HomeCase Studies

Production case studies.

Four production systems running inside client analytics environments. Each entry covers the problem, the technical approach, the tools, and the form of evidence available on request.

§ Shipped systems

Four examples.

Each is a real engagement. The system and client names sit behind approvals; supporting diagrams, screenshots, and before/after outputs are available on request.

/ shipped, not speculative
/ 01 · Forecasting · BigQuery · GA4Live · recurring

Channel and query forecasting.

Forecasting workflows in BigQuery covering GSC query trends, channel traffic, and monthly reporting cycles. Used by analyst and account teams to flag variance against forecast inside the cycle rather than after it. Forecast confidence bands surface the range, not just the point estimate.

120k 90k 60k NOW Jan Jun Dec
BigQueryPythonGSC APIGA4ForecastingReporting
Median MAPE5.4%
Runs / month12
Channels4
EvidenceOn request
Request approved samples
/ 02 · Sentiment · LLM-assistedLive

Sentiment and topic river.

LLM-assisted classification across press, review, and social. The system classifies, deduplicates, and surfaces shifts; nothing is auto-published. Annotated bands track positive, neutral, and negative coverage over time so analysts can interpret context, not just deltas.

Coverage spike · Q2 Topic shift · Q4 Jan Jun Dec positive neutral negative
PythonLLM APIBigQueryNLPStructured outputs
Surfaces tracked11
Mentions / week~3k
Alert classes4
EvidenceApproved decks
Request method notes
/ 03 · AI visibility · AI OverviewsOperational

AI vs traditional search.

Tracking how brands and queries surface across AI Overviews, ChatGPT, Perplexity, and conventional search. Each change captured with date, prompt, and rendered citation so visibility shifts can be reviewed weeks later, not just observed in real time.

Google · organic competitor a · category guide competitor b · review hub your brand · category page forum · long-tail thread AI Overviews · cited your brand · cited 1/3 competitor a · cited 2/3 trade outlet · cited 3/3 prompt: "best [category] tool · 2026" Jan Jun Dec
AI OverviewsChatGPTPerplexitySERPResearch
Queries tracked140+
Platforms4
CadenceWeekly
EvidenceMethod notes
Request method notes
/ 04 · Reporting automationLive

Manual reporting → automated assembly.

Automated assembly of recurring reports and decks from GA4, GSC, and ad-platform data. Replaces a multi-hour manual build with a short analyst review pass. Each step on the before bar is a real stage; the after bar shows what remains.

0h 1h 2h 3h 4h Manual Gather ~60 min Clean ~50 min Build deck ~60 min Review ~40 min Send ~30 min 4h · manual vs Automated Review ~20 min 20m · review
PythonGoogle Slides APIBigQueryAutomationClient delivery
Reports / month36
Time saved~70%
Variance flagsInline
EvidenceBefore / after
Request before / after
§ Supporting documents

What's in the file.

Each case study has documents catalogued and shareable on request. Approvals required for some artefacts.