Halden, Part 1: The Leak
A protein powder brand spends $8,600 a month on Amazon Ads and loses money on almost every order it wins. Part one of a four-part story about how Halden took its advertising back.
A Fluvi story in four parts. This is part one.
Tuesday, 11:40 p.m.
The spreadsheet had nineteen tabs, and Theo Ruiz trusted none of them.
There was the search term report, downloaded from Amazon that afternoon and already out of date. There was a tab of bids he'd changed last week, which he'd meant to check this week, and a tab of bids he'd changed the week before that, which he'd forgotten about. There was a tab called KEEP — agency notes, written by people who no longer worked for the company. And there was a tab called ACoS by campaign, where the numbers turned red above 30%, and almost everything was red.
Halden sells whey protein. Maren Holt started it four years ago in a rented kitchen in Fort Collins, Colorado, and named it after the town in Norway her grandfather came from. Today it's seven people, a co-packer in Utah, and three products on Amazon US:
- Whey Isolate, Vanilla, 2kg at $64.99, the one that pays the bills. Six of every ten dollars Halden makes on Amazon come from this tub.
- Whey Isolate, Chocolate, 1kg at $39.99.
- Grass-Fed Whey, 2kg at $69.99, Maren's favourite and the slowest seller.
Theo runs the ads. He used to coach strength athletes, joined Halden to answer customer emails, and inherited Amazon advertising the way people inherit a dog: someone left, and it needed feeding. When Halden ended its agency contract in October, Theo got nine campaigns, a few dozen keywords, thousands of search terms, and a login.
He was good at it, in the way anyone is good at something they do at midnight. He knew which keywords sold. He knew that "whey isolate" was the search everyone in the category fought over. He knew, roughly, that some of the money was being wasted. What he didn't know was how much, where, and what to do about it that wouldn't take another four hours he didn't have.
The meeting where it stopped being Theo's problem
Priya Nair does Halden's finances three days a week. In the first week of February she put one slide on the conference room TV and didn't say anything for a while.
January is the best month of the year for protein powder. New Year's resolutions, new gym memberships, new tubs. Halden's ads had a big month too. They also had an expensive one.
"Every vanilla tub we sell through an ad," Priya said, "we keep about seventeen dollars and eighty cents after Amazon, after the co-packer, after refunds. We're paying about twenty-six dollars in ads to get each one."
She'd done the math the slow way. In January, Halden paid Amazon roughly $2,700 more for ad orders than those orders earned. In its best month.
Theo pushed back, because he believed it. "Ads aren't just ad orders. When we're on top for 'whey isolate', we rank. Organic sales come with it."
"How many?" Priya asked.
Theo didn't know. Nobody in the room did. That was the real problem. It wasn't that the ads lost money. Maybe they did and maybe they didn't, once you counted the organic sales they pulled in. The problem was that a seven-person company was spending over $100,000 a year on something it couldn't measure, steered by a spreadsheet one tired person updated at midnight.
Maren ended the meeting with one sentence: "I don't want a better spreadsheet. I want to know."
Connecting the account
Theo had seen Fluvi PPC on a forum thread about agencies. What got him to try it was the pricing line: All your data, free. Pay only for live agents. The free plan wasn't a trial. It read the whole account, with no limits on data, and nothing would change on Amazon until he said so.
He connected Amazon Ads on Amazon's own consent screen, then Seller Central, and went to make coffee. When he came back, Fluvi had imported two years of Halden's ad history and finished its first look at the account. The first thing he opened was the list of what it would fix.
A hundred and fifty-seven fixes, each one a specific change with its reason next to it, waiting for a yes or a no.
The biggest pile was the one Theo would never have found on his own. Sixty-four searches were leaking. These were searches like "whey isolate vanilla" and "whey for women" that Halden already had a carefully priced exact keyword for, still being caught by a broad or automatic campaign at a different price. The careful bid wasn't the one that won. For a lot of Halden's best searches, the price Theo had set by hand wasn't the price being paid.
Some of the fixes Theo half-knew about. The duplicate keywords were the agency's doing: the same phrase, "whey isolate", live in both the exact campaign and the broad one, the two bidding against each other in the same auction. Amazon only shows one of your ads per search, so Halden had been driving up its own price.
And the 34 search terms to harvest were shoppers telling Halden, in their own words, what they wanted: searches that had found Halden through broad and auto campaigns, converted, and never got a keyword of their own. "low carb whey isolate". "vanilla whey isolate 2kg". "protein powder for women". Theo had been meaning to go through the search term report for those since November.
