Marketing for SMEs Podcast: Janey Treleaven on AI Adoption for SMEs

I sat down with Janey Treleaven from Intelligence Assist for this episode of Marketing for SMEs. I first spoke to Janey about eighteen months ago, when she was still early into AI implementation.

She’s grown a lot since then, and I went into this chat expecting her to tell me small businesses have no chance with AI. She disagreed with me completely. This one’s worth watching if you want a straight answer on where to actually start with AI, not just what everyone’s saying about it.

If you want help getting your own marketing and ads running properly, that’s what we do at Digital Goliath.

• Watch on YouTube
• Listen on your preferred podcast app

The Free Tools Trap

Jeremy Yang: Going back to that, so now on the networking scene, every week there are people coming in wanting to help small businesses with AI. I don’t think small businesses, we’re talking about a plumber with three vans or a small accounting firm, can afford the implementation.

I don’t think they can afford getting it wrong. So they’re waiting for everything to converge, and once it has, they’ll do it themselves. Love to hear your view.

Janey Treleaven: I understand why businesses feel that way, but I totally disagree. You don’t have to go and build something. AI is kind of like asking, does my business need email? Do I need a finance system?

AI is just the latest thing you have to add, and yes it comes with a cost, but it can be done very affordably using the AI already inside the tools you have.

Most businesses have chosen, just like people choose Android or iOS, whether they’re Microsoft or Google for their workspace. That’s where businesses need to start with AI adoption, using the AI capabilities in what I call your core AI tool.

That’s the day-to-day “help me with something” AI that starts the journey from doing everything yourself, to doing it with AI, to eventually, and this is where it gets expensive, the “do it for me” AI, the agentic AI. You shouldn’t start there. You’ve got to get the foundations right first.

You choose your core AI tool by considering three things: the capabilities it has, the security you need for the type of data you handle, and the knowledge it has access to. Because AI, when it relies on its trained knowledge, that’s when it gets things wrong.

Every AI model has a bias towards the people who trained it and the data they chose to train it on. That’s why American models tend to have a bias towards long dashes and a more verbose style, American professional is very different to Australian professional.

Jeremy Yang: Where’s the misconception when it comes to small business thinking it’s going to be very expensive? So basically what you’re saying is, find a tool, there’s only going to be two or three main ones.

Janey Treleaven: The first question I often ask is, do you use Microsoft or Google? If you’re in Google Workspace, the answer is probably Gemini. If you’re in Microsoft, it’s probably Copilot.

What do you want to use it for? Because the answer may actually be ChatGPT, Claude, or Perplexity depending on what you do. The security side is really important too, because the minute you start moving data around, you open yourself up to risk.

Microsoft Copilot inherits the same permissions you’ve already got set up, so you get enterprise-grade security. Claude is more capable than Copilot, but Copilot has better security because your data stays inside your enterprise environment.

Jeremy Yang: So if there’s a franchise with twenty-two stores that wants everything to remain inside, no competitors touching it, they can’t be playing around on certain LLMs?

Janey Treleaven: No, legit, no. You can, to a certain extent, if the type of data you handle is heavy on PII, you need to be very careful. You might be in breach of regulations.

Companies that do a lot of work with government have to keep data residing in Australia. With products like Copilot and ChatGPT, the data stays here, but the inference that’s happening is overseas.

But as long as, the rule I tell people, and you’ve probably heard this before, if you’re not paying for a product, you are the product.

Jeremy Yang: You guys have a product, yes. I mean advertising, a hundred percent.

Janey Treleaven: But the thing with all these free tools out there is you’re not being advertised to, so it’s not obvious. So what are they getting for it?

Nothing is free. Nothing is ever free. They must be getting something in return. One of the biggest risks at the moment to organisations is those note-taker AI bots that join meetings with you, the Read AI, Fireflies types.

Jeremy Yang: Did you ever think about that?

