MONEYME is a digital lender and one of the leading non-bank challengers in the Australian market. Its UX, marketing and product teams launch new credit products, and change promotions and messages, constantly.
One of those teams recently spent thirty minutes in Claude, connected to Semilattice over MCP, on a positioning change that had been discussed inside the business on and off for months. Nobody had the data to decide it, so it had never progressed. By the end of the session they had explored the change and validated it with their simulated users.
This post explains why the question had stayed open, how we built and validated MONEYME's user model, and how its teams use it. The full story is on the MONEYME customer page.
Decisions arrive faster than research can answer them
MONEYME's teams make judgement calls every day: which promotion to run, how to word a product page, which feature to build next, what to lead with in a campaign. They also use AI tools to work faster, which leaves less time for validation and data gathering.
Research that relies only on surveying or interviewing real people couldn't give the UX and marketing teams enough data for decisions that frequent. The positioning change was one of them: it needed data that nobody had.
A model of 450 everyday Australians
MONEYME's target audience is everyday Australians making financial decisions, so we built a model of that audience.
We worked with MONEYME's team to understand the audience and find the best available data about it. We then used non-personal consumer-finance discussion data from about 85,000 Australians to fine-tune 450 LLMs. Each one models a different customer profile within MONEYME's target audience, rather than one model standing in for the whole audience.
Asking all 450 of them 5 questions takes about 15 minutes.
Check the accuracy before you rely on it
Before we handed the model over, we held back a set of real answers from the data it was built on, asked the model to predict them, and compared the predictions to what those people actually said. We shared the results with MONEYME's team in full, including where the model is weaker, so they could decide how much weight to put on it for each kind of question.
Over 200 predictions in the first few days
Within days of getting their model, teams at MONEYME had run over 200 predictions. A prediction is one question answered by every simulated user in the model.
The teams chose Semilattice partly because simulated users can be put through products and user flows, not only asked questions, and partly because they could run simulations directly in Claude over MCP.
run_simulation
·
A real Semilattice session, run inside Claude via MCP.
The positioning session worked the same way, with the team asking questions in Claude and Semilattice answering them from the model.
"Using Semilattice has been game changing when developing product features. We've been able to source early, frequent, and instant customer feedback that's almost identical to real life customer survey responses. It's been instrumental in helping us make the right product decisions at every stage of ideation and discovery."
Jem Corden, Head of Product, MONEYME
Get a model of your users
Read the full MONEYME story, or get started and we'll build you a custom model of your users.