How agencies run an Amazon PPC audit: read-only access by invitation link, a nine-point checklist with a Claude prompt for each step, and the proposal.
An Amazon PPC audit is usually the first piece of work an agency does for a new account, and often the piece that wins it. Done well, it puts a dollar figure on what the current setup wastes, shows the prospect why, and turns straight into a scope of work. Done badly, it is forty screenshots from the Ads console and a slide that says "optimize bids".
This guide is the version we would run as an agency today: read-only access through an invitation link, a nine-point checklist in which every step is a question you ask Claude over the account's DataDoe data, and a short method for turning the findings into a proposal. The same steps work as an Amazon account audit for a client you already manage, at the start of a quarter or after a bad month.
Many audits stall before they start, because the prospect does not want to hand over a login and the agency does not want to hold one. Amazon's own authorization screens solve that, and DataDoe gives you two ways to use them.
In DataDoe, open Sellers & Vendors and choose Invite External Seller or Vendor. Name the link after the prospect, pick Seller or Vendor, set the access level, and optionally drop the account straight into a Group for that client. The prospect opens the link, picks the marketplace, approves DataDoe on Amazon's consent screen and can add Amazon Ads in the same flow. They never need a DataDoe login. The link works once and expires after 7 days. If you would rather generate links from your CRM or an onboarding form, the same invitation is available through the DataDoe API and MCP.
Choose Read only for a prospect. The access level is fixed when you create the link, and an audit only needs to read.
Some prospects only want to show you their advertising. If they have added your team as a user in their Amazon Ads account, connect it from your side with Connect Amazon Ads only, choose the marketplace and pick their Ads profile. That covers campaigns, search terms, placements and targeting. What it cannot give you is anything that lives in Seller Central or Vendor Central: total sales, and with them TACoS, plus product costs and profit. Ask for the full connection if the audit should judge the ads against the business and not only against themselves.
History loads in two phases. You can work with an account once the first phase is done, but some accounts take more than 48 hours to finish, so book the findings call a few days after the connection, not the same afternoon. In Settings > Data, check that the Ads tables this checklist relies on are switched on: search terms, keyword targeting, placements and negative keywords. Then fix the window. The last eight complete weeks is a sound default, minus the most recent few days, which Amazon still revises.
Each step below says what you are looking for and gives a prompt to paste into Claude with the DataDoe connector added. Run them in order. The first four usually find most of the money.
Start with what shoppers actually typed. The search term data shows each customer query, the keyword or target it matched, and what it cost. Two kinds of waste matter: queries with clicks and spend but no orders, and queries that convert at an ACoS the product cannot carry. The first kind is mostly a long tail of small amounts, so the analysis has to read every row, not just the 50 biggest spenders.
"For the last eight complete weeks, list Sponsored Products and Sponsored Brands search terms with at least 10 clicks and zero orders. Group them by campaign, sum the spend, and turn the total into a monthly figure. Then list converting terms whose ACoS is more than double the account average."
Our guide to the Amazon search terms report explains each column if the prospect wants to follow along. The DataDoe Skills library also includes a PPC Wasted Spend Watchdog that runs this check as one task.
An ACoS figure means little until it sits next to margin. When the prospect connects Seller Central and uploads product costs, DataDoe estimates a break-even ACoS for each search term: the point at which ad spend eats the whole contribution of the sale. Then step back to TACoS, ad spend over total sales, to see whether advertising grows the account or mostly buys orders that would have come anyway.
"Show account-level ACoS and TACoS by week for the audit window. Then list the search terms where ACoS sits above the estimated break-even ACoS, with the spend behind each."
If the prospect's finance lead needs the reasoning, ACoS vs TACoS sets it out.
Look at where spend sits by match type: broad, phrase, exact, and the targeting expressions used by auto campaigns and product targets. An account that puts most of its budget into broad and auto, with proven queries never promoted to exact, pays discovery prices for traffic it already knows converts.
"Split spend, sales and ACoS by match type. Then find search terms with three or more orders inside auto or broad campaigns that are not an exact keyword anywhere in the account."
DataDoe keeps weekly snapshots of negative keywords. Hold them against the waste from step 1: a query that cost money for eight weeks without a single order and is still not negated is the easiest win in the whole deck. Also look for negatives that block queries converting in another campaign.
