Challenge
As a digital challenger in a competitive market, MONEYME is regularly launching new credit products and constantly iterating promotions and marketing messages. At the same time, teams are increasingly using AI-based tools to speed up processes, leaving less time for thinking and data gathering. Existing market-research tools that rely exclusively on surveying or interviewing humans did not give MONEYME’s UX and marketing teams sufficient data to support the frequent decisions they were making every day.
Solution
Simulating MONEYME’s target audience of younger Australians felt like an obvious solution to the teams at MONEYME. They were especially drawn to the ability to have simulated users actually explore products and user flows, and to run simulations directly in Claude using MCP.
At the same time it was clear that, to be of value, MONEYME’s user model had to be highly accurate. To ensure that, Semilattice worked closely with MONEYME to understand their audience and identify the best sources of data about them. Once the model was built, Semilattice used held-back training data for a rigorous validation and shared the results, so the MONEYME team could understand the strengths and weaknesses of their model.
Results
Semilattice used the data sources identified by MONEYME’s team to build a custom user model of its target audience. Using non-personal data from about 85,000 Australians, it fine-tuned 450 LLMs, each representing a rich customer profile within MONEYME’s target audience.
Teams at MONEYME now have a model of their target users they can consult for any decision in minutes. Within days they’d run over 200 predictions. In one instance, a 30-minute MCP session helped the team explore and validate a strategic positioning change that had been discussed on and off but never progressed, for lack of data.
run_simulation
·
A real Semilattice session, run inside Claude via MCP.
