Use cases
The research data problems we are built to solve
Six kinds of engagement, each with a defined way of working, a defined set of deliverables, and checks that make the result defensible.
Pick the one closest to your situation, or combine them. Most projects span two.
01 / Document extraction
From reports and filings to analysis-ready datasets
Turn thousands of pages of reports, filings, and publications into one structured, source-linked dataset.
The challenge
The evidence your question needs exists, but it lives inside annual reports, regulatory filings, and statistical releases: different layouts, different years, different definitions. Copying numbers by hand does not scale, and a table nobody can trace does not survive review.
How we make it defensible
Every accepted value carries its source location, and every check, pass or fail, is reported.
Typically for: Research consultancies, economics and finance groups, policy teams
What we build
- Source inventory and layout analysis across the full document collection
- Extraction pipelines for digital and scanned documents, with every value tied to its page and table
- Unit, period, and currency normalization with explicit, versioned rules
- Reconciliation checks, such as totals, accounting identities, and cross-year consistency
- An exception queue so ambiguous values reach a human reviewer instead of the dataset
What you receive
- Structured dataset with a data dictionary
- Source manifest with page-level references
- Extraction and validation code
- Exception register and validation report
02 / Data harmonization
Harmonizing surveys, waves, and administrative records
Combine datasets that were never designed to fit together, with every mapping and comparability decision on record.
The challenge
Survey waves rename variables, change response categories, and shift definitions. Partner submissions arrive in their own formats. Combining them into one series is less a join than a sequence of judgment calls, and those calls have to be visible to be defensible.
How we make it defensible
Comparability is documented per variable, so analysts know exactly what can and cannot be pooled.
Typically for: University research centres, development research teams, longitudinal studies
What we build
- Variable-by-variable mapping across sources, owned and approved by your researchers
- Crosswalks for codes, geographies, and identifiers, versioned as they change
- Join diagnostics: cardinality, unmatched records, and unexpected row expansion
- Break detection that flags discontinuities for review without deciding their cause
- A comparability register separating fully comparable, conditionally comparable, and source-specific variables
What you receive
- Harmonized dataset and variable dictionary
- Mapping tables and identifier crosswalks
- Transformation code with a decision log
- Comparability and limitations register
03 / Reproducible analysis
Analysis that reruns end to end and survives review
Move a manual, laptop-bound analysis into a versioned workflow that regenerates every table and figure on demand.
The challenge
Many analyses live in a chain of spreadsheets, notebooks, and undocumented steps. The result is credible only as long as the person who produced it is available and remembers every choice. A revised dataset or a reviewer's request can mean weeks of rework.
How we make it defensible
A second person reruns the work from scratch before it is accepted, so reproducibility is demonstrated, not asserted.
Typically for: Funded researchers, labs preparing publications, consultancies with recurring reports
What we build
- Separation of ingestion, preparation, analysis, and reporting stages
- Explicit configuration for parameters, exclusions, and specifications
- Scripted tables and figures generated directly from code
- Pinned environments and tests on sample sizes, exclusions, and reference results
- A fresh-environment rerun by a second person as part of acceptance
What you receive
- Version-controlled analysis workflow
- Regenerable tables and figures
- Environment specification and runbook
- Test results against approved reference outputs
04 / Monitoring & evidence systems
Indicator and monitoring systems that trace to source
Replace scattered partner spreadsheets with one system in which every reported indicator links back to its definition and records.
The challenge
Programme data arrives from many partners in many shapes. Indicators are rebuilt by hand each reporting cycle, definitions drift, and when a funder asks where a number came from, the answer takes days to assemble.
How we make it defensible
Each published figure links to the calculation version and the source submissions that produced it.
Typically for: NGOs and MEL teams, programme evaluators, foundations
What we build
- Indicator definitions turned into versioned, executable calculation rules
- Submission templates and validation, with actionable exception reports for partners
- Deduplication and disaggregation logic with stated treatment of missing data
- Reporting runs that record input versions, cut-off dates, and review status
- Dashboards and automated reports built on top of the traceable layer
What you receive
- Indicator dictionary and calculation code
- Validated submission pipeline
- Versioned reporting outputs
- Dashboards and a reporting runbook
05 / Research software
Research tools and data platforms built around your workflow
Purpose-built applications, APIs, and internal platforms that fit how your researchers actually work.
The challenge
Off-the-shelf tools rarely match a research workflow. Teams end up stitching together scripts, shared drives, and manual handoffs that only a few people understand and that break when staff change.
How we make it defensible
Delivery includes the documentation and tests that let another engineer maintain the system.
Typically for: Research centres, consultancies, teams with recurring internal workflows
What we build
- Requirements drawn from the real research workflow, not a generic template
- Data applications, internal tools, and APIs with automated testing
- Access control and audit trails appropriate to the data involved
- Deployment in your environment or an agreed hosting setup
- User and developer documentation, so the tool can be operated and extended
What you receive
- Tested, deployed application or API
- User guide and developer documentation
- Source code and deployment instructions
- Support terms where contracted
06 / Text & document analysis
Large document collections, structured for analysis
Make policy documents, transcripts, and archives searchable, classifiable, and analyzable, with humans verifying what matters.
The challenge
Thousands of documents hold evidence no one can read at scale. Keyword search misses meaning, and unreviewed automatic classification creates errors that are hard to see.
How we make it defensible
Automated coding is always evaluated against human-verified samples, and the results are reported with their limits.
Typically for: Policy research groups, think tanks, qualitative and mixed-methods teams
What we build
- Collection building from authorized sources, with metadata and versioning
- Text extraction and cleaning that preserves document structure
- Classification and coding schemes designed with your researchers
- Language and machine-learning techniques used only where they fit, with sample-based human verification
- Quality reporting that distinguishes model output from verified coding
What you receive
- Structured, searchable corpus
- Coding scheme and annotation guide
- Classification pipeline and evaluation report
- Documented limitations and error analysis
Your situation does not match one of these exactly?
Most real projects are a combination. Send a short brief and we will shape the right scope with you.