Nanga

Reports, per advertising platform, whether data really landed over a period: the first and last dates observed, the calendar span between them, the row count and the spend, plus the declared accounts and whether their platform ingested anything. This is what establishes that a period is complete, comparable, or covers every platform; requested bounds alone establish nothing. A platform missing from the answer ingested nothing at all over the period, which is not a platform at zero. `observedDaysSpan` counts calendar days between the first and last dates, not days carrying data. Declared accounts cover two platforms only (Meta and TikTok), and an account list that could not be read is named in `unavailable` rather than returned empty. startDate/endDate are inclusive, at most 90 complete UTC days, ending before today.

describe_paid_coverage

Computes named paid metrics (spend, roas, cpa, cac, aov, ctr and the like; names are the registry's lowercase keys) over a period, at campaign, adset or ad level, with this company's own metric registry. Returns the rows, the definitions actually applied, and `unsupported`: the metrics the registry refuses for this company, each with its reason. A refused metric has no value and no substitute, which is not a zero. startDate/endDate are inclusive, at most 90 complete UTC days, ending before today. Optional groupBy (at most 3) and scope narrow the result. Every row also carries the entity of the chosen level, so a groupBy row is that entity's figure and not a total for the group. The requested bounds say what was asked, not what landed; describe_paid_coverage reports what landed.

compute_paid_metrics

Describes the paid advertising data available to this company: without a table, every permitted table with its columns in one answer; with a table, that table alone. Also returns guidance on the company's metrics, the server's current date and, when the server publishes it, `availableMetrics`: the exact metric names compute_paid_metrics and explain_paid_change accept for this company, some of them French words. An absent `availableMetrics` means the list was not published, not that the company has no metric. Requires the consent for advertising figures, Performances and Kily permissions, and unrestricted folder access.

describe_paid_tables

Finds the companies the connected Nanga user can access, by name or slug. A remote connection is bound to the company chosen when connecting, so it returns that company only. Pending invitations are excluded, and `truncated` says the list was cut at the limit.

list_companies

Lists the saved Performances (paid) or Creative dashboards of a company, with their identifiers, within the connected user's module and folder access. Returns dashboard names only, no performance figures; `truncated` says the list was cut at the limit.

list_dashboards

Compares a metric over two contiguous windows of equal length (windowDays, 1 to 45), both ending before today, and ranks at most 20 campaigns or platforms by their SIGNED contribution to the observed change. Returns both windows with their exact dates, the metric's unit and definition, and the election of a top driver in `topDriverStatus`: `elected` when one entity carries the deterioration (that entity is in `topDriver`, with `volumeShare`, its weight among the ranked entities and not a share of the total budget), `none_material` when worsenings exist but none weighs 5 % of the volume, `no_worsening` when no entity worsened or disappeared. The election covers deteriorations only. The ranking shows where a change came from; it does not establish why an entity moved. An entity absent from the recent window is returned as disappeared, not as zero. A status of `custom_conversion_model` means this company counts its online purchases in a model of its own, so this metric is not ranked for it: not computed, which is not an absence of movement. An optional scope (platform, campaignLike, entityIds) narrows the ranking, for example to one platform's campaigns.

explain_paid_change

Reads the company's curated Company Brain: business context, positioning and priorities when available. Uses the same reader as Kily and keeps the connected user's access to shared or internal content. Requires Kily access and the consent for company context. Does not read conversation history or memory notes, retrieve current metrics, or write anything.

get_company_context

Reads the latest saved paid or creative analysis of a dashboard returned by list_dashboards: its findings and recommendations, with the report's own period, generation date and status. It is a stored report, not a fresh calculation. Requires the dashboard's module and Kily access. Does not generate an analysis, query live metrics or change advertising campaigns.

get_dashboard_analysis

Reads what the company's online store collected over a period: revenue, orders, and the currency those amounts are expressed in, with `currencySource`, which says that currency is an ASSUMED reporting pivot: the source carries no currency column, the amounts arrive already converted, and the pivot is deduced, not read from the store. Also returns advertising spend, the revenue the platforms attribute to themselves, and the resulting ratios with their definitions. Store revenue and platform attributed revenue measure different things and are not interchangeable. The status separates figures available, no sales over the period, and no e-commerce connector at all: none of the three is a zero. Blended ROAS divides total store revenue by advertising spend alone, so it is not campaign efficiency. Requires the store sales consent, on top of Performances and Kily permissions. The advertising figures belong to the separate consent for advertising figures: without it, spend, platform attributed revenue and every ratio come back empty and `withheld` names them, which is not an absence of advertising. startDate/endDate are inclusive, at most 90 complete UTC days, ending before today.

get_sales_overview

Runs one read-only SELECT, written in Snowflake SQL (https://docs.snowflake.com/en/sql-reference), over the tables and columns returned by describe_paid_tables, through Kily's protected warehouse reader, which enforces this company and the requested dates on every table. startDate/endDate are inclusive, at most 90 complete UTC days, ending before today. Returns at most 500 rows, and `truncated` says the result was cut. The requested bounds are no proof that data covers them, and a missing row or a null metric is not a zero. Does not read platform APIs or generate a saved analysis.

query_paid_data