AI planning
- Understand the question
- Form an investigation plan
- Select relevant datasets
- Choose useful comparisons
- Decide what to examine next
Data Integrity
A useful interpretation, with all the evidence attached.
Odaq uses AI to plan investigations and interpret their findings. Connected APIs return the values, deterministic tools perform the calculations, and you can inspect the complete path from question to conclusion.
AI plansQuestion to investigation
Tools calculateAPIs to verified results
AI interpretsEvidence to a useful answer
The agent provides the judgment. Your data provides the evidence.
Question to answer
Follow the complete analysis path.
AI plans the analysis and interprets the findings. Connected platforms supply the numbers, and deterministic tools apply the recipe between them.
- Question
- AI plan
- API datasets
- Visible recipe
- Verified results
- AI interpretation
- Answer or export
Visible recipe Verified
How the result was built
- Source
- Google Ads and GA4
- Date range
- Last 90 days
- Filter
- Campaign status equals Active
- Join
- Campaign ID
- Formula
- ROAS equals Revenue divided by Spend
- Sort
- ROAS descending
AI interpretation2 sources
What the evidence means
Spend increased, but most of the additional budget went to campaigns with declining conversion rates. Three campaigns account for 71% of the efficiency loss.
Google Ads · spendGA4 · revenue
Clear responsibilities
Plan with AI. Calculate with tools. Interpret with evidence.
Data execution
- Fetch API data
- Filter and sort rows
- Join and merge datasets
- Calculate formulas
- Return chart and table values
AI interpretation
- Explain what changed
- Identify meaningful patterns
- Compare possible explanations
- Summarize implications
- Recommend next actions
The agent provides the judgment. Your data provides the evidence.
Focused retrieval
Your data is investigated, not dumped into a prompt.
Odaq does not place an entire marketing history into an agent's context and ask it to find something interesting. The agent identifies the datasets required for the question and retrieves them through purpose-built tools.
It can refine the investigation by requesting another breakdown, narrowing a date range, or comparing another platform.
Only what the question needs
- Less irrelevant data competing for attention
- More precise analysis
- More focused interpretation
- Lower context usage on large accounts
- Easier verification of the final answer
Example analysis recipe
Fetch campaign spend from Google Ads and revenue from GA4 for the last 90 days. Join the datasets by campaign ID. Exclude inactive campaigns. Calculate ROAS as revenue divided by spend. Sort from highest to lowest ROAS.
2 sources1 filter1 join1 formula
Visible methodology
No hidden spreadsheet behind the answer.
The visible disclosure can show sources, selected dimensions and metrics, recognized filters and sorting, joins, and calculated columns. Disclosure depth varies by result type.
The recipe remains separate from the interpretation, so the available methodology can be inspected without treating AI prose as source data.
From APIs to visualizations
Charts render returned data, not generated numbers.
Refreshing a view reruns its recipe against the latest source data. It does not ask AI to recreate values from memory.
- 01Create
The agent creates an analysis recipe.
- 02Execute
Odaq runs the recipe against connected APIs.
- 03Verify
Checks review returned rows, formulas, and joins.
- 04Render
The chart or table displays the verified result.
Evidence-backed interpretation
Accurate data is only the beginning.
A correct table still leaves important questions unanswered. Odaq explains what changed, why it matters, which movements are meaningful, and what to examine next.
- Spend increased by 24%
- Conversions declined by 8%
- Mobile landing-page exits increased by 31%
- Brand campaign performance remained stable
The conversion decline is concentrated in mobile non-brand traffic rather than across the whole account. The timing aligns with a sharp increase in mobile landing-page exits. Brand performance remained stable, which makes a broad demand decline less likely. Review the affected landing pages and recent mobile changes before reducing overall campaign spend.
Interpretation is labeled separately from measured facts.Portable evidence
Every conclusion has supporting data you can take with you.
From the table view, download the rows and columns behind the view you are inspecting as Excel or CSV for additional analysis, reconciliation, sharing, or archiving. The file follows the current table state, including visible columns and any active grouping.
The export follows the current table view
- Rows currently displayed by the table
- Columns currently visible in the view
- Grouped results when grouping is active
Multi-step by default
Good analysis rarely happens in one step.
The agent forms a plan, fetches initial datasets, evaluates what they show, and decides what to examine next. Each step is guided by the evidence returned in the previous step.
- 01
Identify when the conversion decline began.
- 02
Compare spend by channel and campaign.
- 03
Find where the additional budget was allocated.
- 04
Inspect traffic quality and landing-page behavior.
- 05
Check tracking changes and operational events.
- 06
Ask platform specialists to test competing explanations.
- 07
Combine the evidence into a cited interpretation.
- 08
Recommend the next investigation or action.
Sub-agents divide the investigation. They do not manufacture or vote on the numbers.
Depth on demand
Start with the answer. Inspect as deeply as you want.
Interpretation and recommendationChart or tableReturned rowsRecipe, filters, and formulasConnected APIs
Executives
Read the conclusion.
Marketers
Explore the supporting visualization.
Analysts
Inspect the rows and methodology.
Teams
Export the supporting view for independent review.
You do not have to audit every answer. You always have the option.
Get the answer without losing sight of the evidence.
Connect your marketing platforms and let Odaq investigate, calculate, and interpret your data through a process you can inspect from beginning to end.