Field notes · Depletion data

Seasonal SKUs vs. real churn

By the Sightglass team6 min read24 July 2026

Every autumn, a chunk of your book appears to be dying. The summer-weighted brands come off taps and shelves, the September file lands thinner than August, and any rule that flags a month-over-month drop hands your reps a list topped by accounts that have done nothing wrong. Call down that list and most of the people on it will tell you nothing has changed.

The cost isn't only the wasted week. A rep who gets three false alarms in a row stops opening the list, and the one account on it that genuinely walked goes unnoticed along with the rest. Separating a seasonal trough from a real loss isn't sophisticated work. It just has to happen before the list goes out, not after.

Why the file bends in autumn

Depletion data records movement from the distributor to the account. It does not record what got poured. An account winding down a summer brand stops reordering it weeks before the last keg blows, so the drop in your file leads the actual season and looks steeper than the sell-through ever was.

Seasonal buying is also lumpy. An account that took four cases of the summer brand every month may take the autumn release in one drop-in load and then show nothing for two periods. Read month over month, that's a collapse. Read against the release calendar, it's a purchase pattern. Both distortions land in the same few weeks, which is why autumn is when a naive churn rule does the most damage.

Compare like month to like month

The fix is unglamorous. For anything with a season in it, stop comparing this month to last month and compare it to the same month a year ago. Month over month tells you an account is doing less than it was. Same month last year tells you whether it's doing less than it does at this time of year. Only the second one is a judgment.

This month vs. last month. Movement, not health. On a seasonal SKU, between August and November, it tells you close to nothing.
This month vs. the same month last year. The honest read on a seasonal SKU. It needs thirteen months of history, which a rolling export window may not carry — so keep every file you pull.
Rolling twelve months vs. the prior twelve. Takes the season out entirely. Good for judging the account; slow to show a break that started six weeks ago. Use it for the relationship, and same-month year-over-year for the SKU.

One trap in like-for-like: a moved release date. If the summer brand shipped in April last year and May this year, April reads as a catastrophe and May as a triumph, and neither happened. When the calendar moved, compare season to season — the whole release window against last year's whole release window — not month against month.

Did the cohort move, or just the account?

Year-over-year tells you an account is down. It doesn't tell you whether that's the weather or the relationship. For that, compare the account to everyone else who bought the same thing.

01Tag every SKU: year-round, spring/summer, autumn/winter, one-off. Do it once — it changes about as often as your portfolio does.
02Pick a seasonal SKU and pull every account that took it in the same period last year. That group is the cohort.
03Measure the cohort's move in aggregate for the current period, same month year over year.
04Judge each account against the cohort, not against zero. Down roughly in line with its cohort is the calendar. Down materially further than its cohort is a question worth a call.
05Run the same cut by distributor. A cohort that goes quiet under one distributor while holding everywhere else is usually a supply story — out of stock at the warehouse, a delisting, a rep who left — and the call goes to the distributor, not the bar.

What they dropped matters more than that they dropped

An account that drops the summer brand in October and keeps ordering your year-round lineup is doing what good accounts do. The tap handle changed with the season; nothing about the relationship moved. The account you actually want is the one where the seasonal SKU came off and the core order quietly thinned in the same period — one less case a month on the everyday beer, no announcement. That account isn't rotating. It's drifting.

The test that matters
A seasonal SKU going quiet in autumn is a calendar event. A year-round SKU going quiet in the same account, in the same period, is a sales event. Rank on the second, use the first as context, and never the other way around.

It's worth doing the arithmetic on how small the real list is. Reading one importer's own VIP exports, we found 630 accounts quietly sliding — 4.8% of the book. A list that size is workable on a Monday. Bury it in seasonal false positives and nobody reads it twice.

Winter releases, and what to exclude

Short-window SKUs need their own rule, because they spend most of the year legitimately at zero. A winter release judged on an autumn file shows nothing — not a failure, an absence. Judged on a March file it looks like a collapse, because the product ended. Neither is a result. A release with a ten-week life can only be read against its own window: last year's November-to-January against this year's, across accounts that could actually get it. A first-year release has no window to compare to at all, so the cohort is the only honest read — and that's a comparison to expectation, not to history.

Depletion data lags — a day or two at best, longer where a distributor reports on a slower cycle — and a file that arrives late makes the most recent period look thin for reasons that have nothing to do with the account. On a seasonal read, that lag falls exactly where the season turns. Wait for a second period before calling anything a loss.
Seasonal SKUs inside their off-window. Not a loss until the next on-window also comes up empty.
Discontinued and delisted items. The account didn't leave; the product did.
Accounts whose entire history is one seasonal SKU. They were never year-round accounts. Work them in season; don't count them as churn in November.
Accounts that changed distributor in the period. The volume is often still in the book under a new account number, and the old row going quiet is bookkeeping.
Package changes. A SKU that moved from 12oz to 16oz reads as one product dying and another being born.
Seasonal venues — ski resorts, stadiums, beach bars, college-town accounts that empty out over the summer. Closed for the season is not lapsed.

Or have it sorted before it reaches a rep

All of this is doable by hand, and worth doing once if only to see how much of an autumn churn list turns out to be weather. It's the doing-it-every-week part that doesn't survive a busy quarter. Sightglass reads the VIP feed you already pay for each night and keeps every snapshot, so the same month a year ago is still there to compare against after it has aged out of the current export window — an export is a transaction file, not an archive. Because that history is there, an account can be read against the same month a year ago rather than against its own last month, and a dropped SKU shows up as a different thing from a core order that is quietly thinning. When the data is too fresh or too thin to support a conclusion, it says nothing. Nothing goes out without a person approving it, and your reporting stays exactly where it is.

Want to see it on your book? Twenty minutes, screen shared. No file required. Book a 1-1 demo