"Which zero-order search terms from the first check are not a negative keyword in their campaign or ad group? And do any negative keywords block search terms that have orders elsewhere in the account?"
Sponsored Products and Sponsored Brands report results by placement: top of search, rest of search and product pages, with an Amazon Business breakdown for Sponsored Products. Placement adjustments tend to be set once at launch and never revisited, so compare conversion rate and ACoS by placement campaign by campaign.
"For every campaign with more than $500 of spend in the window, show spend, CPC, conversion rate and ACoS by placement. Flag campaigns where top of search converts at least twice as well as product pages yet gets under a third of the spend."
A campaign that runs dry at lunchtime misses its evening shoppers, and a tidy ACoS can hide the fact that it was effectively switched off for half of every day. Campaign data carries each campaign's daily budget, so you can count the days on which spend reached it.
"List campaigns where daily spend reached 95% or more of the daily budget on at least a quarter of the days in the window, with their ACoS. Which of them are profitable enough to earn a bigger budget?"
The budget repeats on every row of a campaign, so tell Claude not to add it up across days or placements.
Hour-by-hour performance comes from Amazon Marketing Stream. In DataDoe those tables are off by default and only start collecting once you switch them on, so a prospect rarely has hourly history on the first day. Write it down as a follow-up: turn them on together with the connection, and after a few weeks there is enough data to decide on scheduled bid increases. Our dayparting guide covers that analysis and the rules Amazon now applies itself.
Split spend and sales by ad type and ask whether the mix suits the brand. A registered brand that spends nothing on Sponsored Brands is absent from the top of the results page, and an account where Sponsored Display takes a fifth of the budget with no stated goal deserves a question. One detail matters when you compare them: Sponsored Products credit sales for 7 days after the click, while Sponsored Brands and Sponsored Display credit 14, so their ACoS figures are not like for like.
"Show spend, attributed sales, ACoS and share of spend by ad type for each week of the window, and label the attribution window behind each number."
Searches for the brand's own name convert cheaply because the shopper already chose the product. Accounts that look excellent on ACoS are often spending most of their budget on their own name. Split the queries into brand and non-brand and the picture can change completely.
"Treat any search term containing [brand name] or [product line names] as brand. Split spend, sales and ACoS between brand and non-brand terms by week. What share of total spend goes to brand terms?"
A healthy account defends its name and grows on non-brand queries. This step shows which of the two the prospect is paying for.
A prospect does not buy nine findings. They buy a number, a reason and a plan, so ask Claude to assemble the audit in that order:
"Write a two-page audit summary for [prospect]: the monthly waste figure with its date window, the three findings with the most money behind them, and a 30, 60 and 90-day plan. Put the baseline metrics in a separate section at the end."
Once a client signs, the same conversation can carry out the fixes. Negative keywords, bids, budgets and campaign states are Amazon Ads Actions in DataDoe. Actions stay off until you enable them, most AI clients ask before a change, and every change is logged. Everything Claude can read and do inside an ad account is on our page about Amazon Ads in Claude, and the agency setup is on DataDoe for agencies.
DataDoe is one plan at $97 a month with unlimited Seller Central, Vendor Central and Amazon Ads connections, so adding a prospect costs nothing extra. Each query Claude runs draws AI Tokens from the monthly allowance, and a single-account audit uses a small part of it. There is a 14-day free trial, and there is an account manager for every customer. If you are weighing this against a separate tool per client, see how DataDoe works as Amazon PPC software.
The analysis is an afternoon once the data is in. The wait is the initial load after the prospect connects, which can run past 48 hours on large accounts.
No. The prospect approves the connection on Amazon's own screen through your invitation link, or adds your team to their Amazon Ads account for an Ads-only connection. Nobody shares a password.
Not on a read-only connection. Read only allows data access and read Actions. Write Actions cannot run against that account.
A PPC audit judges the advertising. An account audit adds listings, inventory, account health and profit, which need the Seller Central or Vendor Central connection. Access works the same way; the checklist is longer.
Eight complete weeks is a good default: long enough for search terms to collect clicks, short enough to reflect the setup as it is today. Leave out the last few days, because Amazon revises recent advertising data.