In this episode of Leap to Scale, Greg Ross-Munro and Justin Davis cut through the AI hype and talk about what service businesses can actually do today to get real value from AI. They unpack why many companies are using AI in “back-of-house” tasks like emails, RFPs, and internal operations, but haven’t yet embedded it into their service delivery or customer-facing experiences. The conversation explores practical, low-risk starting points like customer support chat, meeting summaries, and document search, plus a hands-on path to adoption using custom GPTs and workflow automation tools. If you’re a service business leader trying to move from curiosity to capability, this episode gives you a clear place to start.
Episode notes
- Why most service companies are using AI operationally, but not yet embedding it into service delivery
- A look at adoption barriers: risk, governance, trust, job security concerns, and “human-in-the-loop” requirements
- “Hands work vs head work” and why AI changes what can be automated in knowledge work
- Two low-risk AI use cases that deliver immediate value: error reduction and document search with citations
- “Least sexy” AI wins: invoice processing, reconciliation, QA automation, and visual quality control
- Where to start if you have no AI in your organization: customer support tools, meeting summaries, and note-taking
- A practical path from experimentation to automation using custom GPTs and tools like Make.com or n8n
- Why AI fluency across your team matters more than one big AI project
Episode Transcript
AI for Service Businesses: From Hype to Hands-On
Greg Ross-Munro:
Hello, everybody. I’m Greg Ross-Munro. I’m the CEO of Sourcetoad. And today we’re joined by…
Justin Davis:
Justin Davis, Vice President of User Experience at Sourcetoad. It does sound fancy, yeah. You know, I’m actually having a 60 Minute IPA from Dogfish Head, which I have not had in maybe 10 years or so. Not a sponsor of the show, I’m not affiliated, but Sam Calagione, if you wanna give us a call. I don’t even know if he still owns it, it was probably bought by somebody at some point.
Greg Ross-Munro:
That sounds fancy, Justin. Hi. Happy Friday. Do you have a drink?
Not a sponsor of the show, but delicious.
Justin Davis:
Yeah, it’s a good IPA. What about you?
Greg Ross-Munro:
Water, unfortunately. I didn’t think far enough ahead. I was trying to get my headphones right.
Justin, we’re a professional service company. We work for a professional services company. Most of our customers and clients are professional services companies as well, which is why Sara told us to take this topic on today. AI in service businesses seems like something that we’re trying to do all the time for ourselves.
But I feel like, and this is just an anecdotal feeling, that not a lot of our clients are making AI the primary goal of what they’re doing. We’re using AI in a lot of what they’re building, but it still isn’t, for the majority of even our clients, the primary goal. I see that more on the product side. Would you agree with that?
Justin Davis:
Yeah, no, I think that’s absolutely true. Yeah. I mean, I think even as popular as it is among those clients, it’s not their first priority right now.
Greg Ross-Munro:
I saw a statistic from a PwC thing that said 70% of businesses say that AI has improved their productivity already, but only 21% have scaled deployments across all their teams. What is that like? To me, I would say that hints at the fact that they’re using AI more operationally, right? Like they’re using it for RFPs, they’re using it to respond to like write emails quicker or whatever, but they’re really not embedding it in and productizing their business. Like the actual core service is still being handled by human beings.
Justin Davis:
Yeah, I think that, yeah. Back of house, front of house, right? Like back of house operations using it more, and then extending it out to the front of the house.
I think there’s also probably a lot of people looking at it going, wow, this is overwhelming. It seems important and it seems like it can do a lot, and I hear a lot of great things, but I’m not exactly sure how to do this. Like, how do I actually put this into practice?
And I think there are a lot of reasons for that. There’s some mystery around it. There’s political stuff. There’s governance stuff. There’s psychological stuff. There are cultural things. There’s a lot in the mix there in terms of adoption. It’s a little slower uptake, especially on the front-of-house side, for sure.
Greg Ross-Munro:
Yeah, I have a slightly different take, but also kind of on your adoption point there. If you are a service business and the core of generally what you’re doing is human-driven, there’s a certain amount of job security at risk here.
And there’s also the human-in-the-loop problem. Some of these jobs are extremely important. Like if your tax attorney screws something up because an AI hallucinated, that’s not going to end well.
