Nobody loses £600,000 in an afternoon.
That is the first thing worth saying, because it is the part most people get wrong. A loss that size does not arrive as a decision. It arrives as a series of small, individually defensible situations that nobody connected until the year was already over.
I have the full post-mortem in front of me. Six categories of cause. Every one of them was visible, in some tool, to somebody, at the time.
None of them were visible together. That is the whole story.
One: the market moved first
Inflation, flat wages, and a category built entirely on discretionary spend. Garden furniture and barbecues are what people buy when they feel comfortable. In a year when they did not feel comfortable, they bought the cheaper version from a supermarket instead.
This part was not a mistake. It was weather. You do not get to control it and you do not get points for predicting it.
What you do get is a choice about what to do next, and that choice depends entirely on how quickly you see it.
Two: the algorithm reweighted, quietly
Amazon's ranking logic shifted. Listings optimised against the old standard slipped. Nobody sent an email about it.
The pay-to-play effect compounded it. Competitors buying more visibility took ground that organic ranking used to hold. Ad spend that had been adequate the previous year was now, in relative terms, a cut.
Nothing broke. A slope just got slightly steeper, and it kept being slightly steeper for eleven months.
Three: stock ran out at exactly the wrong time
This is where it stops being weather.
Key products went out of stock in August and stayed out. Out-of-stock products cost the business real money during Prime Day alone.
The part that matters is not the lost sales. On Amazon, a stock-out is not a pause. It is a demotion, the kind of compounding failure Fin Academy documents in its System Drift framework. Ranking earned over months decays while you are unavailable, and it does not come back when you restock. You pay for the outage twice: once in the sales you did not make, and again in the position you have to buy back afterwards.
The inventory tool showed a stock-out. It did not show a ranking consequence, because that is not what inventory tools are for.
Four: Prime Day was lost before Prime Day
Sales during the event dropped 17% year on year. The first day was down 36%.
The reason is almost too neat. Nine products approved for deals made no sales at all, while several genuinely high-traffic products were never submitted. The slots went to inventory that needed clearing rather than inventory that could convert.
That decision was made weeks earlier by someone with a spreadsheet and no view of what traffic was actually doing. It was not stupid. It was uninformed, which is different, and far more common.
Traffic across the event was down 17%. By the time anyone could see that number, the event was over. Prime Day performance is decided in the six weeks before it and reported in the two days after.
Five: the categories told different stories
Garden furniture fell 33%. Barbecues fell 11%.
Two numbers, two completely different problems, one dashboard displaying them side by side as though they were the same kind of thing.
Furniture was competitive saturation. Dozens of sellers, near-identical products, individual lines losing up to 90% of their traffic regardless of how well the listing was written. Better keywords do not fix a category with too many people in it.
Barbecues were an internal failure. The underperforming lines had one thing in common: no ad spend, and a higher price than the competition. Products priced above the market with no visibility budget do not sell. That is not a market condition, it is arithmetic.
One of those needed a strategy change. The other needed someone to move a budget. Treating them as the same problem meant neither got solved.
Six: the fees kept arriving
Excess inventory fees on slow-moving stock. Long-term storage penalties. Return rates on furniture doing what return rates on furniture do.
The stock that was not selling was also, quietly, costing money to not sell. That is the compounding part. A bad quarter is expensive. A bad quarter with a warehouse full of the wrong thing is more expensive than the sales figures suggest.
What all six had in common
Look back at the list.
Every one was recorded somewhere. Inventory levels were in a system. Ad spend was in a system. Ranking was visible. Traffic was visible. Category performance was visible. Fees appeared on a statement.
Nobody was working without data. There was more data than anyone could read.
What did not exist was a view where a stock-out in August connected to a ranking decline in September, which connected to a weak Prime Day in October, which connected to a deal submission made by someone who had seen none of it.
Each tool answered its own question correctly. The business needed an answer to a question none of them were asked.
That is not a reporting problem. It is a decision problem, and reporting tools do not solve decision problems no matter how many of them you buy. The Amazon World FAQ covers a lot of this system behaviour in more detail, if you want the mechanics rather than the story.
What actually changed
The recovery was not clever. It was structural.
Stock forecasting stopped being a warehouse concern and became a ranking concern, because that is what it actually is. Deal submissions started being made against live traffic data instead of clearance priorities. Ad budget moved to where visibility was being lost rather than where it had historically been spent. Pricing got reviewed against the market on a schedule, rather than when somebody noticed.
None of that is sophisticated. All of it required seeing more than one thing at a time.
The business came back past £1M.
Why Finnito exists
I did not set out to build software. I set out to stop having that year again.
The tools were never the problem. The problem was that six systems each told the truth about their own small area, on their own schedule, to whoever happened to be looking, and not one of them was ever going to say: this is going wrong, here is why, do this now.
Operators do not need another dashboard. They need the signal early enough to act on, and a clear enough read of it to know what to do. That is what FinFlow is being built to be.
That is the whole thesis. It cost £600,000 to learn.