Oli and the Gotch

Oliver Young: Hello. Hopefully, everyone can hear me okay. Hi Analine, I hope I said your name correctly, apologies if not. I'll tell you what, while we're all waiting—everything's better with a dachshund. There you go, buddy, why don't you sit there? Okay, so we've got a few people coming in today. Hello, how are you? You're on mute, so I can't hear you. Just to make sure that there's not a thousand voices in my ear! But this is Louie Sparkles. As much as I think he's my dog, he's actually my daughter's dog. I'm Oliver, obviously. We're just waiting as we've got quite a few people coming today. We've got Wendy—Wendy and I used to work together. And we've got Deb. Deb is our Chief Technology Officer; that's probably a grand term for one of our coders. He's here to make sure that when I'm talking all things technology, I'm not talking crap. Everyone's on mute except for me and Louie. Can you hear me okay, Wendy? Perfect. Oh, by the way, if you want to chat with me at any point, the chat is in the bottom right corner. Michael is going to help me run things today; he's co-hosting. Michael, you can speak, can't you? Hopefully, people can hear you. Michael Gotch: Yeah, I can. I've been given the powers. Morning, everyone. Oliver Young: Alright, we've got more people coming in. Dan, Jennifer, hello. For everyone who's just come in, this is our mascot, Louie. He's a one-year-old dachshund. I'm going to put you down now, buddy. Do you want to go back to your bed? Okay. While we're still waiting for a few people, it's a few minutes past the hour, so we might get started. Michael Gotch: Yeah, let's do it. Let's jump in. Oliver Young: I know that there are some of you out there who know me because I've worked with you, done a demo for you, or met you at a conference. My name is Oliver Young. I'm a community engagement person. I quite often refer to myself as a "recovering" community engagement person because I'm not on the tools anymore; I'm into software. My co-host is Michael Gotch, who is also a community engagement person. But where I'm more of a facilitator, dealing with angry people, a mediator, and a town-hall style guy, Michael is a community engagement writing expert. The two of us thought that we would get together and start a podcast. Whether this is a good idea or not, you can probably all tell me at the end! I'm the founder of something called Scopo Map, which is an AI community engagement piece of software. Through doing demos, onboarding sessions, talking at conferences, and speaking to peers (I've been doing this for 25 years), it's become pretty clear to me that AI in community engagement is a hot topic. There's a lot of chat around the ethics of AI and security-related issues, but we haven't really been talking about what it means in a nuts-and-bolts, practical way. When I go in and do an onboarding session with a new client, I'll ask the team, "How do you guys use AI?" It's a real mix. Some say, "I use it to draft newsletters and proposals." Others say, "I'm really good at it and I can make it do coding for me." And then there are people who say, "I didn't know it could do PowerPoint slides." It's become clear that a lot of practitioners don't have a systematic, pragmatic framework for how they're going to use it in their everyday practice. I'm going to share some slides with you—and I can tell you that AI did not do these slides. I'd like to say it was all me, but it's actually Michael. He's much better with PowerPoint and writing than I am. Today, I want to unpack AI in practice. I want to talk about tasks that AI can help you with and which stages of engagement they're good for. I want to talk about purpose-built AI tools—specifically the ones I'm familiar with in the Australian practice. I want to show you how AI can help you do work faster and better, and give you a framework for thinking about AI in engagement so you're better informed about what tools are right for your specific projects. Here's one of the things I hear quite often. I'll show what our product can do in terms of drafting, risk assessment, and identifying opportunities, and people will say, "Oh, well, there goes my job." But then they follow up with, "No, actually, I'm fine because AI can't do relationships and it can't do a door knock." And that's true. It can't read the room. It can't build trust. It's not emotionally intelligent, and it's not good at navigating conflict. However, thinking AI won't replace us just because it can't knock on a door isn't a helpful way of looking at it. We talk about engagement as if it's all one thing, but it's actually a collection of different tasks. Only about 30% of what we do is face-to-face engagement. The other 70%—sometimes more—is analytical tasks and drafting. What if we could do that 70% better and faster so that we could spend more time on the 30%? Back when I was consulting, we'd spend the vast majority of our budgets drafting comms plans, newsletters, storyboards, and outcomes reports. Clients actually wanted us out doing face-to-face outreach, but we couldn't afford it because we were spending all this time drafting. If we can use AI to do the drafting quicker and better, that changes the game entirely. But people are overwhelmed because there are so many tools in the marketplace, and they don't know what they do. AI isn't one single thing. Let's distinguish between Generalist AI and Purpose-Built tools. Generalist tools include Gemini, Perplexity, Claude, ChatGPT, and Copilot. You've almost certainly used one of those. They all do some things well, but they have big shortcomings when it comes to community engagement. Then there are the purpose-built tools. I think a useful way to think about these tools is to plot them onto a typical project lifecycle. By doing this, you