Setup
Open Configurations (/configurations). Tabs: Configurations List, Analysis & LLM, Agents. Import (JSON) sits on the right.
Create a configuration
- Create New (Quick Access) or click a card to edit.
- Name and optional description.
- Set configuration type: Analytic, Interactions, Personality, or Financial (help modal shows sample rows).
- Bind a data source:
- MongoDB — database + collection (optional aggregation pipeline). Default collection in samples is often
logging.data_ecosystemruntime_flattenfor runtime logs. - Trino — catalog / schema / table or SQL. Trino is extract-only; report writes stay on Mongo.
- MongoDB — database + collection (optional aggregation pipeline). Default collection in samples is often
- Map fields the type requires (especially Financial — aliases alone are not enough).
- Enable analysis families (or leave defaults). Disable unused families to keep jobs fast.
- Save. You cannot edit a config while a job for it is running or queued.
Import loads a previously exported JSON (metadata stripped). Use this to clone environments.
Analytic
Best for a single wide table (events, customers, transactions).
Enable on Analysis & LLM (or the form):
| Module | What it produces |
|---|---|
| Main analysis | Profiling, correlations, histograms — Main analysis |
| Categorical | Tests and charts for categories — Categorical |
| Actuarial | Risk, extremes, survival-style metrics — Actuarial |
| Feature engineering | ML-ready features — Feature engineering |
| Time series | Off by default. Needs a datetime column — Time series |
Optional: LLM interpretation on the analysis agent. Time Series also has its own page for datetime, grain, journeys, and event association.
Interactions
Full walkthrough: Interactions Analytic.
Pair calls (ecosystemruntime) with responses (ecosystemruntime_response) by UUID. Mongo connection and database come from Source. Then:
- Set calls and responses collections.
- Discover Predictors — pick specific predictors or Analyze all. Required before Run on Mongo.
- Optional date range and attribution window (hours, default 168).
- Optional declared experiment groups (control + treatments).
- Enable families on Analysis & LLM. Marketing mix is off by default.
| Family | Focus |
|---|---|
| General performance | Campaign & predictor — also channels, convergence, economics, fairness |
| Offers and catalog | Offers & catalog — also experimentation |
| Customer engagement | Customer uptake |
| Market intelligence | Market intelligence |
| Marketing mix | Adstock / saturation — off by default |
| Entity clustering | Offers by category — on by default when enough catalog rows |
Acceptance is uptake, not purchase or causal lift. LLM interpretation cannot change metrics. After a simulation, point this config at the same logging collections.
Personality
Requires transactions with customer, amount, merchant/category, timestamp. Optional score / campaign collections (often suggested from the source name).
| Family | Focus |
|---|---|
| General / spend / money | Overview, spend, money |
| Drift | Transitions |
| Category patterns | Txn-scan co-occurrence |
| Calendar | Weekend, DOW, daypart |
| Recommendations | Lifecycle cards |
| Intelligent campaigning | Playbook / export |
| Customers / data quality | Change reasons |
Needs enough history for drift. Not a substitute for Spend Personality scoring in Runtime.
Financial
Map identity, time, and recognized revenue. Map cost if you need ROI/CAC. Industry profile: generic, banking, telco, or gambling.
| Family | KPIs (when mapped) |
|---|---|
| Unit economics | ARPU / ARPA / ARPPU |
| Returns | ROI, ROAS, CAC |
| Value | Observed LTV |
| Profitability | Margin, cost-to-serve |
| Growth | NRR, GRR, churn |
| Time / breakdowns / distributions / comparisons | Time, breakdowns, distributions, comparisons |
| Campaigns | Attribution |
| Ledgers | AR/AP/GL |
| Industry pack | Profile-specific |
| Data quality | Mapping coverage |
ROI is skipped, never invented, if cost/investment is not mapped.
Dimensions (product, region, customer, division, category, account, type) slice measures. Hierarchies such as region → category → product run when two or more levels are mapped.
Analysis & LLM tab
After Save, use Analysis & LLM for the selected config:
- Agent LLM interpretation
- Module LLM interpretation (Personality / Interactions / Financial)
- Per-type family checkboxes (same families as above)
Save Settings applies to new jobs, not in-flight ones.
Agents tab
Embedded test console: bind the saved config, pick a specialist, ask questions over jobs and reports (MCP tools). Same catalog as the Agents sidebar page.
After a job completes, each family has its own report page under Reports.