Skip to content
AdCrunch
Esc
↑↓navigate↵open⌘Jpreview
On this page

Read your accounts

Ask an AI agent which ad accounts you have, what runs in them, and what they spend and return. The agent reads what AdCrunch stores, and it changes nothing.

Ask an AI agent about your connected ad accounts: which accounts you have, what runs in them, and what they spend and return. The agent answers from four tools. Each tool only reads, so no question in this job changes a campaign or spends money.

Before you start

  • Connect a provider. Meta, TikTok and Google Ads give campaigns and metrics. What each provider supports tells which provider gives what.
  • The agent needs the scope observe:read. Auth & scopes tells how the agent gets it.

1. Find your ad accounts

Which ad accounts do I have in AdCrunch?

The agent calls list_advertisers. You see each ad account that your organization connected: its name, its provider, its currency, and whether it is active. The agent keeps the id of each account, so that the next steps name the account by its id and not by its name.

To see the accounts of one provider only, name the provider: “Which Meta ad accounts do I have?”

2. See what runs in an account

List the active campaigns in the EU account, with their daily budgets.

The agent calls list_entities for that account. You see each campaign with its status, its objective, and its budget in the currency of the account. A campaign whose ad sets have the budget shows no budget of its own.

To go one level down, name a campaign: “Show the ad sets of the Q4 Acquisition campaign.” The agent then lists the children of that campaign. Each provider has its own word for this level: an ad set on Meta, and an ad group on TikTok and on Google Ads.

A long list comes one page at a time. The agent asks for the next page when it needs more rows.

3. Read all the settings of one campaign

Show me all the settings of the Q4 Acquisition campaign.

The agent calls get_entity with the campaign that it found in step 2. You see each field that AdCrunch stored from the provider for that campaign, in the words of the provider, such as its start time. Use this step when the list of step 2 does not show the field that you need.

4. Measure performance

How much did each campaign in the EU account spend last week, and what was the ROAS?

The agent calls query_insights. You see one row for each campaign, with its spend and its return on ad spend for that week. When the agent sends no dates, AdCrunch reads the last 15 days.

Ask a follow-up question to change the view:

  • “Show it day by day.” The agent adds one row for each day.
  • “Now by ad set.” The agent reads the ad sets in place of the campaigns.
  • “Show all of it in euros.” The agent asks AdCrunch for euros. AdCrunch converts each day at the European Central Bank rate of that day, and then adds the days up. Each row then shows euros. With no currency, each row shows the currency of its ad account.

Guardrails

  • Read-only. No tool of this job changes your ad accounts.
  • Your organization only. An ad account that your organization does not own gives an empty answer. For an entity of another organization, get_entity answers not_found, the same answer as for an entity that does not exist. So the agent cannot find out what another organization has.
  • The data comes from AdCrunch, not from the provider. The four tools read the copy that AdCrunch stores. The ingestion of that copy starts each day at 00:00 UTC. So a change that you make in the provider’s own interface shows after the next ingestion. Right after you connect a provider, the first ingestion is still in progress, and the answers can be empty or incomplete.
  • A budget can be old. Meta does not mark a campaign changed when only its budget changes, and AdCrunch does not follow a budget change on Google Ads yet. So a budget that someone edits in the provider’s own interface can be old.
  • Each provider reports different metrics. A metric that a provider does not report reads 0. Google Ads reports no reach, for example. Before you compare one metric across two providers, read what each provider supports.
  • A failure has a code. When a call fails, the agent gets a code and a message. Errors lists the codes.
  • The same rules for each tool. What to expect states how fresh a stored answer is, and how AdCrunch states money.

Was this page helpful?