Rethinking Amazon PPC — Segment-First, Hourly Bidding with Christoph Söhnlein

A wide gifts catalog that straddles multiple categories, price points, and margins sounds like a nightmare for Amazon Ads—unless you turn the whole account into a set of clean, measurable segments. In this episode, AMALYZE founder Christian Kelm interviews practitioner Christoph Söhnlein on how his team retooled PPC around per-target math, transparent logs, and seasonality-aware time windows, all without rebuilding their campaign structure.

Why Christoph abandoned portfolio-averaged PPC for segments

Söhnlein’s company sells personalized gifts across categories and events: Christmas, Valentine’s Day, baptisms, weddings, and more. That means heterogeneous margins and wildly different demand curves. Historically, he trialed “a lot of tools.” They weren’t all bad—but for his use case, they were slow or imprecise, and most optimized at the wrong level: aggregating ACoS at the portfolio or campaign layer.

His pivot was AMALYZE’s segment model. A segment is a goal-driven slice of the account—think “personalized men’s gifts” with its own target ACoS and time windows. Inside, every target (keyword, ASIN, category, auto match types) is optimized toward the segment goal individually. No averaging across 100 keywords to hit an aggregate ACoS while some targets are under-spent and others hemorrhage.

Christian frames it with a kitchen metaphor: a segment is the cake, but every crumb (each target) must taste like the goal. No salty piece balancing a sugary piece to a passable average.

Build on data from day zero: Ads API first, then tooling

Before using AMALYZE Advertising in earnest, Christoph connected the Amazon Ads API early at Christian’s insistence. That call aged well: it unlocked deep historical data the tool could immediately leverage for modeling. If you’re considering a switch, this is step one.

They also stress that the tool is not a black box. There’s no “AI”—just math. Every change is explainable and logged. In fact, AMALYZE is one of the heaviest global users of Amazon Marketing Stream, reacting to customer behavior in near-real time rather than fixed “Prime Day” calendars.

Onboarding that teaches a new mental model

The hard part isn’t clicking “create segment.” Christoph: that’s minutes. The hard part is thinking in segments. AMALYZE’s onboarding is hands-on (regular reviews with Christian/Michael), and the Discord channels (one-to-one plus a group Beta/Advertising channel) act as a living playbook. Christoph shares that even throwaway questions get 10–15 substantive replies from other practitioners.

He also highlights ongoing nudges from the AMALYZE team—pings about anomalies, reviews of logs, and proactive feedback. That matters when you’re unlearning habit-formed workflows like global bid bumps before a holiday.

Protocols and proof: every change, and why it happened

Söhnlein’s favorite feature is the protocol. It shows every adjustment with the reason. He tweaked two flights, checked an hour later, and saw 7,500 target-level bid changes applied. Moments later, Christian refreshed: the log already showed 7,931. In March alone, Christoph’s account saw 1,188,482 bid-evaluation checks.

Other tools truncated logs (some stopped at 100 lines), killing traceability. Here, changes are double-checked for application—no “we tried to set €1.60” when the console still runs €1.20. You can follow the chain from segment settings to per-target math to the final applied bid.

“I can finally see every change and why it happened. From one hour to the next, 7,500 bids moved—and I know exactly why.” — Christoph Söhnlein

Math over myth: bids follow ACoS ≈ CPC / (CVR × price)

Christian and Christoph keep returning to first principles. Expected ACoS is approximately CPC divided by conversion rate times price. If your CVR drops or your average cart value halves (e.g., customers shift from 2-pack to 1-pack at season end), your ACoS will spike at the same CPC. The tool recalculates hourly at the target level to reflect that reality.

This rigor can be sobering. Early on, Christoph complained that a beloved keyword was bid down so far it “wasn’t really participating.” The math held: given current CVR and price, a higher bid would bust the ACoS. You can fight that by extending time windows (to include delayed conversions) or adjusting goals, but you shouldn’t fight the math with wishful thinking.

Time windows as levers: attribution and lookback

Two often-misused controls become strategic when you’re segmenting properly:

• Attribution window (e.g., 30 days): For gifts and higher-consideration items, a lot of sales attribute after 14 days. Christoph sees accounts where 20–30% of revenue lands between day 15 and 30. Extending to 30 days lets you bid assertively without “buying data” blindly—because late revenue is in scope.
• Lookback window (e.g., 90/180/365 days): For slow or episodic products, longer lookbacks keep the model from swinging on sparse, noisy recent data. Christian notes Amazon’s console now shows 90-day per-target performance; AMALYZE can operate beyond that, up to a year if needed.

Pairing these windows with hourly recalculation means bids ramp into demand and decay out of it, using data that actually reflects how customers buy your products.

