Data to AI™ / Gemini/DV360
Feed DV360 pacing, reach and YouTube metrics into scripts, notebooks and test logs from the terminal or VS Code, with Gemini asking before each data lookup.
Better answers, fewer tokens.
Gemini CLI
Which insertion orders showed ads to the same people too often this month?
Retargeting Display reached 96,500 people at 11.2 impressions each, more than three times the 3.4 of Brand Awareness CTV. Prospecting Video sat in between at 7.8.
| Insertion Order | Unique Reach: Total Reach | Unique Reach: Average Impression Frequency |
|---|---|---|
| Retargeting Display DV360 Reach/Frequency by Insertion Order | 96,500 | 11.2 |
| Prospecting Video DV360 Reach/Frequency by Insertion Order | 412,000 | 7.8 |
| Brand Awareness CTV DV360 Reach/Frequency by Insertion Order | 1,180,000 | 3.4 |
Reach is shown per insertion order and not added up, because DV360 reach is estimated.
Dataddo extracts Google Display & Video 360 (DV360) reporting through a flexible report dataset with report types such as Standard and YouTube, plus a reach and frequency dataset and an advertiser list.
Pull Line Item Pacing Percentage, Budget Segment Budget and Total Media Cost (Advertiser Currency) per Line Item for yesterday and feed them to scripts/pacing_alert.py with --dry-run.
A programmatic engineer tunes alert thresholds in Gemini CLI against real values from yesterday, sees exactly which line items would have triggered, and only then schedules the job.
From DV360 Reach/Frequency by Insertion Order, write Insertion Order, Block Start Date, Unique Reach: Total Reach and Unique Reach: Average Impression Frequency to notebooks/data/frequency.csv.
A media analyst works in VS Code with Gemini Code Assist in agent mode and charts frequency for each insertion order block by block, so no cumulative estimate is stitched onto another.
Compare TrueView: View Rate and Earned Views per YouTube Ad Group for the 7 days before and after 12 Oct, when the new cut-downs went live, and append the table to docs/youtube-tests.md.
A video operations lead keeps the team's creative test log in git, so every asset change has its before and after numbers next to the brief that requested it.
Data to AI™ is priced per seat, like the AI assistants your team already uses, with ~25% off on annual billing. Need tailored contracts or volume pricing? Talk to us about Enterprise.
Serve governed business data to Claude, ChatGPT and other AI tools over MCP.
$10.49/month
per seat
3 seats minimum
Flexibly deploy Dataddo in any cloud or hybrid environment.
Custom
We have a payment model that works for you
| Without Dataddo |
With Dataddo
|
Outcome for you | |
|---|---|---|---|
| Scripts that need pacing numbers | A script waits for a scheduled offline report, parses the downloaded file, and breaks the day someone renames or reorders the report's columns. | Gemini CLI requests Line Item Pacing Percentage by its field name from the scheduled DV360 source, and you confirm each tool call or pick Always allow for that tool. | Field names instead of file parsing |
| Reach rows in a notebook | A notebook loads a year of reach and draws one continuous line, quietly joining estimates that DV360 calculated for separate periods. | Each row Dataddo returns carries Block Start Date and an Exceeds 93 Days flag, so Gemini writes one CSV row per block and the chart keeps periods apart. | Periods plotted separately |
| What Gemini is able to see | Nobody on the team can say which reporting data a terminal assistant on someone's laptop has been reading. | After sign-in, Gemini reaches every flow or Context attached to an AI destination or AI Model in the Dataddo account, so the attached list is the scope to review. | Scope you can audit |
| Credentials on a laptop | Reporting credentials get pasted into .env files, where one careless commit can publish them. | Gemini CLI keeps the Dataddo sign-in in ~/.gemini/mcp-oauth-tokens.json and renews it automatically, outside the repository you are working in. | No secrets in the repo |
| Token consumption | Gemini spends tokens exploring raw DV360 columns, sampling rows and retrying until it understands the data. | Field definitions and relations are supplied up front, so the model skips the exploration and goes straight to the answer. | Fewer tokens per answer |
| Answer quality | Prone to wrong joins, invented fields and numbers that are hard to check. | Answers grounded in defined fields. In the Dataddo benchmark, 78.5% of questions were answered correctly, against 65.5% with plain CSV files. | Better answers |
The DV360 datasets marketing teams use most, with their real field names. Gemini queries them by name through the Dataddo semantic layer.
| Dataset | Metrics and attributes |
Ask Gemini
|
|---|---|---|
| Google Display & Video 360: delivery and cost |
|
Write daily Impressions, Clicks and Total Media Cost (Advertiser Currency) for the last 60 days to a CSV for my spend forecast notebook. |
| Google Display & Video 360: breakdowns |
|
Which Insertion Order names do not match the naming pattern defined in our campaign config file? |
| Budget and quality |
|
Can you export Line Item Pacing Percentage for every line item as JSON so the pacing alert script can test its thresholds? |
| YouTube report |
|
Did TrueView: Views per YouTube Ad Group drop in the days after we swapped the video assets? |
| DV360 Reach/Frequency by Insertion Order |
|
List Unique Reach: Total Reach per Insertion Order and Block Start Date so my notebook can plot each block on its own. |
| DV360 Advertisers |
|
Which Display Name values changed since the advertiser list we last committed, so the dashboard labels can be updated? |
We asked 43 pre-registered questions of three real datasets - Google Search Console, HubSpot deals and Google Ads - and delivered the same rows to the same model in four ways. Only the data layer changed.
of questions answered correctly - the best of the four delivery methods tested
fewer confidently wrong answers - 14.0% of runs against 24.0% and 26.3%
more correct answers on messy CRM data - better on 8 questions, worse on none
Same model, prompts and rows in every condition; 759 graded runs, ground truth frozen before any run. On Google Search Console alone, documented CSV files did as well as Data to AI™ - the advantage comes from messier data such as HubSpot deals.
Read the benchmark →Create a Google Display & Video 360 source in Dataddo with a Standard report and DV360 Reach/Frequency by Insertion Order, attach it to a Google Gemini destination or AI Model, and add Dataddo to ~/.gemini/settings.json with httpUrl https://headless.dataddo.com/mcp-data. Restart Gemini CLI and run /mcp auth dataddo to sign in.
This setup relies on no built-in link between Google products. Dataddo runs your DV360 reports on a schedule, and Gemini CLI reads them through the Dataddo Data Access MCP server.
With the url key, Gemini CLI falls back to SSE, which Dataddo does not support. Add the server under mcpServers with httpUrl and no headers or oauth block, then restart and check /mcp.
Yes. Put the server in .gemini/settings.json inside that project instead of ~/.gemini/settings.json, which applies to all projects. Gemini Code Assist in VS Code reads the same file.
No. DV360 reach comes in blocks of up to 93 days, and those blocks cannot be summed. Ask Gemini for one row per Block Start Date and keep each insertion order separate.
Your Dataddo email and password, with MFA if prompted; Google and Microsoft SSO are not offered for this sign-in. Gemini CLI opens the page in a browser on the same machine.
No. Access is read-only over the data Dataddo extracted, through a semantic layer: Gemini asks for entities and fields by name, Dataddo never accepts SQL from it, and nothing goes back to DV360.
