The journal
Notes from the team building the AI.
Long-form writing on Amazon PPC, autonomous bidding, EU multi-marketplace dynamics, and what a decision trace actually contains. Not a content marketing operation. Just what we are seeing on the accounts we read in simulation.
- 7 min readROAS
ROAS and ACoS are the same number upside down. Here is which to use.
ROAS is ad revenue ÷ spend; ACoS is its exact inverse — a 25% ACoS is a 4x ROAS, every time. So the choice is a translation, not a strategy. The real point is what both are blind to: your margin, and your total revenue.
Read the piece - 8 min readNegative keywords
Amazon negative keywords: the cheapest profit lever you are not pulling.
Brands that actively maintain negative keywords cut wasted ad spend by ~28% and lift ROAS ~22% (Tinuiti). The mechanism is not clever — it is weekly hygiene most accounts skip. Where the waste hides, the rule for negating without starving discovery, and why a flat spend threshold is the wrong line.
Read the piece - 8 min readACoS
What is a good ACoS on Amazon in 2026?
The 2026 average ACoS sits near 32%, and under ~28% beats the market — but quoting the average as a target is how sellers throttle growth or defend a losing campaign. The benchmark is a thermometer, not a goal. The numbers by stage, why break-even is the only ACoS that judges a campaign, and the margin that decides the line.
Read the piece - 8 min readPrime Day
Prime Day 2026 PPC: the pre-event checklist that protects margin.
The Prime Day mistake that costs most is not under-bidding — it is running bids hot into a stockout, or forgetting to wind spend down the day after. Inventory buffers, break-even ceilings, per-marketplace caps, and the ramp-down everyone skips.
Read the piece - 9 min readRufus
How Alexa for Shopping (Rufus, COSMO) is rewriting Amazon PPC.
Amazon retired standalone Rufus in May 2026 and folded it into Alexa for Shopping. With an AI assistant mediating 15–20% of mobile searches and reading intent over keywords, PPC shifts from ranking for a word to winning a conversation. What changes, and what does not.
Read the piece - 8 min readTransparency
Transparency, the first pillar.
Pacvue, Perpetua, Teikametrics — three good tools, three black boxes. Mirox treats explainability as the product, not the disclosure. What that looks like in practice on a single bid.
Read the piece - 8 min readSimulation Mode
Simulation Mode: preview the AI, then trust it.
Every other PPC tool on the market demands trust before evidence. We invert it. Watch Mirox decide on your real account for fourteen days. Nothing moves. Then you decide.
Read the piece - 10 min readSafety
Seven layers between an AI and your ad budget.
An autonomous PPC bidder without circuit breakers is a runaway training set with a Stripe key. Here is the safety architecture we built before we let Mirox bid live, and the eighth gate we added when sellers started losing inventory.
Read the piece - 10 min readIntelligence
Sixteen agents, one objective: your profit.
Most "AI" PPC tools are a single optimisation engine wearing a wig. Mirox is sixteen specialised agents — Tactician, Strategist, Sentinel, Semantic — coordinating on a blackboard. Here is what each one does and why the architecture matters.
Read the piece - 9 min readPricing
Why we will never charge a percentage of your ad spend.
Pacvue charges 3% of spend. Teikametrics 3% over the first ten thousand. Agencies, anywhere between 10 and 20 percent. The structure means everyone in your ad pipeline is paid more when you spend more. Mirox is paid the same.
Read the piece - 8 min readEU PPC
EU-native is not a translation layer.
Every "EU-friendly" Amazon PPC tool was built for the US and shipped to Europe with a localisation pack. The +28% ACoS hit on UK→DE copy comes from that assumption. Mirox was built the other way around.
Read the piece - 11 min readPillars
What autonomous PPC intelligence actually means.
Transparency, Simulation Mode, Safety, Intelligence, Aligned Pricing, EU-native. Five pillars, one company. Why none of them works without the others, and why the combination is the category we created rather than the category we entered.
Read the piece - 8 min readEU PPC
Why German Amazon campaigns underperform UK ones — and what the data shows.
UK→DE copy without localisation costs sellers a documented +28% ACoS and −42% CTR. The fix is not translation. It is per-marketplace bidding architecture.
Read the piece - 9 min readACoS
Where 30 to 40 percent of your Amazon ad spend actually goes.
Sellerqi 2026: the average Amazon seller wastes between 30 and 40 percent of their PPC budget. We map where the leak is, why dashboards miss it, and what a decision trace surfaces that a CSV cannot.
Read the piece - 10 min readTransparency
Why we built a PPC AI that explains itself.
Black-box optimisation worked when the question was "did it beat the baseline." It does not work when the question is "why did the bid change last Tuesday." Here is what a decision trace actually contains, and why a tool without one cannot be defended to a CFO.
Read the piece - 7 min readFounding Beta Program
We are looking for ten Founding Members.
Ten EU Amazon sellers, hand-picked. Free Simulation Mode on your real account, then go live when you are ready — free for the whole beta, no card. The terms inside.
Read the piece - 9 min readMulti-marketplace
Expanding between the US and EU on Amazon: the cross-marketplace PPC playbook.
The costliest expansion mistake is cloning campaigns across marketplaces and assuming the bid math travels. It does not — UK→DE copy carries a +28% ACoS penalty. The playbook in both directions: US into the EU, EU into the US, Canada and Mexico included.
Read the piece - 8 min readSponsored Brands
Sponsored Brands in 2026: bid for new-to-brand, not ACoS.
Judging Sponsored Brands on ACoS throttles the exact spend that builds your brand. Its job is acquiring new-to-brand customers and growing branded search — outcomes ACoS cannot see. The metric set that actually fits the ad type, including the Rufus shift.
Read the piece - 9 min readBudget
How much should you spend on Amazon Ads in 2026?
There is no fixed "right" budget — the correct spend is whatever keeps TACoS in target while every marginal click clears break-even. How to size it in 2026 as CPCs rise, and why a percentage-of-spend tool fee quietly pushes the number up.
Read the piece - 8 min readInventory
Your best-selling ASIN is your biggest PPC liability.
A bidder that ignores inventory keeps paying for first page right up to the stockout — then bills you to re-earn the rank you lost. Trellis puts the average stockout cost near €18,000. Inventory-aware bidding is the fix, and almost no general-purpose tool does it.
Read the piece - 8 min readAmazon Ads Agent
Amazon launched Ads Agent. Here is what it does and does not do.
Amazon’s native AI campaign assistant saves 30–40% of management time and improves ACoS 12–18%. It is genuinely useful — and it is not profit-aware, not inventory-aware, and structurally motivated to grow your spend. The honest line between a campaign assistant and a profit model.
Read the piece - 8 min readCPC
Amazon CPC benchmarks 2026: by category, ad type, and marketplace.
The average Sponsored Products CPC in 2026 is about $1.22, with competitive terms past $3.00. But the headline average is the least useful number here. The benchmarks by category, why the US figure misleads EU sellers, and the only CPC that matters: your break-even.
Read the piece - 9 min readAI PPC
How to evaluate an AI Amazon PPC tool: a nine-point checklist.
The label "AI" has stopped meaning anything in Amazon PPC. Nine questions — starting with "can it explain a single bid?" — separate a real profit model from an autobidder wearing a wig. A buyer’s checklist before you sign.
Read the piece - 8 min readTACoS
ACoS tells you efficiency. TACoS tells you the truth.
ACoS measures one campaign. TACoS measures whether advertising is growing the business or renting it. A 18% ACoS can hide a falling contribution margin. Here is the difference, the 2026 benchmarks, and how to bid on profit instead of a proxy.
Read the piece