So a lot of the actual work that’s being done, I think, is people are very nervous about farming it off to an AI. Maybe the thing is that a lot of the time they don’t want to take the intelligence part out of the loop. They want to take the procedural stuff out. By that I mean intake of a new client, process an invoice quicker, those low-hanging fruit.
Or they have some huge spreadsheet. A lot of businesses work on some huge spreadsheet that has all these calculations, and that’s procedural mathematical stuff that works really, really well. So where do you think there might be little jumps between those procedural things?
I know I can make my business five times more efficient if I can turn this spreadsheet into a self-service SaaS product and have my customers log in and do this part themselves and charge a subscription. Great.
That’s what we do. That is a no-brainer. That’s going to pay for itself.
But then where do you put AI into a system that is going to analyze something like that? What would you do? Do I sound like a crazy person?
Justin Davis:
Like say more about what you mean by how it will analyze something.
Greg Ross-Munro:
Yeah, okay. Let’s say, for example, I’ve got a giant spreadsheet that runs my business for balancing planes. I’ve got a bunch of cargo planes that are converted. I’ve got old commercial passenger jets that I’ve converted into cargo planes.
I’ve got a rail system. I have to balance them all out. I’ve got a spreadsheet that shows where I need to put certain weights inside the plane so the plane is well balanced and flies.
Now I’ve got a device, an app on my iPad, which I can do it much quicker and easier on. I can send it to more people and nobody screws with my spreadsheets.
So someone comes to you and says, “Oh, you should be using AI to do this.” I don’t know. Is that a good use case?
Justin Davis:
One way that I think about this is Jared Spool, who created UI User Interface Engineering, used to talk about two types of work that we do, which is head work and hands work.
I kind of think of that as hands work, and that’s where automation has primarily lived. A lot of traditional automation has been moving data around, copying and pasting, and doing procedural pieces.
LLMs and AI are really the first time that we have been able to think about automating head work. I think that’s interesting from a couple different angles.
One is that it’s hard to think about it. A lot of major shifts in technology go through a phase where people go, this is new and such a different way to think that we all have to get our heads around what this means.
Another piece is that as humans, head work is where a lot of knowledge workers feel their value lives. Not necessarily in hands work.
So there’s resistance because it pushes on the value we thought was unique to us.
Greg Ross-Munro:
Dude, I’m looking at what I use it for to do my job and I’m like, maybe I’m not going to write a speech by hand ever again. I’m going to go first to the AI and get that to outline it for me. Maybe I’m not that good of a public speaker. Never was.
Justin Davis:
Right. It changes the nature of work.
But here is a practical takeaway. You could do an exercise with your teams and get on a whiteboard and do the “how do you make toast” exercise.
If you do that for some of your main business processes and map them out, you can label the parts: is this head work or hands work?
Hands work can often be automated with traditional automation. Head work is where we can ask what type of thinking is happening. Is it analysis? Report generation? Summarization? Spotting patterns?
Then you can think about what model and what prompt and what kind of context engineering you might use with an LLM to solve that problem in an automated way.
Greg Ross-Munro:
Yeah, I love that. I think there are two “how to make toast” things service companies should always be thinking about.
One is error reduction. If you have a traditional process, having an AI look at the result and call out potential issues is a really easy place to start. The downside of false positives is low.
The other is reading a ton of documentation and making it searchable, and being able to retrieve data later. If it returns the answer, it should also return a link back to the original documents so someone can fact check it.
Those are two easy low-hanging fruits for building systems into pre-existing procedures.
But there’s also a lot of off-the-shelf stuff. So, what is the least sexy AI use case that has interested you recently?
Justin Davis:
The least sexy AI case…
Greg Ross-Munro:
I’ll give you one. We have a client who has to process huge amounts of invoices. Three years ago, we would have written OCR and built rules for every single format, and it would have been a nightmare.
Now we’re processing tens of thousands of these invoices through the system and reconciling them in 90 seconds rather than a human taking 5 to 10 minutes. It runs all day every day and it costs three cents an invoice. Not sexy, but the time it frees up is huge.
Justin Davis:
Not sexy, but extremely useful.