Janey Treleaven: What they’re getting is the video, an image of you they can use to clone, voice samples, and they could be looking in the background of your video for passwords and things.

Free is not okay. That’s the first thing I tell every business I engage with. And unless you pay even a dollar, there’s no law to protect you either, because under contract law in Australia, as long as you’ve paid a dollar for something, there’s an obligation on them to provide what they promised.

Jeremy Yang: So it’s interesting, because I’ve looked at your website and I know all the great work you’re currently doing with larger businesses too. I interviewed Chris Melotti not long ago. I thought coming in here you’d be like, small businesses, waste of time, don’t do AI. So there is demand?

Janey Treleaven: Yesterday I did a session with a branding company, they do merchandise and exhibitions for small businesses. They’re two people, working with some major clients.

It’s all about identifying, if you get time back, what are you going to do with that time? Do you want to spend more time talking to customers? So let’s work out what we can delegate to AI.

No matter the size of your business, there’s always the mundane stuff you don’t want to be doing. That’s the stuff you start asking, how do I get AI to either do this for me, or help me do it faster?

Where AI Falls Short

Jeremy Yang: When we were talking two years ago, spreadsheets weren’t even very good, you couldn’t upload stuff. One thing in my business was always analytics. I still find to this day that AI isn’t very good at joining the dots.

Janey Treleaven: Two of the key terms in AI are large language model and natural language processing. Both contain the word “language.” AI is good for the qualitative, not the quantitative, not yet.

How they’ve started to solve for that is by giving AI the ability to write and execute code. It’s like having a data scientist in your pocket. You ask the question, and it needs to be in the right mode to write the code, execute it, and give you those insights.

For anyone on Microsoft with Copilot Premium turned on, not the basic version, there’s an agent in the Microsoft 365 Copilot app called Analyst. If you go to Analyst first instead of starting a generic chat, it already has the right capability to answer questions and even generate new spreadsheets.

It’s writing Python and SQL and executing on those. It’s not relying on its trained knowledge or trying to please you.

Jeremy Yang: One experience I keep coming back to, when we send reports and quotes out, clients now run them through AI instead of reading them, and it spits out five “actionable” recommendations that are often very bad. Do you run into that a lot?

Janey Treleaven: One of the main things AI is used for is to summarise. It’s always sitting there in Gmail or Outlook asking, would you like me to summarise this? But how does it know what’s important to you?

It doesn’t have that context, so it’ll likely miss something. You have to layer in enough instructions. What do you want the summary for? What is the context it’s recommending within?

This is why letting AI models rely on trained knowledge isn’t a good idea in business. I’ll ask you a question, do you think the earth is flat?

Jeremy Yang: No.

Janey Treleaven: Do you think one times one equals two?

Jeremy Yang: No.

Janey Treleaven: Did you know there are people out there who believe both of those things, and they’ve written a lot of content about it? That content gets used by companies to train their models.

So the model’s training believes there’s a possibility the earth is flat and one times one equals two. In the moment when the AI is doing a task, reasoning and inferring from what you’ve asked, it might use that.

That’s why you have to control the AI. You have to ground it to the knowledge you want it to use, and you need to allow it to RAG (retrieval augmented generation), retrieve before it answers.

Those are your two control mechanisms. I’m going to ground you to just this knowledge and you answer only from this, or I’m going to allow you to go ask a friend, go out to the web or into my files and retrieve before you generate.

But you can’t just say go look at the World Wide Web, it’s full of fake news. You need to control it by saying, RAG only to these verified sites. If you’re asking for an opinion on something for the ATO, make sure it’s looking at the ATO website.

Don’t just ask it to go out and tell you things. You’ve got to give it the instruction, treat it like an intern, and be really clear about what the rules are. We call those guardrails.

Jeremy Yang: “Guardrails” simplifies everything you just explained into one word. But people need to know that deep explanation, because when you just say guardrails, people move past it without understanding.

