Thousands of survey responses, themed, scrubbed, and analyzed, in a single project
Launch by Lunch built ELIXR a research project in Claude that turns raw open-ended survey data into themed, privacy-safe, analysis-ready findings, with a human checkpoint at every step. Work that used to be slow and manual now runs end to end, and it is reusable for every new client.
6 hrsmanual coding time, before & after
- Industry
- Workforce & caregiving research
- Team
- 2 co-founders + small research & ops team
- Engagement
- Research project build
- Timeline
- May → June 2026
The challenge
Incredible research insights, done manually, that couldn't scale
ELIXR helps organizations understand the caregivers in their workforce, work that hinges on reading thousands of open-ended survey responses and turning them into honest, usable insight. Doing that well is painstaking: every response has to be read and themed consistently, scrubbed of anything that could identify a person before it ever reaches a client, then woven together with the quantitative data.
It was repetitive, time-consuming, and high-stakes, one missed identifying detail in a client-facing file is a real privacy risk. It also hit a ceiling. Growing the client base or the volume of responses meant adding people and time, not just pressing go. ELIXR needed a way to do this faster and more safely, without giving up the human judgment the work depends on.
What we built
Three skills that move raw data to client-ready, with checkpoints
A reusable research project in Claude that's easy to replicate client over client.
Themer: consistent coding at scale
Reads every open-ended response and codes it against ELIXR's own theme taxonomy, the same lens applied to thousands of open-ended responses in a cycle, with a human checkpoint to confirm the themes before anything moves forward. Responses submitted in Spanish are translated inline into English and coded with the same lens, no separate workflow.
PII Scrubber: safe by default
Automatically finds and removes identifying details to produce a clean, client-facing copy. It over-redacts on purpose and logs every change, so only the rare genuinely ambiguous case lands in a short human review queue.
Analysis: qual + quant, together
Merges the coded themes with the survey crosstabs and the team's intake knowledge into staged findings, with built-in guardrails (small-group suppression, significance gates) so the deliverable is never built on partial or unsafe data.
Reusable for every new client
The skills resolve each client's data by a simple naming convention, so a new study is a fresh workspace and a couple of swaps, not a rebuild. The team owns and maintains it going forward.
The results
Faster, safer, and built to scale across clients
- Built to scale across clients: a new study is a setup, not a from-scratch build, so the project replicates client over client.
- Gets smarter every cycle: as ELIXR grows its theme taxonomy and learns across client projects, the system compounds and improves over time.
- Privacy protection runs by default: identifying details are removed automatically, with every change logged and a tiny review queue (0.3%) for edge cases.
- Consistent, defensible coding: every response is themed against the same taxonomy, no drift between cycles or reviewers.
- Accurate and consistent in two languages: all 2,300+ responses, including those submitted in Spanish (translated inline into English), are coded with the same lens.
- Human judgment kept where it matters: checkpoints at every stage keep the team in control of what ships.