Audit Timing: Seasons, Sales and When Not to Judge

Judge a funnel against its own season's shape. Audit deep work in representative trading weeks, not during sales, launches or stock crises; verify repairs against the same season's baseline segment; and keep a written calendar of the year's distortions (promotions, seasonal peaks and troughs, campaign flights, the trade's calendar) attached to the baseline, so a future reader can tell weather from climate. Most 'mystery' movements in small-store funnels are calendar events that nobody wrote down. The calendar is four lines a quarter and it retires more false findings than any tool.

List yours honestly: retail has its peaks and its January; services have weather, budget cycles and school terms; software has renewal seasons and conference months; every business has its promotional calendar and its stock or capacity cycles. Add the irregulars as they are scheduled - a price rise, a range change, a marketplace promotion - because a distortion known in advance is data, while the same event discovered during a read is an argument. The calendar lives with the baseline and is consulted before any audit date or read window is approved.

The deep audit belongs in a representative stretch - ordinary traffic, ordinary range, ordinary prices - ideally six to eight weeks before your peak season, so its fix list ships and verifies before the peak makes everything unreadable and precious. The hour audit runs monthly regardless, with its findings window-flagged per the mistakes guide. Auditing during the peak is not forbidden; it is specialised: peak audits hunt mechanical defects and capacity failures (the site under load, the checkout under volume), and their rate findings are quarantined until a normal window can try them.

Sometimes a repair ships at a season's edge and its window straddles the change. The honest options: extend the read until a comparable clean window exists inside the new season and compare against the new season's prior-year or early baseline; split the verdict (kept on mechanism and qualitative evidence, rate verdict deferred); or revert-to-baseline comparisons using the same weeks last year where the business has the history. What is not available is pretending the boundary is not there. The log entry names the boundary and the option chosen, so the verdict's confidence is part of the record.

Same-period-last-year is the small store's seasonal control, and it misleads when the business changed shape between the years: new ranges, new channels, a price architecture replaced. Use year-ago for shape (the week's pattern, the season's depth) rather than for level, unless the shape held. Where the business is young and last year is another country, the honest control is the season's own early weeks plus the distortion calendar's notes - thinner evidence, labelled as such, beats a confident comparison to a business you no longer run.

At the year planning sitting, place the four deep-audit windows, the peak-season defect watch, and the baseline refresh dates on the same calendar as the trading events they must avoid or serve. Measurement planned in advance survives contact with a busy year; measurement improvised around one does not happen, and the year passes with opinions where the funnel file should be. Four windows, written down in January, is the entire ask - and it is the difference between a business that knows its seasons and one that is surprised by them annually.

Trading seasons are only half the calendar; the other half is the business's capacity to act. An audit delivered into the quarter when the one technical person is migrating platforms, or the owner is opening a second location, produces a fix list that ages unworked - and an unworked list teaches the business that audits are theatre. Place the deep audit where its list can ship: the review rhythm should hold one honest conversation a year about capacity, setting the audit count and sizes to the hours that genuinely exist. A smaller audit fully worked compounds faster than a grand one admired and shelved.