I’ll also say, with a lot of people I’ve spoken to about AI, a lot of it is more of a front. When I worked with you on that discovery session, with all those icons, that’s real. That takes real R&D.

You had alarms at 2am because there was a beta release and you wanted to be part of it. That’s real on-the-ground stuff, and that’s what people need.

Janey Treleaven: When I set up Intelligence Assist, the first thing I said to myself was, I’m not building software and I’m not selling software. I’m selling AI adoption.

We all have subscription fatigue from streaming services. Software as a service came with a lot of subscriptions, it’s great, it’s helped, but we don’t want more and more software.

A lot of businesses trying to do what I do are actually a front for software, they’re trying to sell you their AI software. You don’t need new software. You need to use the AI in the software you’ve already got.

The Triangle of AI Adoption

Jeremy Yang: I think it’s very unique. People might see a wall of icons and get excited, like it’s a skillset game. But business owners with real budgets want to know which ones you use.

Janey Treleaven: Every business, regardless of size, is pretty similar. They all have front, customer-facing functions, how they find and engage customers.

They have a core product or service they deliver, and a back office where they do all the business-sustaining functions. In a small business it might be two people doing everything. In a large business it’s layers and layers of the same thing.

Before I started Intelligence Assist, I worked at Commonwealth Bank, and they’re not that dissimilar. Every business needs to make that Microsoft versus Google decision, get a secure core AI tool, and then train people.

I call this the triangle of successful AI adoption. You need the balance of value, the tools, what is the AI actually going to do. You need controls in place so people know what they’re allowed to do. And you need literacy, teaching people.

AI’s a great leveller. I’ve seen very old, elderly people really embrace it, and young people too, everybody’s on the journey, just at different points.

The hardest thing for most organisations is that variety of people, those who are really against it and fearful, those who are just waiting to be told, those who’ve experimented and didn’t think it was that great, and those who are really proficient.

Jeremy Yang: That’s the thing for me, because I’ve never worked in a big organisation. The teams you started in India, the amount of people-skill involved, that’s deep, because there are people who think it’s going to replace their job.

Janey Treleaven: Absolutely. I’ve done enough of it now that I’ve developed a few techniques. AI is not a technology-led project, it needs to come from process and people. That’s how you have to tackle it.

The technology is secondary to what the AI is actually going to do. It’s a new team member, you’re going to give it work, train it, give it feedback, and keep investing in it.

The second thing is around fear. People have realised AI’s good, but it’s not that good, so they’ve stopped worrying about it replacing them. What I find people more worried about now is that it’s going to take away the things they like doing.

Jeremy Yang: Can you give an example?

Janey Treleaven: Some people enjoy the writing process, writing letters. Some people like doing analytics the old way, or design, or coding.

The Right Way to Get Started

I tell those people, start with the thing you know how to do, because if you don’t know how to do something, you can’t verify the results. You have to keep the human at the core of everything it does.

If you’re not a lawyer and you try to get AI to be your lawyer, you can’t verify it because you don’t know what you’re looking at. So start with the thing you do know how to do, and also the thing you don’t enjoy doing.

Then it’s often just about getting over that first hurdle. Something I really hate hearing is people using the threat of, if you don’t get on board with AI you’re going to get replaced.

That’s actually what’s going to cause you to get replaced, not the AI itself. I actually think it’s the people who use AI badly that are going to get replaced first, if you don’t verify its results and make sure it hasn’t hallucinated.

Jeremy Yang: You’ve used it carefully, your hand-ins don’t require double-checking, I’m not picking little mistakes out.

Janey Treleaven: Part of what holds people back from doing that verification is what I call AI slop, low quality, over-reported content. It’s exhausting reading all of that.

The problem isn’t the AI, it’s you. If you say “write this for me” and don’t say “I just want five bullet points,” it’s going to go over the top and do too much.

