whole reason we can be in business

is because we manage the risks more

291

: 00:15:47

effectively than our competitors.

292

: 00:15:51

So if we can have AI help us to understand

what those risks are from its general,

293

: 00:15:57

you know, accumulated knowledge across all

the large language model, if we can have

294

: 00:16:02

it, help us understand what we could do

to streamline or to fine tune our process.

295

: 00:16:09

To make it more effective, then

we can get some value out of it.

296

: 00:16:13

We can get some real value out of it.

297

: 00:16:15

We can get it to draft something

and highlight the areas and

298

: 00:16:19

areas where we need to work

harder on it and so on like that.

299

: 00:16:23

But it isn't necessarily going

to have a structured audit

300

: 00:16:27

relevant to our business in there.

301

: 00:16:29

They're gonna be general risks, they're

gonna be generic benefits, they're

302

: 00:16:33

gonna be generic bits of software it

might re might suggest, and so on.

303

: 00:16:39

The, it's not gonna be asking, what did,

what risks does this manage, what benefits

304

: 00:16:43

does this particular process provide?

305

: 00:16:46

You know, you've, you risk building

a noncompliant system if you only

306

: 00:16:50

rely upon AI helping you in creating

in, or creating it for you, which.

307

: 00:16:57

Anthony Perl: So John, let's wrap

up this discussion on AI because we

308

: 00:17:01

could probably talk for hours about it

'cause there's so much involved in it.

309

: 00:17:04

But I want to bring it back to

the beginning question and ask

310

: 00:17:08

where you sort of see it at the

moment and where you see it going.

311

: 00:17:12

AI for systems, good thing, bad thing.

312

: 00:17:15

What's the view?

313

: 00:17:17

John Tonkin: I think AI for systems

is a fantastic thing and you just

314

: 00:17:21

have to treat it very carefully.

315

: 00:17:23

You've gotta be using it.

316

: 00:17:25

With your eyes wide open, you have to

say, alright, treat it the same way

317

: 00:17:30

as if your 12-year-old child had come

up and said, daddy, I've mapped how

318

: 00:17:34

you do this, how this system works.

319

: 00:17:37

You'd be going through and you're looking

for all the things that it does well.

320

: 00:17:40

You're looking for all the things

that it may not have done well.

321

: 00:17:43

You're testing everything you can see

to make sure that it manages all the

322

: 00:17:47

risks and achieves all the benefits

that you and your team understand.

323

: 00:17:52

But it definitely needs to be.

324

: 00:17:54

It can speed up things.

325

: 00:17:56

It can put a little angle on the

table that you didn't see before.

326

: 00:18:00

You know, it can ask a question that,

wow, that's an interesting thing.

327

: 00:18:03

I never thought of that

very important point though.

328

: 00:18:07

You know, we've gotta have that in there.

329

: 00:18:08

But really, if I were wrapping it

up, I would say, if know, we can

330

: 00:18:12

use AI to systemize, but only to

assist with the systemization.

331

: 00:18:18

You can't use it at this stage.

332

: 00:18:19

I would say definitely not to say,

here's how you run your business.

333

: 00:18:23

You might use it to say, here's what

I should be covering in my business.

334

: 00:18:27

What systems should be included,

but not exactly how to define them,

335

: 00:18:32

how they work for your business.

336

: 00:18:34

There's too much that's unique

to you and your business in your

337

: 00:18:38

country, in your area, your sector,

your industry, whatever else.

338

: 00:18:43

That need to be created by you.

339

: 00:18:45

That's my view at the moment.

340

: 00:18:46

Certainly can't wait till it changes.

341

: 00:18:50

Anthony Perl: Before you go, don't

forget to hit the subscribe button on

342

: 00:18:53

whatever platform you are tuned to.

343

: 00:18:55

Thank you for listening to the

Systems From The Box podcast featuring

344

: 00:18:58

John Tompkin from Brain In A Box.

345

: 00:19:00

Details on how to contact John and

his team are in the show notes.

346

: 00:19:04

Along with other valuable

links you don't want to miss.

347

: 00:19:07

This podcast is produced

by podcast Done for you.

348

: 00:19:10

I'm Anthony Pearl.

349

: 00:19:11

We look forward to your

company on the next episode.