3.5x ROAS Improvement for a Multi-Million Home Goods Seller

THE AMAZON MARGIN SQUEEZE

Here is the reality of selling home goods on Amazon in 2026: You are competing against thousands of sellers offering identical imported products, and Amazon's referral, FBA, and storage fees consume **35% of your gross sales**.

The client was a high-volume home goods brand doing $4.8M annually. They had plenty of organic volume. But their paid marketing was hemorrhaging cash:

They were spending 22% of their total revenue on advertising just to keep their ranking. Their ACoS sat at a painful **91%**. They were buying sales, but making zero profit from them.

When they came to us, they wanted to target new keywords. They wanted to increase their PPC budget to $120k to “scale sales.”

I told them: Your ACoS target is a pipe dream. You are bidding on terms that Alexa doesn't care about. We need to restructure your catalog for COSMO, automate your bids, and optimize for TACoS, not ACoS.

PHASE 1: THE COSMO SEMANTIC ENRICHMENT

Amazon's search engine transitioned fully in early 2026. The classic keyword-matching engine is gone. It was replaced by **Alexa for Shopping**, powered by the **COSMO common sense knowledge graph**.

COSMO does not look at your keyword stuffing. It evaluates **intent, scenarios, and constraints**.

The client's listing for their signature “Heavy Duty Kitchen Utility Cart” was filled with: *“utility cart, rolling cart, kitchen island cart, organizer cart, storage cart.”*

But how do users search in 2026? They say to Alexa: *“Show me a kitchen cart that can fit in a 12-inch gap next to my fridge and hold a heavy KitchenAid mixer.”*

COSMO maps: - **Constraint**: *“12-inch gap”* (Width < 30cm) - **User Scenario**: *“hold a heavy KitchenAid mixer”* (Load capacity > 25 lbs, stable top surface)

We restructured their entire listing catalog to inject semantic nodes into their bullet points and description:

By adding explicit dimensions, load constraints, and user-scenarios, Alexa immediately indexed the utility cart as the primary match for these long-tail contextual queries.

PHASE 2: THE SP-API BIDDING AUTOMATION SCRIPT

Amazon ad campaigns bleed budget because of latency. Keywords that stop converting sit unchecked for days.

We built a Node.js automation script that connects directly to the Amazon Selling Partner API (SP-API) using secure OAuth.

// scripts/amazon-sp-api-bid-adjuster.js
import { SellingPartner } from 'amazon-sp-api-sdk';

const spClient = new SellingPartner({
  region: 'na',
  refresh_token: process.env.AMAZON_REFRESH_TOKEN,
  client_id: process.env.AMAZON_CLIENT_ID,
  client_secret: process.env.AMAZON_CLIENT_SECRET,
});

async function optimizeBids() {
  try {
    // 1. Get Campaign Report (Last 7 Days)
    const reportData = await spClient.callAPI({
      api: 'reports',
      operation: 'getReportData',
      path: { reportId: process.env.CAMPAIGN_REPORT_ID }
    });

    const keywords = JSON.parse(reportData);
    
    // 2. Adjust bids dynamically based on conversion thresholds
    for (const keyword of keywords) {
      const clicks = keyword.clicks;
      const orders = keyword.orders;
      const currentBid = keyword.bid;
      const acos = keyword.acos;

      // Rule A: High clicks, zero conversions -> Trim aggressively
      if (clicks >= 15 && orders === 0) {
        const newBid = Math.max(0.10, currentBid * 0.70); // Cut bid by 30%
        await updateKeywordBid(keyword.id, newBid);
      } 
      // Rule B: ACoS is exceeding target (e.g. > 45%) -> Trim moderately
      else if (acos > 0.45) {
        const newBid = Math.max(0.20, currentBid * 0.85); // Cut bid by 15%
        await updateKeywordBid(keyword.id, newBid);
      }
    }
  } catch (error) {
    console.error('SP-API Optimization failed:', error);
  }
}

async function updateKeywordBid(keywordId, targetBid) {
  return spClient.callAPI({
    api: 'advertising',
    operation: 'updateKeywords',
    body: [{ keywordId, bid: parseFloat(targetBid.toFixed(2)) }]
  });
}

The script ran as a cron task every 6 hours. It automatically evaluated ad spend data at the keyword level.

Instead of waiting for weekly manual campaign reviews, non-converting keywords were down-bid within 6 to 12 hours of crossing the conversion waste threshold.

PHASE 3: TARGETING TACoS & COHESIVE SCALE

We transitioned the client's goal from Sponsored ROAS to **TACoS (Total Advertising Cost of Sales)**.

Because Alexa for Shopping recommends products to users based on listing authority, PPC ads play an upstream discovery role. A buyer clicks an ad, closes the tab, and asks Alexa to purchase the product organically later.

Looking only at last-click ROAS makes you cut campaigns that are actively driving organic and voice search discovery.

Here is the outcome of the bid automation and semantic indexing:

By reducing paid ad spend and letting organic voice search scale, they recovered **$37,400 per month** in pure bottom-line margin.

That is the difference between vanity PPC scaling and margin optimization.