Jeremy Yang: One of the final things on my list, you said you don’t like the phrase “human in the loop,” you’d rather say “human in the core.” What other clichés have you moved past?

Janey Treleaven: Maximise your investment in the tech stack you already have, because AI is more about the people and process around the tech than the actual tech.

Lay the AI foundation first. Don’t try to build a system that’s going to do everything for you, by the time you finish, the technology’s moved on. Get the foundations right first, get the core AI tool in, get everybody using it.

Get that productivity boost, but be patient with the boost. Right now, most people are using AI to do fifteen percent, they’ll start writing an email, think maybe AI could write this better, and go backwards and forwards.

What we need to get to is AI literacy that gets you to the next step, where you only have to do the five percent, because you go straight to the AI tool, and it already knows how to do that task for you.

Once you get to that eighty-five-fifteen split, then you can get to the “do it for me” stage, where the AI is monitoring for an email arriving and writes the draft for you, or sees a new order come into your CRM and kicks off a whole bunch of other tasks.

Jeremy Yang: You’re probably the most experienced person I’ve spoken to about this, and someone I actually trust. I see stuff on LinkedIn where people say they’ve got fifteen agents working for them, but I don’t know if it’s true, or if they’re just selling a course.

Janey Treleaven: We do at Intelligence Assist practice what we preach. We’re punching well above our weight with the clients we’re working with, partly because we use AI every day.

Every one of my emails is read by an AI agent and it drafts a response for me. I don’t have an assistant. We always challenge ourselves, could we have used AI to do that better.

Jeremy Yang: I love this chat so much. We should catch up in six months. For businesses out there, what’s the ideal client, or your passion work right now?

Janey Treleaven: It’s coming in and looking under the hood, that’s my favourite part, being let in to do a blueprint discovery workshop. We always kick off with a three to four hour workshop, I call it peeling away the onion.

It’s a two-way workshop where I get to teach a few things people didn’t know before, get everybody in the room a little more literate, and excited. Anyone who says “nah, I can’t do it,” I see that as a personal challenge.

I love finding all the opportunities and presenting them back a couple of weeks later. But you can’t do everything, it’s a journey. You’ve got to start somewhere, there’s no right or wrong.

Jeremy Yang: On a small business, I guess the business owner is the one contact. On a medium-sized business, is it the COO, the CTO, an operations person?

Janey Treleaven: The operations people, people in culture, it’s amazing the different people who approach us.

Jeremy Yang: Can you tell people again how they find you and the best way to get hold of you?

Janey Treleaven: Best way is to jump onto our website, intelligenceassist.com.au, and book a discovery call with me. I always take those calls, you’ll probably learn something in half an hour.

Thank you, Intelligence Assist, think IA for AI, and just through our website. You can also find me on LinkedIn.

Jeremy Yang: Cool, thanks again for coming on the show.

Janey Treleaven: Thank you very much.

• Start with your existing tools, not new software. Most small businesses don’t need to buy new AI products, they need to properly use the AI already built into Microsoft or Google Workspace. Pick your core AI tool based on capability, security, and knowledge access.
• Free tools have a hidden cost. If you’re not paying for a product, you’re the product. Note-taker bots and free AI tools often collect video, voice, and data you wouldn’t otherwise hand over, so free isn’t actually free.
• Ground and RAG your AI to stop it hallucinating. AI trained on the open internet inherits bad information along with good. Control it by grounding it to verified knowledge and only allowing retrieval from trusted sources.
• Adoption is a people and process problem before it’s a technology problem. The biggest barrier isn’t the AI itself, it’s fear, unclear instructions, and lack of training. Start with what you already know how to do so you can verify the AI’s output, and treat the rollout as a journey, not a single event.

---

This conversation is from Marketing for SMEs with Jeremy Yang, a show for Australian small business owners on Google Ads, Meta Ads, and everything in between. Available on your preferred podcast app and YouTube.