An assistant grounded in your own scorecards — not a general chatbot with your dashboard bolted on.
Half the questions a client asks are not on the report. Why is cost per lead up on one page and down on another? Which of these findings actually matters this month? What changed since the last run?
Answering means reading across several tables and holding the account in your head — quick if it is your only client, slow if it is your twelfth.
Ask Eleven answers from the stored scorecard data, so its answers are about your account rather than about advertising in general.
A plain question about the client you are looking at.
The assistant reads your stored scorecard data through defined tools rather than guessing from general knowledge.
A hard ceiling on calls and tokens per message, plus rate limiting, so cost is predictable by construction.
The tool-calling loop is capped, and the endpoint is rate-limited. Without a limit there is no upper bound on what a stuck retry or a scripted endpoint could bill.
It reads the same snapshots the dashboard renders, so an answer can be checked against the report rather than taken on trust.
The assistant works within the client you are viewing, so tenant isolation applies to it exactly as it does everywhere else.
It is a language model reading your data, so it can be confidently wrong — particularly when asked to speculate about causes rather than report figures. Treat it as a fast way to interrogate the report, not as a source of truth to forward to a client unread.
No. Data is sent to the model to answer your question and is not used to train it.
There is a hard cap per message and rate limiting on the endpoint. That ceiling exists specifically so the answer to this question is no.
No. It reads and explains. Changes go through Ads Manager, where there is a dry run and an explicit action.
Ask in plain English and get an answer grounded in your own data.
Get started