Financial Institutions
Insurance Agency Performance & Loss Ratio
How an insurer separated genuine agency underperformance from bad data — and why the retention programme had to stop using loss ratio as a proxy.
View projectIndependent analytics practice · Ottawa
We are a small, founder-led practice. The person who scopes your project is the person who does the analysis and presents the result.
Who we are
Elexipdata is a founder-led analytics practice based in Ottawa. We started in 2018 with a simple observation: most analytics projects fail before the model stage — in the question, the data or the handover.
We stay deliberately small. The person who scopes your project is the person who does the analysis and presents the result. For specialist work — data engineering, design or sector expertise — we bring in trusted collaborators, and you always know who is accountable for what.
That model is not for every buyer. If you need a large team on site for a year, we are not the right fit. If you need a clear answer to a specific decision, delivered by the people who did the work, we probably are.
“We would rather tell you that analytics cannot answer your question than sell you a model that cannot either.”
What we are hired for
Each of these draws on the same four capabilities — data analysis, business intelligence and visualization, model building, and AI automation — applied to a specific decision.
01
Who is leaving, why, and which intervention is worth the cost. We build churn scores around retention capacity and customer value — not around a model leaderboard.
Related work02
Default, delinquency, fraud and counterparty risk. Explainable scores, reason codes, threshold policy and monitoring that can survive an audit.
Related work03
One definition of revenue, margin or utilisation; a data model that reconciles; dashboards that operations and finance both trust.
Related work04
Private LLM assistants, retrieval over your own documents, document processing and predictive workflows — with evaluation, permissions and human review.
Related workHow we work
Who will act, what action is available, and what would change their mind? If we cannot answer that, we do not start modelling.
Grain, keys, missingness, sentinel values and leakage. Most disagreements about results are really disagreements about data.
A transparent baseline people trust beats a complex model nobody can explain. Complexity has to earn its place.
Row counts, totals, a second method and a holdout period. We do not ship a number we cannot reproduce.
Documentation, thresholds, monitoring and training. The work should keep running when we leave.
Selected case studies
Financial Institutions
How an insurer separated genuine agency underperformance from bad data — and why the retention programme had to stop using loss ratio as a proxy.
View projectHealthcare
A pre-discharge readmission score built around patient-level validation, discharge-disposition leakage and the care team's real follow-up capacity.
View projectCross-industry
How an operations team replaced scattered documents and shadow AI use with a permission-aware assistant — citation-first answers, a rerunnable evaluation set and content governance that keeps it honest.
View projectTell us what decision you are struggling to make. We will tell you whether analytics can help, what data we would need, and what a first phase should look like.