Late last year, UNE’s SMART Region Incubator delivered a Graduate Certificate in AI Innovation, developed and taught by Dr Matthew Wysel. Participants were asked to bring a problem to the course that they would then attempt to address via AI.

The following interview conveys how Rob Hughes, the University’s Digital Marketing Coordinator, identified a UNE problem that found an AI solution as the course unfolded. Rob and his colleague, Marketing Content Coordinator Sharleen Schier, found that technology implementation is relatively easy. It’s accurately identifying the problem, and bringing people along with the solution, that takes time.

ROB HUGHES, DIGITAL MARKETING COORDINATOR:

I wanted to do the Graduate Certificate in AI Innovation to expand my knowledge of AI and get an official qualification.

To get a UNE funded spot, I had to choose the organisational problem I wanted to address.

I came up with three or four ambitious, highly technical applications that I wanted to build. I took them to the my colleagues and asked: which one?

The team said, ‘none of them’, not because they weren’t cool ideas, but because they weren’t what was important.

It became clear that the best thing I could do was to fix Sharleen’s workload, because what Sharleen does is crucial and her workload is insane.

SHARLEEN SCHIER, MARKETING CONTENT COORDINATOR:

I’m responsible for the Course and Unit pages in UNE’s online Prospective Student Catalogue (PSC). I’m also responsible for the Discipline and Area of Interest pages.

To explain how these pages interact: when a student visits www.une.edu.au/study, they are taken to the PSC. If they know the course they want to study, they can search directly for it from the PSC home page. Or, if they aren’t sure about what courses might suit them, they can select an area of interest, like Animals and Agriculture, and then drill down to disciplines, and then to specific courses within those disciplines.

Each of these levels has some description to help students understand what’s on offer and to guide decisions.

There are more than 200 courses displayed on the PSC, and almost 2000 units. That adds up to a lot of assets to maintain and keep updated, because they are dynamic, and change over time.

My job is to build the marketing content for the PSC, and all that sits underneath, and to make sure that courses are described to students in a way that is exciting and appealing, and gets them through to making an application.

That work is handled by just me, working across all the Faculties and Schools, and the sheer quantity of material can be overwhelming.

At times I’ve had to carry solo the workload of nearly two full-time positions. Then Rob stepped in and made my life easier.

RH:

Working with Sharleen, I started working on a bot intended to fulfil one task: to draw on information from the UNE academic governance system, CourseLoop, and create draft marketing content in Sharleen’s voice.

Sitting behind every course is material like course proposals and course descriptions, which provide an administrative overview of what a course is about and why it’s relevant to our students.

A time-consuming part of Sharleen’s job has been to convert that material into compelling marketing content designed to sit within the PSC in a way that appeals to prospective students and encourages them to click “apply”.

My project was to make an AI bot that can interrogate all the background material and turn it into marketing content in the right length and format, and importantly, in a style that imitates Sharleen’s writing style.

It’s work very suited to AI. Generative AI chatbots are only as good as their training material. If you train them on the right stuff, they can generally turn out something that is almost as good as the training material.

Because Sharleen has written so much, I had an absolute wealth of material on which to train a bot. I ended up training the bot on 12 of Sharleen’s favourite pages of content.

SS:

There’s still a bit of work to do on the bot’s output – some massaging and fact-checking – but it is alarmingly accurate. It’s a bit of uncanny valley seeing your own tone of voice parroted back at you.

The way I write is informed by evidence-driven research. It’s not just, ‘this is how Sharleen likes to write content’. It’s based specifically on what appeals to prospective UNE students, who we’ve researched extensively to understand how they want to be spoken to. And the AI is excellent at mimicking the way I write to appeal to them.

The AI supports my own writing process too. I might be stuck halfway through the sentence and I can ask the bot to give me three different endings so that I can determine what I think is the most effective.

It’s almost like having a colleague to bounce off. In Marketing, we’re all specialists in our respective areas, and everyone is more than busy just doing their job. There’s not a lot of time to soak up someone else’s time to say, hey, what do you think of this sentence? The bot is surprisingly good at being that colleague.

RH:

The other aspect of the AI bot is its support for analysing the unique selling points (USPs) of our courses.