For me, I’ve been thinking a lot about quality control and QA. I’ve been experimenting with using LLMs to do automated QA against web apps and websites.
It takes a lot of work out of writing manual test cases and clicking through flows. It can understand the context, see the page, and do the checks automatically.
I’m also interested in quality control with visual recognition. LLMs are getting better at recognizing objects in video. What used to be expensive and hard is becoming lower cost.
You can imagine QA on assembly lines where it flags abnormalities in packages and products.
Greg Ross-Munro:
That’s really cool. My brain goes to very dirty places thinking about what it’s going to say it finds in these manufacturing processes, but we’ll move on.
There is also so much off the shelf. Chat bots, ticket triage, email drafting tools, note takers. If somebody wasn’t using any of these in their professional service firm, what do you pick up first?
The reason I ask is because everyone should be doing something. Get reps in, learn how the tools work, see where you can trust and where you can’t.
So you own the company, you’re the COO of a financial compensation company, and you don’t have one piece of AI in your entire organization. Where are you starting?
Justin Davis:
I’m probably starting with customer support because it makes so much sense and the tooling is mature. Chat bots are a no-brainer. It’s low cost and can save a ton of time.
Internally, meeting summarization is massive. Meeting notes and transcripts are critical.
And data is really important now. Recording meetings and conversations is rich with information.
Before LLMs, it was kind of useless to record meetings. No one goes back and rewatch them. Everybody says they will. No one ever does.
But now you can ask questions against those meetings and it makes everything more efficient.
Greg Ross-Munro:
We record every meeting because sometimes somebody says you didn’t say something. We scrubbed through hundreds of hours of meeting notes to find out, this is where we said it and you agreed.
Justin Davis:
Right, and then we do. It is true.
Greg Ross-Munro:
This segues nicely into the next point. How do you start to grow this from a development standpoint? How do you start looking at processes you can automate?
Finish my sentence? No, you please. Sorry. You go ahead.
Justin Davis:
You know what I think you should do. Start using ChatGPT to do those things. Don’t build stuff yet.
Greg Ross-Munro:
That’s what I was going to say. Go learn how to build GPTs, which is a terrible name.
If you go and look up how to build a GPT, you can go into ChatGPT, click GPTs on the left-hand side, and create a tool. You give it instructions.
For example: “You are a financial analyst. I’m going to feed you a transcript of a conversation we had with a new client. Based on everything in your instructions, build the framework for your first report.”
If you use that GPT over and over, then the next question is, what’s the next GPT I can build?
Maybe that produces the summary, then you build another one that takes that and puts it into your document form, or builds the spreadsheet you need.
If you have a few of these, you can start stringing them together with a tool like Make.com or n8n. Or you hand it over to a company like us and we use something like Langflow.
Then that becomes the basis of your automation using LLMs.
Tell me I’m wrong. Was that what you were going to say?
Justin Davis:
Yeah, no, you’re dead on. That’s exactly it.
In some ways, there’s a metaphor here where ChatGPT is like Excel in the automation process. Often when a business starts to automate a thing, you turn to Excel, build formulas, and prototype.
ChatGPT lets you prototype different types of work than you can do in Excel, but the point is the same. You don’t have to invest a lot of money to start.
Your entire organization should be fluent with AI tools and encouraged to use them in daily work. Take care of security and governance, but participate in the wave.
As people use it, they will discover things you can’t think of, and those ideas will bubble up through the organization.
Greg Ross-Munro:
Yeah, I couldn’t have said it better myself. Go build GPTs. Get everybody an OpenAI or Claude account. It’s like 20 bucks a month.
It’ll be the best thing you invest in your entire business ever.
If you’re not paying $20 a month to help build Sam Altman a better bunker for when the apocalypse comes, you are doing your business a disservice.
Help Sam build that bunker and get your team on the AI bandwagon. Get some training for them around safety and policy and all that stuff.
Do not be afraid of these tools. Start using them yesterday if you can.
All right. Well, that was good. Thank you for your time. I’ll see you next time. See you next week.
Justin Davis:
Absolutely. Here to talk AI or anything anytime. Looking forward to the next time. See you guys.
Greg Ross-Munro:
Everybody take care.