He accepted the duplicate pauses that night. It took four clicks. He left the rest for the morning, because the next screen was the one that changed how Halden thought about its money.
Four tiers
Fluvi's classifier had read every keyword, every search term and every competing product in the account and sorted each one into four tiers, separately for each of Halden's products:
- The product itself. "halden whey isolate", "whey isolate 2kg".
- An adjacent need. "protein powder", "lean protein powder", "whey for women".
- Its domain. "clear whey", "protein shake".
- Unrelated. "yoga mat", "shaker bottle", "gym bag".
It wasn't a black box. Theo could open any tier, see what was in it, and move anything he disagreed with. A tier he changed stayed changed. And beside every keyword sat its search volume and, for the first time anywhere Theo had looked, Halden's own organic rank for it.
Then he filtered to tier 4.
"yoga mat". "shaker bottle". "gym bag". "protein bars", which sounds close until you remember Halden doesn't make bars. "whey pills", which isn't a thing. Then he grouped the last 60 days of spend by tier.
Eighteen percent of Halden's ad budget, about $1,550 a month, was going to searches that had nothing to do with protein powder. They brought in 4% of the orders, mostly flukes.
Meanwhile the searches that named Halden's own products, the shoppers who already knew what they wanted, got only 14% of the money and brought in almost a third of the orders.
Theo stared at that chart for a long time. He'd always known some of the budget was wasted. He'd pictured a few hundred dollars, a rounding error. He hadn't pictured a fifth of it.
What a click is actually worth
The third thing Fluvi showed him answered Priya's question, or at least the part of it that could be answered.
For every target in the account, Fluvi had worked out a bid: what one click is worth to Halden, from Halden's own numbers. It read the price, Amazon's fees, the cost of goods and the refund rate from Seller Central. It read how often each keyword converts, adjusted for where its clicks were shown, because a keyword that mostly appears at the top of search looks better than it is. And a keyword with only a few clicks didn't get a bid based on a few clicks' worth of luck. Its rate was blended with its family's: the same phrase on other match types, its tier, the product, the account.
There were two suggestions for every target, one click away on its bid:
- Standard, to coast: the most a click is worth if you only count the ad order it brings.
- Growth, to push: the same, plus the organic orders an ad order tends to bring with it, read from Halden's own share of organic sales.
Theo read it twice, then called Priya over.
On "whey isolate protein powder", Halden's best keyword, they were bidding $2.94. Counting only the ad order, a click was worth $2.62, so Priya was right: by her math, Halden was overpaying. Counting the organic orders that come with it, about five for every ten ad orders on the vanilla tub, a click was worth $5.19. Theo was right too. On this keyword, Halden could afford to pay far more.
"So the good keywords aren't the problem," Priya said.
"The good keywords are fine," Theo said. "It's everything around them." The junk in tier 4. The leaking exact terms. The duplicates bidding against each other. The leak wasn't where anyone had been looking.
Fluvi hadn't changed a single bid. The suggestions just sat there, with their workings shown, for Theo to use or ignore. Nothing changes until you do.
The second meeting
Theo brought four screenshots to the next Monday meeting. He didn't make a slide.
Priya read the tier chart first. "So we can stop the leak."
"We can stop this leak," Theo said. "This week. I'll negate the unrelated stuff, fix the duplicates, harvest the search terms, set the bids. And then next week, there'll be new search terms, and new bids that need moving, and new junk to negate. It doesn't stay fixed. That's the part that's been killing me."
Maren had been quiet. "So the question isn't how to fix it," she said. "It's how to stop fixing it."
That was the right question. Halden's account was cleaner by Wednesday: search terms harvested, duplicates gone, leaking exact terms fenced off. But a clean account isn't a smart one. Amazon's auctions move every day, and Theo's attention didn't scale.
What Halden needed wasn't a better spreadsheet, or another agency, or Theo staying up later. It needed something that would do what Theo would do, if Theo could watch every keyword, every day.
In Fluvi, that thing is called an agent. In Part 2, Theo draws his first one, and spends two nervous weeks watching what it would have done before he lets it touch a single bid.
Next: Halden, Part 2: The First Agent.
Want to see your own account the way Halden first saw theirs? Run the free audit. Nothing changes on Amazon until you say so.
About this story: Halden and its people are fictional, the same brand you'll find on our PPC product page, and the numbers are illustrative. The screens are the real Fluvi app, running on Halden's demo account. Every feature in it works the way it's described.
