Finding Demand for a New Event Format
The useful output wasn't a topic cluster. It was an event the business could run next week.
- Client
- The same cafe, before it became a lounge. Independent work.
- My role
- I collected and analyzed public local posts and recommended the event format.
- Result
- Recorded foot traffic doubled in the quarter after the launch.
- How it was measured
- The quarter after the launch against the quarter before. Not a controlled test.
Result: foot traffic doubled within one quarter of launching the language-exchange nights, and the format stuck.
The problem
A cafe needed a reason for people to come back on a schedule. The question was which event had real local demand but few places offering it, and whether the business could try it without committing to a big program first.
What I did
I collected public Reddit and Meetup posts about social events, language learning, and recurring meetups, along with messages from public local community group chats. Then I analyzed the text in aggregate, counting which events people mentioned, which phrases came up together, and how people reacted, to compare three concepts: language exchange, karaoke, and trivia. I also trained text models on the combined collection.
Language exchange stood out as underserved: steady interest and few events offering it. So I recommended a small test: run language-exchange nights, watch attendance, and compare the next quarter with the one before.
How it was measured
Recorded foot traffic in the quarter after the launch, compared with the quarter before. It's a before-and-after comparison, not a controlled test, so season and other changes at the business aren't separated out.
How the business used it
The finding changed what the business sold. Marketing moved from "come visit the cafe" to a scheduled social reason to show up. The copy got casual, energetic, and multilingual, along the lines of "Welcome to the worst language class you'll ever take and the best party you'll find this week."
The events also gave the business a repeatable unit for testing. I used designated holdout and treatment days to compare event formats, audiences, and marketing approaches.
The business grew, and an asset sale brought in a second owner with more capital. The new owner turned it into an event-based lounge built around the same idea and kept me on retainer for attribution, pricing, and performance tracking. My later analysis for the lounge found karaoke and trivia under-offered with strong interest, and it now runs language exchange and trivia nights.
Built with (for technical readers)
Python with NLTK, spaCy, and Hugging Face models for phrase counts, topics, and sentiment.