The accounts missing from your depletion report
A depletion report can only list accounts that bought something. When an account stops ordering altogether it doesn't appear as a zero — it stops appearing at all. No row to sort, no blank cell to notice, no flag to filter for. It was in last year's file and is nowhere in this one, and nothing in the report says so. The customers you are losing are, mathematically, the ones you cannot see in the standard report.
That is not a flaw in your reporting. It is what a transaction file is: a record of transactions. An account that transacts appears; an account that doesn't, doesn't. The gap becomes visible only when you hold this period's file up against a stack of earlier ones — the one piece of work that never reaches the top of a Monday.
Why the gap is invisible
Three things stack.
Volume. A single export runs to tens of thousands of rows — one line per account, per product, per period. One importer's monthly file that we worked through came to roughly 30,000 rows naming 13,109 accounts. That is not a document; it is a database wearing a spreadsheet's clothes. The honest ceiling of what anyone does with it on a Monday is: open it, wait for the filter dropdowns to populate, sort by cases descending, read the top forty, close it. Everything read that way bought recently and bought a lot. The accounts that took none are not in the file to scroll to.
Absence. Every other problem in your book is a value: a number that fell, a rank that moved, a margin that thinned. You can sort it, filter it, put a conditional format on it and let it find itself. A lapsed account is not a value. It is a missing key, and there is no column called 'stopped'. To see it you need a second list — the accounts you had before — and something that compares the two. No single export knows what last year's file said, so the information exists only across files: a different kind of work from reading one.
It is also why the largest losses are the quietest. An account that halves its order still files rows, and rows land in a variance column. An account that leaves files nothing at all.
Timing. Quarterly is too slow, and not for analytical reasons. Draft lines, cold-box doors and shelf facings are finite and get reallocated continuously; a handle that went to somebody else in March is not free again in June because you noticed in June. Winning space back means displacing whatever replaced you, and asking a buyer to reverse a decision they have already explained. On a quarterly cadence you hear about the last order one to three months after it landed — by which point the call has stopped being a save and become a fresh sell.
Do it by hand, once
You can build this in a spreadsheet, and the first build is an afternoon. Pull twelve months of depletions, one export per period, and keep every file. The archive is the whole trick: no single export reaches back far enough to show an absence, so whether this report exists in a year is decided by whether the folder starts today.
Two traps sit in the middle of that; both manufacture ghosts.
Refresh it monthly, a few days after the new export lands, right behind the archive pull. The file is monthly, so a weekly rebuild mostly re-reads the same numbers; the flagged list is what gets worked through the weeks that follow. The refresh takes about twenty minutes, nearly all of it on account keys you haven't seen.
What to do with the list
A list is not an outcome. The report is finished when a named person has a named account and a first sentence. Two hundred rows in a shared folder is the same as no report, and costs more.
The order of those last two matters. The distributor rep knows what your export cannot: whether the account is still open, whether the order simply stopped being placed, whether the buyer left in April. Turning up at a bar to ask why they stopped buying, when the answer is that the truck stopped coming, spends credibility you will want later.
Telling a lapsed account from a closed venue is mostly signals you already have, plus a few that are public record.
What one importer's data showed
We ran this comparison across more than a year of one importer's own depletion exports — their data, our analysis. None of it had surfaced in their monthly reporting, and nothing was wrong with that reporting: it was doing what a transaction file does.
Read the middle number carefully: it is not 5,085 lost customers. A book that size accumulates venues that closed, venues that changed hands, one-time buys, and accounts a distributor re-keyed. The honest statement is narrower and still large — five thousand accounts that ordered in the prior year did not appear in this file, and 80,932 cases sat behind them. Nobody could say which share was dead weight, because nobody had the list.
The other half of the finding is about accounts still in the file — sliding rather than gone.
The 630 are the reachable ones: still filing rows, still taking deliveries — and roughly one account in twenty was carrying essentially the whole annual decline. The distributor figure is a different warning. Not every quiet wholesaler is a problem — plenty are dormant by design. But a quiet distributor takes every account behind it dark at once, which is why that column is worth reading before anybody works the account list.
Nobody at the company knew any of this, and not because anyone had been careless. The data was correct and sitting in files they already owned. What was missing was an afternoon to line a year of them up next to each other, in a quarter that had no spare afternoon in it.
Or have it done every night
That is the weakness of the manual version. The build is genuinely an afternoon. What it doesn't survive is a quarter with a price increase, a distributor transition and a national account review in it. A month gets skipped, then another; the keys go stale, the history test starts failing quietly, and the sheet goes on producing a number nobody quite trusts.
Sightglass reads the VIP depletion feed you already pay for, every night, and keeps every snapshot, so history accumulates well past any single export's window. It watches for the gaps, allows for the reporting lag, and hands each rep a short ranked list of their own accounts — the reason in plain numbers, the outreach already drafted in your house voice, one to the account and one to the distributor. Nothing sends until a person approves it, and approved mail goes from the rep's own mailbox, so replies come back to them. It arrives in Slack, in an email digest, or in the dashboard.
Leaders get the same data as conclusions rather than a grid — lapsed accounts, distributors gone quiet, concentration, territory by territory — every figure traceable to the file it came from. When the data is too thin to support a conclusion, it stays quiet rather than inventing one. Your VIP subscription and your existing reporting stay where they are.