Case studies / 04
120 household case studies coded to one standard, with a quote behind every claim
For a UNICEF Zambia child poverty study, we turned 120 interview-based case studies of female-headed households into one consistent evidence base that report writers could query and check.
- Client
- Heidi Loening-Voysey, lead consultant
- For
- UNICEF Zambia, child poverty study
- Period
- August to December 2025
- Our role
- Data and evidence subcontractor to the lead consultant
- Service
- Data System Build, delivered as a subcontractor
- Status
- Delivered and handed over. Report accepted by UNICEF
The problem
The study produced 120 long narrative case studies from interviews with female-headed households in Mongu and Kasama. Coding one by hand, at the depth the report needed, took 75 to 90 minutes.
Speed was only part of it. Different analysts could code the same story differently, and writers had no quick way to get from a finding back to the household and the words behind it.
What we did
Before any coding, we set a ten-theme structure, field rules and a data dictionary that every case had to follow. Missing information was recorded as missing, not guessed, and unclear passages were flagged for review.
AI did a first pass on each case. Every claim it extracted had to carry its supporting quote and its reasoning, and each case was checked before it went into the database. The evidence base was built in Excel and Google Sheets, the tools the team already used, with formulas for comparisons between the two sites, ranked lists and report tables. Writers could ask questions of it in plain English.
We handed it over with a written procedure and two training sessions, so the team could keep running it.
What we built
- A ten-theme schema and data dictionary
- An AI-assisted coding process with a quote and a reason for every claim, and review before entry
- An evidence base in Excel and Google Sheets covering all 120 cases
- Report tables, site comparisons and summaries
- A plain-English query tool for report writers
- QA and reconciliation checks
- A standard operating procedure and two handover sessions
The work in figures
- 120 case studies
- 10 themes
- 2 sites, Mongu and Kasama
- About 15 minutes per case, including checking, against 75 to 90 minutes by hand
- 120–150 analyst hours saved across the study
What changed
Every case followed the same structure, so findings could be compared across themes and sites. A line in the report could be followed back to the case, the quote and the cell it came from. The team kept the process after we left.
Limits
The coding structure was set for this study's ten themes. Another study needs its own schema before the same process can run.