Seasonality managed, without manual whiplash

In the old world, Christoph mass-bumped bids before an event (e.g., +€1.00 across “Valentine’s” campaigns), then mass-cut them afterward (−€1.00). Collateral damage was inevitable: evergreen queries like “gift for dad” were over-cut and starved right when they should have stayed active.

With per-target math and time windows, CPCs climb into peaks and ease back out automatically. Christoph saw Valentine’s CPCs rise to around €1.40 by Feb 14, then glide down to ~€1.30/€1.20/€1.00 over the next week. In December, when Prime-by-Merchant delivery cutoffs flipped the PDP promise to “arrives after Christmas,” the protocol showed 6,000+ targets adjusted within a single hour (10–11 a.m.) to de-peak immediately.

Christian adds an anecdote: the system reacted a week before Black Friday/Cyber Monday when shoppers started spiking—not because a calendar said “Prime Day,” but because Amazon Marketing Stream showed customer behavior. Amazon even called post-Prime Day to ask about the strategies behind the growth; the answer was simply:

• price × CVR-driven math,
• stream-driven reactivity,
• and no funnel theatrics.

Structure-agnostic migration (keep your campaigns)

One blocker for many teams is structure debt—the fear they must rebuild years of campaigns to fit a new tool. Christoph worried about killing existing auto/harvesting “pipelines.” AMALYZE overlays any structure; there’s no forced naming conventions. You can start with a single segment, run simulations, and expand granularity later.

“Flights” layer in scheduling and pacing inside segments. Simulation guards against bad surprises by modeling the first up-to-1,000 target changes for your chosen attribution and lookback windows before you go live.

The community effect—and operating cadence

Christoph runs a one-person Advertising team. With segments set, he checks the tool two or three times a week. When major changes loom (new events or big assortment moves), he invests a few hours, often after an AMALYZE review nudges him to re-check assumptions.

On Discord, he regularly gets a dozen-plus thoughtful replies from other users. That peer review helps avoid local maxima and keeps the mental model sharp.

Root-cause detection beats “more bids”

Because the logs surface cause and effect, many “PPC problems” turn out to be merchandising or market problems:

• A top seasonal product cratered due to a wave of 1-star reviews—customers didn’t realize they had to peel off a protective film. The fix was better on-product labeling, not a bid tweak.
• Accounts drifted where bids and targets no longer matched today’s pricing or competition. In one review, conversion-rate assumptions were ~10% when reality was ~3.5%, and the market had been reshaped by brands launching official accessories. The right move: refresh PDPs, redo keyword research, and reset segments—not crank spend.
• “Data buying” with €5–€6 starting bids isn’t necessary. If you do start high, the hourly math will yank bids down quickly—AMALYZE even tested €999 bids and watched them normalize within a day. But better is to set sane initial bids informed by your ACoS math and windows.

Numbers that tell the story

• 7,500 target-level bid changes in one hour (across two flights) after a settings update; minutes later the log read 7,931.
• 1,188,482 bid-evaluation checks in March for Christoph’s account.

• 6,000 target adjustments between 10–11 a.m. on a December day when delivery cutoffs flipped PDP messaging.

• Valentine’s CPCs peaked near €1.40 and fell back within a week—no manual mass-edits.
• Holiday budgets scale 4–6×, with ACoS holding at goal.
• Christoph now spends 2–3 short sessions per week in the tool; the rest is handled by segments and hourly math.

Practical workflow tips from the conversation

• Connect the Ads API early—even before you switch tools. History is leverage.
• Start with one segment and a clear goal. Run the simulation before you go live.
• Choose attribution and lookback windows by product behavior (e.g., 30-day attribution for gifts; 180–365-day lookbacks for slow movers).
• Watch the protocol, not just ACoS. Fix PDP issues and assortments before turning knobs harder.
• Keep segmentation lean. Most users start overly granular, then consolidate to fewer, cleaner segments.
• Don’t fear structure debt. Keep your campaigns; let segments govern goals and bids.
• Expect CPCs to climb into peaks and fall out automatically. Resist global bid hammers.

Bottom line

Christian Kelm and Christoph Söhnlein’s talk is a candid field report on what serious Amazon Ads operations look like when you abandon portfolio averages and optimize at the target level inside segments. The stack is simple but demanding: early API data, hourly math, transparent logs, and thoughtful time windows. The payoff is big—stable ACoS through seasonal surges, automation that reacts to real shopper behavior, and a workflow lean enough for a one-person PPC team.

Christoph’s verdict: AMALYZE does exactly what he needs—no black box, no magic, just the control and visibility to steer a sprawling gifts catalog with confidence.