Ditching 'Hockey-Stick ACOS' for Per-Target Control — with Denis M. Klug

Christian Kelm brings back Real Talk by skipping the fluff and pressing on the only question that matters: what did you actually change in your Amazon PPC, and what moved the numbers? Denis M. Klug—seller since 2014, Amazon since 2016, founder of the Viking‑themed apparel brand North Legendary—lays out his workflow after switching his ads to AMALYZE. It’s a candid tour from wasteful averages to per‑target control, seasonal timing, and unit‑economics discipline.

Denis in 10 years: seller to SEO operator

• 2014: starts selling online; 2016: launches on Amazon with print‑on‑demand tees; evolves into North Legendary (textiles, Viking niche). First ASIN went live in 2017 and still sells.
• Today: Denis oversees the brand (no longer packing boxes), and runs an agency focused on Amazon SEO. He’s tested “a lot” of ad tools and agency setups in the past decade.
• Candid stat: by his own audit he’s burned ~€60,000 in ads through “optimizations” and tools that didn’t deliver.

Why a “structureless” PPC intake worked here

Christian’s first point: most ad tools require you to rebuild to their template. Denis: AMALYZE was the first to say “we’ll take your account as it is.” No forced campaign taxonomy. Segments can cover an entire account if you want. That alone removed weeks of migration risk.

Under the hood, Christian explains why this can work: the system doesn’t aim for an average ACOS across a bundle; it bids every target (keyword/ASIN/category) independently to the target ACOS you set. That’s what makes rigid, prescriptive structures unnecessary.

“It blew my mind that every single keyword can have its own target and price—it’s not one average anymore.” —Denis

Kill the hockey stick: per‑target ACOS, not averaged

Christian’s diagnosis of most accounts: the infamous “hockey‑stick ACOS.” Averages hide over‑ and under‑performers; you hit 20% on paper while paying 120% on generic money‑pits and 1% on brand terms. AMALYZE steers bids at the target level to a specific ACOS goal (not just a portfolio average), so each target earns or loses budget on its own merits.

Christian’s quick self‑audit if you want to see your hockey stick today:

• Export your Search Term Report.
• Plot CPC vs. performance (even F11 auto‑chart on Windows) or use a Pivot chart.
• You’ll see a flat zone (no sales = infinite ACOS = wasted spend) and a long tail—the stick.

Match attribution windows to buying behavior

Denis’ product mix spans gifts and textiles. He set 14‑day attribution after observing carts that linger before checkout. That single change helped the system credit delayed sales and bid more rationally.

Christian’s rule of thumb:

• 1 day for immediate‑need products or micro seasons (Valentine’s Day itself, flash promos).
• 7–14 days for typical consumer goods and gifts.
• Up to 30 days for higher‑consideration items (think robot vacuums) or if your customers often wait for payday.

Changing attribution window can materially shift apparent ACOS without touching bids. One case: a client stuck at 20% target wanted 15%. Switching to 30‑day attribution (data supported that >20% of sales landed after day 14) instantly reflected 15% ACOS because the sales had always been there—just outside the prior window.

Use the Betrachtungszeitraum as a control surface

Beyond attribution, Denis uses the Betrachtungszeitraum (the analysis lookback window) as an active lever:

• Static window to lock in a historic period (e.g., last year’s Easter week) and pre‑seed bids a month in advance.
• Dynamic window that grows from a chosen start date as you near an event (e.g., sliding toward 21 December, the last relevant ship date).

Workflow Christian recommends for movable feasts like Easter:

1. In March, set bids from last year’s mid‑April data (static). Let it run.
2. Right after Easter, switch the window back to “from now” (static) for a hard reset.
3. After a week, revert to a dynamic window so the system learns on fresh signals.

Simulate first, then apply—and verify hourly

Before pushing changes, Denis previews with AMALYZE’s simulation. The system shows projected sales, ACOS, and spend deltas for the chosen segment. After execution, an hourly protocol (powered by Amazon Marketing Stream) logs each target‑level change with the “why.” A full history timeline lets you correlate performance humps with the exact settings you changed (e.g., “we switched attribution from 14 to 30 days on this date”).

Denis’ reaction to the first all‑in simulation: disbelief—followed by the numbers matching reality.

Expect ACOS bumps when price and CR move

Christian’s math mantra: ACOS ≈ Bid / (Conversion Rate × Price). Two implications:

• If you change price or conversion (A+ Premium, images, coupons), ACOS will shift even if bids don’t.
• Around holidays, deliveries pause and purchases delay. Expect a temporary ACOS spike and a CR dip; trailing sales later pull ACOS back in line. Judge the system by the actions it takes (bids vs. goals), not by a single‑day ACOS screenshot.

Denis cites Spring Deals: price cuts drove a short‑lived ACOS uptick, then the module steered back to target as delayed sales closed.

Unit economics per keyword—and turn off Amazon’s dynamic bidding

One painful but profitable exercise Denis ran mid‑migration: compute the maximum CPC per product/keyword that still leaves margin, given actual conversion. He stopped letting Amazon’s suggested bids dictate spend and disabled Amazon’s dynamic “raise/lower” so the precision bids from the module weren’t overridden.

