Halden, Part 3: Momentum
Budgets running dry by 2 p.m., a rival's launch knocking Halden off page one, and a container stuck at port. Halden stops defending and starts pushing. Part three of a four-part story.
A Fluvi story in four parts. This is part three; start with part one.
Dark by two o'clock
The budget problem looked simple. halden-vanilla.exact.t1 had an $80 daily budget. On three of the last seven days it had spent all of it by early afternoon and switched itself off until midnight. Raise the budget, done.
Theo almost did exactly that. Then he looked at where the money was going, because for the first time he could see it.
Amazon shows a Sponsored Products ad in three places: top of search (the first row of results), rest of search (further down the page) and product pages (the "products related to this item" rows). Each campaign has one base bid per keyword, and you can raise it by a percentage at the top of search. Halden had never touched that. The agency's notes said "leave at default".
The campaign's placement numbers said otherwise. Almost half its money went to product pages, where it sold at 44% ACoS. By the time the after-work shoppers arrived, there was nothing left for the top of search, where the same keywords sold at 19%. The budget wasn't too small. It was being spent in the wrong place.
The chocolate 1kg's exact campaign had the opposite problem. Its top of search was fought over by Ironpeak and running at 46% ACoS.
Theo could fix both by hand. He'd have to check them again next week, and the week after, on every campaign. He knew how that story ended.
The campaign agent
Fluvi's third kind of agent manages campaigns rather than keywords: placement adjustments, audience boosts and daily budgets. Theo drew it the way he'd explain it to a new hire:
Three forks, on every Products campaign. It ran in simulation for a week. Then it went live.
The vanilla exact campaign's top of search was selling at 19%, so the agent set its adjustment to +50%. The campaign kept running dry, so the agent raised its budget 20%, then 20% again: $80, $96, $115. It stopped going dark at 2 p.m.
The chocolate campaign's top of search went to 0%. Its spend shifted to rest of search, where Ironpeak wasn't fighting, and its ACoS dropped from 46% to 29% in three weeks.
A month later, the vanilla campaign's placements looked like this:
Nobody at Halden decided those numbers. Theo decided the rules, and the numbers followed.
Six in the morning, seven at night
Theo had always suspected something about Halden's customers. They're gym people, like him. They don't shop for protein powder at 2 a.m. They shop at 6 a.m. in the car park before a session, and at 7 p.m. after dinner when they notice the tub is nearly empty.
Fluvi's dayparting agent let him act on it. He could start from Fluvi's default schedule for the US marketplace, built from two years of research on Amazon shoppers: lower at night, higher in the morning. Instead he looked at Halden's own sales by hour and painted his own week.
Weekday mornings, +20% at 6 and +30% at 7. Weekday evenings, +30% from 7 to 9. Overnight, 30% less, easing down from 10 p.m. Weekends have no peaks at all. Theo's athletes, it turned out, sleep in on Saturdays and don't buy protein on the way to brunch.
The bids shift on the hour, every hour, on every campaign the agent covers. They move from the base bid the other agents set, so the agents don't fight each other. The bid agent decides what a keyword is worth. The dayparting agent decides when it's worth more.
Within a month, Halden's overnight spend was down by a third with almost no lost orders. The money went to 6 a.m., where it bought more.
Ironpeak's launch
Now the problem from the end of Part 2.
Ironpeak launched a new 5 lb tub in the second week of March, priced under Halden's vanilla tub and backed by what looked like an enormous launch budget. Within four days, Halden's daily rank check showed the vanilla tub sliding on "whey isolate protein powder", its best keyword, from 3rd to off page one.
In the old world, Theo would have noticed this in a week or two, as a dip in sales he couldn't explain. In the new one, the bid agent saw it the morning it happened. One of its forks was built for exactly this: buried in organic? Push. It raised the bid 15% the next morning, and again two days later, until paid clicks did what they're for: more sales on that search, which is what moves organic rank. Once the keyword was back on page one, the fork stopped firing and the everything-else fork took over again, aiming at 25% ACoS like every other keyword.
Theo found out about the whole thing afterwards, reading the log on a Sunday. He'd been coaching that week. Every keyword's page shows its bid and its organic rank on one chart, so the story is there to read:
Who else is in the room
Ironpeak's launch taught Halden something else. Defending is fine, but the searches you already know about are only part of the market.
Fluvi's Competitors view finds the products that keep showing up next to yours in Amazon's results, and scores how much of your search territory they share. For the vanilla tub there were three that mattered:
Northfield Labs, the closest match. Ironpeak, the loudest. And Pulse Nutrition, a lean whey for women, sharing 23 search terms with Halden's vanilla tub. That was the same signal the harvesting agent had found in Part 2 with "protein shake women", arriving from a completely different direction. Two features, reading two kinds of data, both pointing at a customer Halden had never marketed to. Maren had the new main image, with a woman training in it, live by the end of the month.
Hover over a competitor's shared terms, and you see every search you both fight over, with their rank and yours. Ironpeak's list had a line Theo read three times:
On "halden whey isolate", a search for Halden by name, Ironpeak ranked first. That was the conquest Theo had guessed at in Part 2, now in black and white, and the reason the brand fork sat at the top of his bid agent.
The other tab was the one that grew the business: the searches Ironpeak ranks for where Halden never shows at all.
"lactose free whey". "vanilla protein powder". "whey isolate 5lb". Thousands of searches a month, all of them shoppers looking for a tub a lot like Halden's, and Halden with no ad and no organic rank on any of them. Theo added them as keywords. The classifier put them in tier 1 and 2, so the bid agent picked them up the next morning. No new rules needed. That's the part Theo liked most: every new keyword fell into rules that already existed.
The container
The last test of Part 3 had nothing to do with advertising.
In mid-April, a container of whey isolate sat at the Port of Long Beach for eleven days longer than planned. Halden's co-packer couldn't fill the vanilla 2kg order without it. Inventory on Amazon dropped to 9 days of cover, then 6.
Running ads on a product that's about to go out of stock is one of the most expensive mistakes in Amazon advertising. You pay to send shoppers to a listing that will show "currently unavailable" next week. Worse, if you sell out with ads running hot, your organic rank drops while you're out, and you pay again to win it back.
Maren found out on her phone, in an airport, between flights to a trade show. She didn't open a laptop. Halden's Fluvi account was connected to her AI assistant over MCP, so she typed one sentence:
Six campaigns paused on Amazon before her flight boarded. Ten days later, when the stock checked in at the warehouse, one more sentence turned them back on. Every agent picked up where it left off.
Halden didn't sell out. The vanilla tub's organic rank barely moved. Priya put the ads they didn't run on a nearly empty listing at about $1,900.
Where Halden stood
By the end of April, three months after Priya's slide, the account looked like it belonged to a different company. Most of the Sponsored Products money now went to the top of search, the placement that sells, and product pages, once the biggest drain, were down to a sixth of it.
ACoS on Amazon US was 27%, at break-even before counting a single organic sale. Ad sales were up a third on January's pace. Tier 4 spend was under 4% of the budget. Theo was spending about six hours a week on ads, down from twenty-two, and most of that was reading logs and deciding what to try next.
Halden had stopped bleeding. It had found new customers. It had defended its best search against the biggest brand in the category, and won.
What it hadn't done yet was grow on purpose. In June, Halden would launch its first new product in two years. Six weeks after that came Prime Day.
In Part 4, Halden goes for it.
Previous: Part 2: The First Agent. Next: Part 4: The Win.
See which products sit beside yours, and the searches they win that you've never bid on. Connect your account for free.
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.
