Sunday night an order came into my collectibles storefront. One buyer, five vintage action figures from the same 90s line, a nice basket with my volume discount applied. The kind of order you screenshot.
Some context for readers who know me from my main business: the storefront is a side business. I run it with my son on nights and weekends, in the hours left over after the day job. Keep that in mind; it is where this story actually lives.
Wednesday morning I pulled the figures to pack them. I found two of the five.
Here is what the order history showed once I stopped trusting my dashboard and started reading receipts. Three of those figures had already sold in August 2025, under an older order, shipped and gone. At some point after that sale, their inventory counts silently reset themselves back to one. Nobody touched them. No restock, no return. The listings just woke back up, sat there looking sellable for eleven months, and then sold again.
Sold twice. On hand: zero.
Owning it first
Before touching any systems, I emailed the buyer. Not a canned "unfortunately your order has been cancelled" template; an honest note that I was doing a warehouse check, that three of the five had a records problem on my end, and that they could either take the two I had in hand with a partial refund, or cancel the whole thing.
They chose the full cancel. $64.84 refunded, all five lines. I thanked them, told them more figures from the same line are hitting the store in the next couple of weeks, and put a shipping credit on their next order. A refund costs me the sale. The apology done right cost me nothing and probably kept the customer.
The fork every small operator knows
Then came the real decision. I had two bad inventory rows. The cheap move is obvious: fix those two rows, feel careful for a week, move on. Every reseller who has ever double-sold something knows that move, because until recently it was the only affordable one.
This is the part where running a side business is different in kind, not just in size. A business that gets your broken-up evenings does not give you the long uninterrupted stretches where patterns jump out at you. Silent failures thrive in that gap. The honest fix is safeguards that watch when you cannot, but safeguards are systems, and nobody quotes you an inventory reconciliation system for a side business at a price a side business can pay. You ARE the reconciliation system, and you run whenever you happen to remember.
I run this business with an AI assistant doing the heavy technical lifting, and that changes the economics of the fork. "Do it properly" stopped meaning a five-figure line item on somebody's proposal and started meaning the rest of the same day.
So the same day, we built the class fix:
- A resurrection alarm. Anything that sold and later reappears with stock trips a flag. That is the exact failure that bit me, and it can never happen silently again.
- Sold-but-still-listed detection. Anything with a completed sale that never got archived gets archived automatically, with deliberate exemptions for preorders and restockable items.
- A refund watch. Refunds that never resulted in a relist or a restock get surfaced instead of evaporating.
- One-way sync from the storefront to my master database. The store is the source of truth; the spreadsheet follows it, never the reverse.
- A platform automation that archives any listing the moment it hits zero. We tested it live: zeroed an item, watched it archive itself in about twenty seconds.
The first run paid for itself immediately
The part I did not expect: the first full reconciliation pass found three MORE ghosts I did not know I had.
A refund from April 2025 where the item came back but never got relisted. That one was sitting in a bin, real and sellable; it is now live again at $109.98. A figure that sold in 2023 through an old channel with no location record anywhere; shelf check found it, records now honest. And one figure that appears to have sold off the books entirely; archived, flagged as revivable if it ever surfaces.
One incident, honestly investigated, audited my last three years of inventory history in an afternoon. The double-sell also synced 115 stale cells in my master database that had drifted from reality without any of them ever causing a visible problem. Yet.
What I actually learned
The incident is the cheapest audit you will ever get. The $64.84 refund bought me a finding a consultant would have charged real money to produce: my sales records and my inventory records had no referee. Wasting that lesson on a two-row patch would have been the real loss.
Fix the class, not the row. The two bad rows were symptoms. The class was "nothing reconciles sales against stock." Patch symptoms and you will meet the class again on a bigger order, with a worse customer.
AI does not just make the work faster; it changes which work is affordable. For a side business especially. The careful version and the quick version used to be different budgets. Now they are the same afternoon, which means "we will build that properly someday" is increasingly just a habit left over from the old economics.
I still had to catch the problem, read the order history skeptically, write the customer email, and walk the shelves. The judgment stays human. But the distance between "I should really have a system for this" and having the system running dropped from a someday project to a Tuesday.
If you run anything on the side, ask yourself one question: when a sold item reappears in your stock count, does anything notice? Until Wednesday, my answer was no.
~James
This article first ran on LinkedIn. Comments are open there. Back to all writing.