Seven stories of repetitive work handed to a system, from newsletter assembly to prioritization and fraud detection, the judgment left with a person.
Two engagements sixteen years apart, a newsletter and a content-decision layer, share one rule: automate the low-leverage manual work when it costs more than the tool, and keep the editorial call with a person, since capacity and objectives stay human judgment.
When card non-payments surface weeks too late to recover, this shows how a layered, in-house scoring system put a screen on each acceptance decision and pulled one shop’s losses from hundreds a year to fewer than a dozen, then fell quiet enough to need little upkeep.
Lean operations, in this piece’s sense, is an operating model rather than a toolbox or a cost cut: find the structural cost inside a process, re-engineer its mechanical layer, and free the human time for work worth doing, so the saving outlasts the fix.
You cannot assume a decision system built for an ad-funded media site fits a lead-driven training business: the scales stopped comparing, the conversion data ran sparse, and the manual upkeep is what pushed the tool from a one-off into a back-office product.
Prove the mechanism by which a decision-support algorithm scales heterogeneous KPIs logarithmically and anticipates the gain of each update, so a reader can see how a repeatable prioritization call becomes a ranked, human-actionable list rather than a claimed black box.
Prove that AI-era operational relevance is evidenced by two real engagements (AI-assisted content production, Simplébo 2023 and Aprovall 2024) showing the same systems-building instinct on each side of the moment AI became mainstream (one just before, one just after) rather than a late pivot, scoped to exactly what those engagements demonstrate.
Prove, with a decision-support algorithm built for one media site’s editorial prioritization, that a system designed once against real data can outlive the engagement that funded it and keep serving other operations.