realize they are doing very different things, even if they're all intended for "community engagement." Let me run through some of these tools: Scopo Map (my tool): Started as a stakeholder mapping tool. We integrated it with an AI named "Bruce." Think of Bruce like a grad. He helps you draft proposals, do stakeholder analysis, and draft comms plans, deeply rooted in ABS (Australian Bureau of Statistics) and geolocated data. It sits primarily in the "Plan and Prepare" phase. CE Canvas: A relatively new project management tool. It helps teams build engagement plans using templates, maps stakeholder influence, tracks project timelines, and assigns team responsibilities. Open Point / Granicus (Engagement HQ) / Go Vocal: These live in the "Engage" phase. They host online consultation websites, surveys, interactive maps where communities can drop pins, and they manage submissions. (Granicus is what "Bang the Table" became). Hello Lamp Post: A British application now in Australia. It's a text-based AI chat tool. They put QR codes on posters or pavements, and citizens can text an AI assistant to give feedback or ask questions. Local Loop: Sends consultation opportunities to local residents via targeted digital channels, using things like online storyboards. Otter AI / Firefly AI: Recording and transcribing tools for workshops and meetings. They automatically generate summaries, themes, and action items. Community Labs / Thematic / Covert Lens / Simply Stakeholders: These sit in the "Analyze and Evaluate" space. Anyone who remembers doing engagement 20 years ago remembers coding a thousand paper feedback forms into Excel. These tools analyze large volumes of feedback to detect patterns, categorize open text, and organize submissions. (Simply Stakeholders, for example, is the modern evolution of Darzin). Michael Gotch: Hey Ollie, a quick question from the chat. Will we be providing the slides and the recording afterwards? Oliver Young: 100% we will. Michael Gotch: And a question from Lucy: Do these software tools build knowledge based on your own projects, or is each project new? Oliver Young: My understanding is, does the information share across a range of projects? With a tool like Simply Stakeholders, that's probably true. For our project (Scopo Map), each project is new, but our AI, Bruce, has a learning model that makes him learn from the specific data and prompts we feed him, not from other people's data. Michael Gotch: I think the key point we're trying to convey here is just pulling apart this amorphous pool of "AI tools for engagement." It helps a lot just to place them somewhere in the project lifecycle. Even if they spill over into different phases, having that framework helps you sort out what you actually need. Oliver Young: 100%. Thank you, Michael. Let's look at why you wouldn't just use Generalist AI tools for this. They're great for some things—ChatGPT is good for drafting, Claude is good for deep numerical analysis, Perplexity is great for searches and sourcing, and Gemini is highly multimodal. But they come with inherent risks. For example, during the Northern Rivers Recovery, a contractor used ChatGPT to crunch a whole lot of feedback forms that contained people's contact details, and that data got out into the public. People are rightfully concerned about uploading IP and security to generalist AIs. Also, generalist AI hallucinate. Purpose-built tools have safeguards and "harnesses" built around them to significantly reduce hallucinations. A generalist AI has access to massive amounts of data, which means if you ask it the same question three times, it will give you three slightly different responses. They also have relatively short-term memories; if you have a long chat string, it starts to lose the context of where you started. Let me give you an example comparing ChatGPT to a purpose-built tool (Scopo Map/Bruce). I asked ChatGPT: "What is unique about the area 100 meters either side of Victoria Road compared to the suburb of Rozelle?" ChatGPT gave me a comparison using "typical SA1 patterns for inner west arterial corridors." It wasn't actually looking at Victoria Road; it was looking at roads in the Inner West like Victoria Road, including places like Parramatta Road, and mashing the data together. ChatGPT said the median weekly income was $1,062. Bruce looked at the actual SA1 tables and found it was $3,028. ChatGPT said the dominant professions were hospitality and retail. Bruce correctly identified professionals, managers, and trades. ChatGPT said the area was 60-75% apartments. Bruce identified it was actually 28%. ChatGPT said 40-50% of people were born overseas with massive linguistic diversity. Bruce correctly noted that 65% were born in Australia, overwhelmingly English-speaking, with a small percentage speaking Mandarin, Greek, and Italian. If I relied on ChatGPT, I'd be planning to hire a bunch of translators for a highly diverse apartment-dwelling population, when in reality, I'd be engaging mostly middle-aged, professional, English-speaking people living in terraces and townhouses. Let's do a live demo of what our AI, Bruce, can do in practice. Michael Gotch: Quickly, Alison asks: "Do you sell this to property developers and real estate agents? They would love this sort of data." Oliver Young: Yes, I am working with some people in the property and real estate industries. Michael Gotch: Another from Ashley: "Does Bruce only deal with Australian data, or could it be New Zealand?" Oliver Young: Currently, it's Australian ABS 2021 data. However, places like Canada, New Zealand, and the UK have very similar federal census formats and geolocated post office data, so we are absolutely looking at those markets. (The US is harder because they collect census data state-by-state with different