He also did a targeted reset:

• Paused the worst campaigns (some north of 100% ACOS).
• Rebuilt a few with sane starting bids derived from margin math.
• Didn’t hand‑price every keyword—too many terms—but used averages where necessary while the automation converged.

Result: ACOS fell faster, stability improved, and he didn’t need to micromanage daily.

What not to ignore: Buy Box, dead targets, 0.33% CR

Christian’s war stories that saved Denis time:

• Buy Box mismatch: advertising FBA while FBM holds the Box (or vice‑versa) makes ads look broken; check Box ownership before blaming bids.
• Zero‑volume targets: the module can “set without data” at segment level (small, cautious steps), but if a term has no traffic, bids can’t create demand. Confirm search volume and relevance first.
• Brutal CR example: at 0.33% conversion you need ~300 clicks per sale—no CPC will make that math work. Cut or fix the listing.

TACOS: reporting north star, terrible control loop

Denis doesn’t operate to TACOS day‑to‑day. Christian’s stance:

• Use TACOS monthly/quarterly to judge your organic/paid mix (50/50 suggests you can likely double ad spend and still hold the same mix).
• Don’t try to bid to TACOS at the target level. You don’t have per‑target organic sales in the Ads console; mixing in coupons, external traffic, and post‑click discounts corrupts the control loop. It recreates the hockey stick.
• Attribution straddling month‑ends further skews TACOS timing: costs hit this month, sales land next month in Seller Central but are credited back to the ad click date.

Results: fewer hours, better margins—not just “more revenue”

• Time: Denis went from roughly two workdays a week doing bid math in Excel/Python to about 3–4 hours a week reviewing logs and simulations. He now opens AMALYZE instead of Seller Central and checks in Mondays and Fridays.
• ACOS: portfolio average trended from ~40% toward sub‑20%. Some segments already hit targets; others are converging.
• Profit over revenue: Denis didn’t chase top‑line at any cost. He accepted lower volume on generic sinkholes (e.g., “T‑Shirt Herren” previously at 120–140% ACOS) to protect margin, while still pushing when launch velocity or reviews were strategically needed.

Christian underlines a common misconception: you can’t demand 5–10% ACOS and expect the same absolute sales as at 30–40%. Market competition sets a floor; if competitors won’t follow your bid hikes, volume won’t magically appear.

Market reality check: CPC inflation and irrational bidders

In apparel, Denis’ retrospective shows CPCs up ~80% vs. pre‑COVID. Many competitors still “can’t do the math,” bidding beyond product margin—so expect auctions to be noisy. On the flip side, some sellers quietly have their best years because they built resilient processes: clean data, tight content, stable supply, consistent bidding—rather than chasing a mythical perfect campaign structure.

Roadmap: harvesting and recommendations without the spreadsheet pain

Denis still runs “data‑gathering” auto campaigns and a monthly script to extract new terms and ASINs. Christian previews what’s next so he can retire that script:

• Recommendations from search term data and live performance.
• One‑click actions: add to the right campaign, pre‑calculated bid, or add as negative exact.
• Full de‑duplication and classification: detect exact/phrase/broad dupes, check if a candidate already exists elsewhere, and allow cross‑type promotions (e.g., graduate from Exact → Phrase/BR where appropriate).

The idea is simple: move every relevant target out of auto and into explicit control—priced correctly for its own ACOS goal, not merely the “good” ones that already show sales.

Onboarding, support, and how decisions actually get made

• Onboarding took Denis ~2.5–3 hours with Michi, recorded for replay. The first 5 minutes set up a segment that covers the whole account; the rest aligned on goals, attribution, windows, and language (what “push” means: launch? rank? volume?).
• Post‑onboarding: a check‑in two weeks later, another after a month. Discord for day‑to‑day questions. Group sessions twice a month to share patterns (e.g., Easter timing, Spring Deals).
• One actionable fix from onboarding: disable Amazon “raise/lower” dynamic bidding so the module’s per‑target math isn’t undone.

What changed in Denis’ weekly workflow

• Review simulations before major switches (attribution/window/targets).
• Scan hourly logs and the history timeline for anomalies after price or promo changes.
• Quarterly: sanity‑check TACOS and the organic/paid split, not to steer bids but to sense overall health.
• Keep auto campaigns only as harvesters until the recommendations pipeline absorbs their output.

Bottom line

Per‑target control beats pretty structures. Denis stopped averaging away the truth, matched attribution to how his customers actually buy, pre‑seeded seasons with the right lookback data, and grounded every bid in margin and conversion. He traded two days of spreadsheeting for a few focused hours a week—and swapped a 40% average ACOS for a path toward sub‑20% without chasing vanity revenue.

Quote it how Denis did: “We didn’t grow revenue; we grew profit.” If that’s your goal, stop arguing about the perfect campaign template. Get ruthless about targets, windows, and unit economics—and let the logs and simulations do their job.