Finding New Keywords for Amazon PPC: A Repeatable Process for Sellers and Vendors

Traditional "keyword harvesting" in Amazon Advertising is effectively dead. While many sellers still rely on standard search term reports to periodically migrate performance winners from auto to manual campaigns, this reactive approach creates data silos, wastes budget, and fails to handle the granular complexities of modern Amazon PPC. To achieve true scalability, advertisers must transition to a recommendations-based system that prioritizes immediate controllability and systematic source exhaustion.

Why Keyword Harvesting is Obsolete

The classic harvesting model follows a linear, slow path: a search term appears in an Auto campaign, it eventually hits a threshold of "success" (like five sales), and the advertiser then moves it to a Broad match campaign. Weeks later, it might graduate to Phrase, and eventually to Exact.

This process is fundamentally flawed because it ignores the nature of the data collected. When a customer converts on a specific search term in an Auto campaign, you have just acquired "Exact" data for that specific target. Waiting weeks to move that term through a Broad-to-Exact funnel is a waste of time and money. Furthermore, standard harvesting often focuses only on "winners," leaving underperforming search terms to continue draining the budget within the Auto campaign rather than isolating and correcting them.

The modern goal should not just be finding new words, but "drying out the sources." This means identifying every relevant target within an Auto or Broad campaign and moving it into a controllable unit where the bid can be managed precisely according to its specific performance.

The Logic of Simultaneous Transfer

A sophisticated PPC strategy bypasses the traditional funnel in favor of simultaneous transfer. When a relevant target is identified with sufficient data, it should be moved into Broad, Phrase, and Exact match types at the same time.

The reasoning is rooted in data precision. If the search term "blue pen" converts, we know its exact performance. By placing it immediately into an Exact match ad group, we gain 100% control over the bid for that specific term. By placing it simultaneously in Broad and Phrase (with corresponding negative match types to prevent cannibalization), we allow the system to continue exploring variations (like "blue pens for students") without losing the efficiency of the original discovered term.

The Problem of Rule-Based Rigidity

Many tools use arbitrary rules: "If clicks > 10 and Sales > 1, then move keyword." This logic fails to account for the actual Target ACOS (Advertising Cost of Sales). A keyword with only two sales and a 5% ACOS is a massive win that should be pushed immediately, yet many harvesting rules would ignore it. Conversely, a keyword with 20 sales but an 80% ACOS needs to be isolated just as quickly—not to be deleted, but to be bid-adjusted so it can perform at the desired target.

The Mathematical Engine: Bid Calculation

Finding a keyword is only half the battle; the other half is setting the right bid from the first second it enters a manual campaign. Professional Amazon advertisers should rely on the standard bidding formula to ensure every new recommendation starts with a "mathematically correct" bid:

Bid = Target ACOS × Conversion Rate × Selling Price

By using data from the source campaign (e.g., the Auto campaign), you can calculate exactly what the bid should be to reach your target ACOS in the new manual campaign. This removes the guesswork and the "learning phase" that typically burns through the budget when launching new targets. With the integration of Amazon Marketing Stream, this data can be further refined by time-of-day performance, ensuring that recommendations aren't just based on what works, but when it works.

Structural Integrity and "Clean" Campaigns

A recommendation engine is only as good as the campaign structure it feeds. To utilize a high-frequency keyword discovery process, the account must be organized into logical, manageable units.

Match Type Separation: Mixing Broad, Phrase, and Exact match types within a single ad group is a recipe for failure. It makes negative keyword management impossible and muddies the data.
Price Point Consistency: Never mix products with significantly different price points in the same ad group. Because the bid is a function of the price, a 10 € product and a 50 € product require different bids even for the same keyword. Mixing them makes the "average bid" ineffective for both.
Source-Target Alignment: Before transferring a keyword, the system must verify that the SKUs in the source campaign (e.g., Auto) match the SKUs in the target manual campaign. If you learn performance data from an FBA SKU and transfer that keyword/bid to an FBM SKU with a lower conversion rate, the bid will be too high, leading to an inflated ACOS.

The Multi-Directional Learning Loop

Keyword discovery shouldn't only flow from "Auto" to "Manual." A sophisticated system employs a cross-diagonal logic where every campaign type learns from every other type:

Forward Learning: Auto campaigns find new search terms for Broad, Phrase, and Exact.
Backward Learning: An Exact match campaign converts on a misspelled version of a keyword or a plural. This variation should be fed back into the Broad and Phrase structures as a new target, with appropriate negatives applied to keep the Exact campaign "pure."
Self-Enrichment: A Broad match campaign identifies a specific long-tail phrase that is performing exceptionally well. The system should recommend this as its own isolated target within the same structure to allow for unique bit control.

This "hygiene" aspect of keyword management ensures that the account doesn't just grow in volume, but in structural clarity. It prevents "duplicate" targets—where the same keyword is active in 20 different ad groups—which scatters data and makes it impossible to reach statistical significance on any single data string.

Scaling in International Markets

The greatest leverage for a recommendation-based process is found in international expansion (pan-EU or North America). Most sellers struggle with keyword research in languages they do not speak.

By relying on a data-driven recommendation system, the language barrier is removed. You do not need to speak French or Italian to know that a specific search term has a 12% conversion rate and a 4% ACOS. The system identifies the relevance based on customer behavior (data) rather than linguistic knowledge. It kemudian transfers these terms and applies the correct bid immediately. This allows for a "brutal advantage" in foreign markets, as you can scale advertising as quickly as the data permits, without waiting for manual translations or market research.

Practical Takeaways for PPC Management

To implement a repeatable process for keyword discovery, prioritize the following technical adjustments:

Exhaust the Auto Campaigns: Treat Auto campaigns as research environments. Your goal is to identify every search term that generates a click and move it to a manual structure where it can be bid-managed.
Implement "Negative Hygiene": Whenever a keyword is moved to an Exact match group, ensure it is added as a "Negative Exact" in the source Auto or Broad campaign. This "dries out" the source and forces Amazon to find new search terms.
Avoid "Negative Kill": Do not simply set underperforming keywords to negative. Transition them to an Exact match group and lower the bid using the ACOS/CR/Price formula. This allows you to capture "cheap" sales that would otherwise be lost.
Limit Recommendation Batches: Processing too many keywords at once can be overwhelming. Focus on batches of 500 recommendations per cycle to maintain manual oversight and ensure structural integrity.
Audit for Duplicates: Regularly check if the same target is being bid on in multiple ad groups. Consolidate these into the highest-performing data string to improve the speed of your optimizations.

The True Lever: Relevance Over Content

Many advertisers spend thousands on improving content (images, A+ content) in hopes of a 5% increase in conversion rate. While important, the real lever is finding more relevant targets where you can achieve your Target ACOS.

If you have 100 keywords performing at your target, your sales are capped by the search volume of those 100 words. If you use a recommendation system to find 1,000 relevant targets—even if some have low search volume—the cumulative effect on your "flat" sales growth is much higher than a minor tweak to a product image. In a competitive landscape, the seller with the most granular and well-managed keyword list usually wins.

Bottom line

Effective keyword discovery is a continuous cycle of drying out automated sources and transitioning data into controllable manual units. By applying mathematical bid formulas and maintaining strict structural hygiene, advertisers can scale their reach without sacrificing profitability. Moving beyond manual "harvesting" to a data-fed recommendation system is the only way to maintain a competitive edge on Amazon today.