questions). (Oliver begins screen share of Scopo Map) Oliver Young: Okay, so I'm going to talk to Bruce. "Hey Bruce, can you map a 100-meter alignment either side of Victoria Road from Terry Street to Robert Street in Rozelle, please?" (Bruce generates the map instantly) There's your map, automatically done. Anyone who has ever struggled with a letterbox distribution map, photocopying a street directory and drawing a line with a red texta—no more. All you need to do is talk to Bruce. All the data on the screen comes from Australia Post and demographic benchmarks against the state average. When we first built this, there was no AI. People thought the maps were cool, but let's be honest, a lot of comms people aren't great at crunching quantitative numbers. When we created Bruce, he made sense of all this data. I can ask him: "Hey Bruce, what's unique about this area?" He goes away and gives us five key points. Notice his language—he has a specific Australian vernacular. He has a backstory: he's a middle-aged urban planner who lives in Melbourne with a dachshund. Why did we give him a persona? Because when you explain a role to an AI, you get a much more focused response. We also trained him on engagement handbooks, IAP2 standards, and local government handbooks so he understands what community engagement actually looks like. Next, I ask: "Hey Bruce, I'm going to do a consultation around a high-density development in this area. Draft me a media analysis about any opposition to high-density housing in Rozelle." Bruce uses Perplexity to search the web for real, current context. He brings back media sentiment regarding the old Balmain Leagues Club site and the Rozelle Interchange. He gives me the constraints, counter-narratives, and links to all the source articles. That's your draft media analysis done. Next, I ask: "Hey Bruce, create sample personas for this area based on the data." He creates synthetic people based on the actual demographics, outlining their interests and engagement styles. Then I say: "Hey Bruce, roleplay being each persona and answer: What are the three main things you're concerned about regarding a new high-density development?" Basically, you are running a virtual focus group. We've tested this process about 14 times on completed real-world projects. When we matched Bruce's generated themes against what the public actually said during those consultations, the accuracy was between 83% and 97%. Finally, I ask him to draft a consultation strategy, including milestones, techniques to reach hard-to-reach voices, and resourcing. He pumps it out in seconds. Is it 100% ready to go to the client? No. Is it 70% of the way there? Absolutely. Mick, you're the writing expert. Tell me about taking that 70% draft and getting it ready for the client. Michael Gotch: What I see all the time in engagement is static data. People will include demographic breakdowns in a report but do absolutely nothing with it; they treat it as if the data speaks for itself. It doesn't. Why this tool is useful for me is that it turns numbers into meaning. When I'm writing early drafts, the exact wording doesn't matter as much as having robust building blocks of information that I have confidence in. It forces you to take a critical eye, extract what's meaningful, and put it into your own words. It's a huge deal for sense-making. Oliver Young: Exactly. Traditionally, we'd just look at "quick stats" for an LGA, but we know suburbs change from street to street. Getting this granular data is half the battle. To wrap up: Doing things faster isn't enough. If everyone just does stuff faster, clients are going to say, "Why am I paying you $6,000 for this outcomes report when I know you pumped it out of an AI?" Profits will shrink. It's not just about doing things faster; it's about doing them better. What do we do with this extra time and insight? We can start benchmarking: Who actually responded to our project? Was it representative of the community, or was it just 80% angry middle-aged people? We can identify who we didn't reach and design targeted engagement to reach those missing voices. When finding the right AI tool, ask yourself: What problem are you trying to solve? Is it planning, engaging, or analyzing? Are you looking for time-saving, or superior outcomes? What's the size of your project and budget? And who are your stakeholders? (For example, don't try to get an 80-year-old to interact with an AI QR code on a lamppost). The good news is, clients still need us. AI still hallucinates. It sounds smart, but sometimes it's hollow. There's real, tangible value in human practice. Michael Gotch: One final question from Ashley: "Would Bruce operate as an agent for an organization, or is it a generalist tool specific to engagement? What are the security limitations for using Bruce versus a custom-created agent?" Oliver Young: Think about Bruce like a grad. You can upload documents to him—storyboards, RFPs, comms plans, templates—to inform his drafting. From a security point of view, all those documents are held strictly on our servers in Sydney. They are not shared with other accounts. They are deleted within 48 hours unless you choose to save them to your private document library. He only uses ABS data, Australia Post data, and the data you give him. He only goes out to the web when you explicitly tell him to. The security risks are very low. Thank you all for coming. We're going to be running these masterclasses for the rest of the year, bringing on the developers of the other applications we talked about today so they can showcase their own tools, and talking to practitioners using AI in real case studies. I'll send out the slides and the recording to everyone via email. Thanks again, and I'll see you soon! Michael Gotch: Thanks everyone. Cheers.