We can input URLs of similar courses listed by other universities, and the AI will scrape the relevant data, run an analysis to identify variations between our courses, and come up with table we can use to identify our USP.

It helps us understand what other universities regard as the key properties and outcomes of their courses, so that we can then devise a strategy to differentiate UNE’s course in the market.

SS:

I’m really limited in my ability to visit every single university’s website to see whether we can pitch our courses differently.

The AI brings all the data together for us. I don’t have to go onto individual websites and write down that the duration of UNSW’s course is 0.5 years, the duration of University of Newcastle’s course is 0.5 years…

It’s a huge time saver, and necessary because one of the most important parts of writing good marketing content is understanding your USPs. To properly sell the course, we need to know whether we’re offering a unique product, or whether the way that we are teaching it or the experts that we’re working with are special in some way.

That means understanding what the rest of the market is doing, and gathering that intelligence manually is very time-consuming.

RH:

We calculated that overall, the AI saves Sharleen about 18 hours per new course she needs to develop content for. That’s time freed up from the grunt work of the initial translation that she can spend on improving quality.

It took me about a week of working with the team and Sharleen to narrow down what we wanted and to understand the problem and what sort of solution we needed to design.

The actual building of the bot took me 16 hours. I haven’t touched it since.

SS:

The time freed up by AI gets fed back into fine-tuning our content, and picking up work that’s been left on the backburner due to resourcing.

Things like updating student testimonials and gathering new course images; those things are my responsibility as well. And there simply hasn’t been the capacity for doing that work.

Last year UNE created more than 20 new courses. That’s about a 100% increase on previous years, and researching and writing the content for all those courses didn’t leave much time for anything else.

This year there have already been five or six new courses with approved business cases, so I expect that by the end of the year we’ll probably hit the 15 to 20 mark.

AI helps us to claw back some time so that I’m not just producing content; I’ve got capability to finesse it and review some other aspects of our online experience for incoming students.

RH:

I think what made this particular project work was collaboration with the team, and specifically with Sharleen, so that I could see a specific issue from that person’s viewpoint and the figure out how I could bring the technology to the table.

If your technology doesn’t solve a real person’s pain point, then it’s just another piece of technology.

Working with the people who are going to be affected by the technology also makes the onboarding process a lot smoother.

SS:

The way Rob approached things with me was fantastic.

He was saying, I know that you’re hurting here. If I’m building something, what would you want it to focus on so that it would alleviate the most stress for you?

The fact that he came to me from that perspective made that really positive experience for me. It was oooh: this person doesn’t want to take my job. He wants to help me with my workload issue.

I understood exactly how it was going to help me specifically in my day-to-day job. It wasn’t just this vague thing that I have to add to my list of things to do.

I bought into it immediately. I was happy to do the work, and give Rob the information he needed, because I could see from the outset how this was going to be a huge benefit to me in the long term.

RH:

We have something of a resourcing crisis at UNE. We can put AI to work clawing back human time lost to poor processes, so that we can get back to doing the things that only people can do.

There is so much work just in closing the resourcing gap that I think it’s going to be a long time, if ever, before we see AI actually taking away valuable work from people.

I think it’s also important to recognise that AI can be very bespoke. We can make bots to automate the smallest processes with very little overhead. And once we’ve really thought about a problem, and made a bot to resolve it, that bot can be used on similar problems elsewhere in the university, or adapted to address something similar but not the same.

SS:

I’m very curious about what AI can do, so I’ve started looking at how it can help me finesse descriptions new units.

The unit coordinator will write down a general overview of a unit, and it’s up to me to massage that into something that is designed to be really appealing to students. I keep all the information factually correct, but give it more of a spin, and adapt it to the same tone and style we use for courses.

I wanted to see whether the bot Rob built could do that work. It clicked onto it super fast. It hasn’t actually been trained on any unit descriptions, but it’s following that key UNE tone and style because it knows that’s how we expect the output to look. It’s really quite clever in how it’s used the information fed to it and applied it to another application.

It’s still not getting it all right out of the box. Like the course content it produces, I still have to fact-check it and make sure it hasn’t introduced some completely random information – which it occasionally does.

That’s one of the reasons that I don’t worry about my job. AI is never going to be perfect. There is still work to do, even if the bot has started it.

IMAGE: Generated by AI