Commercial trials and proportionate commitment 5 min read
The practitioner
Jaspreet Qezvathi
The first question I should have asked was when the model could know its strongest input. Instead, I asked whether its score was high enough. Our retrospective renewal test correctly classified 182 of 200 contracts, and I approved the 91% result for a management discussion. The model was intended to predict thirty days before a rental ended. A renewal invoice used as an input had been created after that point. It was not an early signal. It was part of the later outcome leaking into an earlier decision.
I approved a renewal model after it correctly classified 182 of 200 retrospective contracts. The 91% score looked like a reason to move quickly. A review then found that one input was a renewal invoice created after the date on which the prediction was supposed to be made. The model had access to part of the answer. We withdrew the performance claim and rebuilt the evaluation before allowing it to influence customer decisions. My approval, not only the model's design, needed correction.
I am Jaspreet Qezvathi, fictional CDO of Qazvori Workcycle Hire in Vadodara. We rent manually pedalled cargo cycles to adult business users through nine regional depots, with 178 employees and ₹48 crore FY 26 revenue. Eighteen data colleagues work through three leads within ₹2.8 crore FY 2026–27 operating expense and a separate ₹1.1 crore capital programme. Mira is CEO and Harish leads operations. Workshop, financial, legal and operating owners retain their own decisions; our models do not automatically refuse hire or set prices.
My twenty-nine-year career began in survey analysis. The questions have become more consequential, but the discipline still starts with how the evidence was produced. Completed hires cannot reveal every request that an unavailable cycle prevented. A small randomised reminder trial can offer a useful signal without settling a full rollout. A depot report can expose a business's activity even after its name disappears. I want a decision-maker to see those limits before a clean chart turns them into an assumption no one remembers making.
A new evaluation instead of a better-looking claim
A ₹41 lakh rebuild restricted inputs to those available at the intended prediction date. A separate prospective shadow cohort produced 154 correct classifications from 200 contracts, or 77%. It was a different cohort, so the figures do not establish a measured fourteen-point decline. Baselines, error consequences and commercial usefulness still required assessment.
Completed hires did not reveal the unmet request
Across the same twenty-eight days, two selected depots recorded 120 and 80 completed hires. The second had cycles available on only sixteen days. A ₹32 lakh observation programme was commissioned before expansion decisions; we did not extrapolate a demand total or call the first depot the stronger market from those counts alone.
A trial result kept its uncertainty
Four hundred eligible business accounts were randomly assigned to 200 receiving a reminder offer and 200 following the usual process. Forty-two and thirty-four renewed in the same eight-week window. The observed four-percentage-point difference supported further planned evaluation, not a settled claim of commercial benefit or an immediate full rollout.
Current focus
Evaluation at the real decision date, availability-aware demand research, commercial experiments and careful reporting of customer activity.
Conversations I welcome
A first external board contribution involving data-supported service decisions, after preparation and consent. IICA registration has not begun; assessment is not completed, no exemption. No statutory seat; twenty-four provisional days.
Beyond the page
The record behind the perspective.
Formation
Bachelor of Science in StatisticsTevnuli College of Statistics · 1997
As CEO, I saw Jaspreet withdraw the model score he had approved and explain why the later shadow result answered a more relevant question. He did not turn the corrected evaluation into a claim that deployment was already justified.
Mira QevnuliChief Executive Officer, Qazvori Workcycle Hire — fictional colleagueThe full record ↗
Contact
A first external board contribution involving data-supported service decisions, after preparation and consent. IICA registration has not begun; assessment is not completed, no exemption. No statutory seat; twenty-four provisional days.
