Anonymous Insurance Company

In Property & Casualty insurance, pricing accuracy is a direct driver of profitability — mispriced policies either erode margins or lose customers to competitors. This SME insurer had accumulated meaningful policy and conversion data but lacked the ML expertise to extract value from it. Underwriters were setting premiums manually, with adjustments constrained to less than 2% — a range too narrow to reveal how customers actually respond to price variation. Without data on conversion behavior at higher price points (5–15%), the company had no foundation for a statistical pricing model. The result: revenue systematically left on the table, with no clear path to fixing it.

Tribe AI began with a structured discovery phase, embedding a product manager experienced in high-value ML scoping to align stakeholders on the highest-priority use case. The team identified dynamic pricing optimization as the greatest near-term value driver. Because the company lacked training data at meaningful price variation ranges, Tribe designed an intermediate live experiment rather than a conventional A/B test: most policyholders saw adjustments of up to 5%, while customers with high predicted conversion likelihood were exposed to changes of up to 15%. This segmented approach — based on the hypothesis that high-intent customers are less price-sensitive — simultaneously generated the training dataset needed for a full predictive ML pricing model and produced measurable business impact from day one. The experiment ran in production against live policy data, with controls for distributor variability and changing market conditions.

The intermediate pricing experiment delivered 2.5% premium lift company-wide and reached positive ROI within the first week of launch — an outcome the Tribe product manager described as rare in 20 years of pricing work. Projections based on experiment data indicate that expanding the dynamic pricing model to the full policy book would yield a 7–12% lift in premiums across the company. Beyond the numbers:

• Underwriters shifted from manual, narrow-band adjustments to model-informed pricing decisions
• The project revealed additional book-of-business insights and identified new areas where ML could improve decision-making
• The company's leadership, self-described as "insurance experts, not statisticians," moved from ML skepticism to treating data science as core to business strategy

• Lead with a scoped discovery phase. Before any model development, align business stakeholders on the highest-value ML use case — this prevents wasted cycles on technically interesting but low-impact work.
• When training data doesn't exist, design an experiment that creates it. A live pricing experiment can generate immediate ROI while building the dataset a future model requires.
• Use conversion likelihood to segment experimental exposure. High-intent customers tolerate larger price moves, enabling wider price variation data without significant churn risk.
• Fast ROI builds organizational trust. Demonstrating measurable results within the first week unlocks executive buy-in and opens the door to broader ML initiatives across the business.