Prescient AI

Use the Prescient AI connector to sync marketing mix modeling (MMM) output — including halo effects — and platform-reported spend into your Athena lakehouse. This is a Blotout custom connector built on the Airbyte Connector Builder (declarative manifest).

API access is not self-serve

API access must be enabled on the account by Prescient (request through the Customer Support Manager). Until then, the Settings → API tab has nothing to generate. Budget lead time.

Prerequisites

• A tag with Blotout Cloud set up (green checkmark in the Tags table)
• Prescient AI API access enabled on the account (requires CSM request)
• An API token generated by an org admin under Settings → API — displayed once; if lost, generate a new one
• Be aware: tokens can be channel-scoped — a scoped token returns a subset of channels with no error

Create the pipeline

Follow the shared Create a pipeline steps until the source catalogue appears, then select Prescient AI.

Source configuration

On the Source step, configure the connector:

Field	Purpose
Schedule Type	How the pipeline runs — typically Scheduled
Sync Frequency	How often Airbyte syncs (for example, Every 24 hours)
Source Name	Name for this pipeline — used in the Athena schema <SOURCE_NAME>_<ENV>

Authentication

Every request carries Authorization: apikey <token>. Note: the scheme word is apikey, not Bearer.

Configuration fields

Field	Type	Required	Default	Description
api_key	secret string	Yes	—	Prescient AI API token. Sent as Authorization: apikey <token>.
start_date	string	Yes	—	Earliest marketing day to pull (YYYY-MM-DD). This is the full history window — set it to the start of the client’s Prescient history. Not the incremental cursor.
end_date	string	No	Today	Leave blank for scheduled connections.
lookback_days	integer	No	7	Rewinds the processDate cursor only — guards against restatements landing mid-sync. Does NOT control how far back marketing days are read.
sales_channel	array	No	Both	ECOMMERCE / RETAIL filter for modeled_metrics. Empty = both.
base_url	string	No	https://api.prescient-ai.io	The connector appends /graphql.

Supported sync modes

• Incremental Sync — Append (modeled_metrics, reported_metrics)
• Full Refresh (models, channel_names)

Why the cursor is processDate, not reportedDate

An MMM platform restates history — Prescient re-models on new channel connects and ships data patches. A reportedDate-based cursor would never re-read older days. The connector pins startDate to config.start_date (the full window, every request) and advances only processDate (“changed since”), so each sync asks: everything in the whole window that has been re-modeled since I last looked.

Supported streams

Stream	Cursor	PK	Sync Mode
modeled_metrics	processDate	reportedDate, sourceChannelName, sourceCampaignId, metricName, target, targetChannelName	Incremental
reported_metrics	processDate	reportedDate, channelName, campaignId	Incremental
models	—	name	Full Refresh
channel_names	—	id (constant)	Full Refresh

modeled_metrics — the MMM output

Modeled attribution per campaign per day, including halo effects. Rows are long/EAV — one row per metric, so a single campaign-day produces many rows.

metricName is one of eight values:

	REVENUE	CONVERSIONS	LEADS	SUBSCRIPTIONS
FIRST_ORDER_	Direct, last-click-style credit
SECOND_ORDER_	Halo credit on another channel

sourceChannelName = what was spent on. targetChannelName = what got the credit (populated on SECOND_ORDER rows only; null for FIRST_ORDER).

metricValue has no intrinsic unit — join target → models.name → models.unit to determine if it’s revenue, customers, or subscriptions. Never sum across models without checking the unit.

reported_metrics — platform-reported spend

What the ad platforms said was spent, before Prescient models anything. Join to modeled_metrics on campaign + date to get modeled return against actual spend.

models

One row per model: name and unit. Joins to modeled_metrics.target. Also the check stream.

channel_names

A single row listing the channels this token can see. Use it to verify token scope — a scoped token returns a subset with no error.

Downstream notes

• Schema: prescient_ai_<client>_prod.*
• Deduplication (both metric streams are Append; restatements arrive with different values):

SELECT * FROM (
  SELECT *, ROW_NUMBER() OVER (
    PARTITION BY reportedDate, channelName, campaignId
    ORDER BY _airbyte_extracted_at DESC
  ) AS rn
  FROM prescient_ai_<client>_prod.reported_metrics
) WHERE rn = 1

• To pivot modeled_metrics from long to wide:

SELECT reportedDate, sourceChannelName, sourceCampaignId, target,
       SUM(CASE WHEN metricName = 'FIRST_ORDER_REVENUE'  THEN metricValue END) AS first_order_revenue,
       SUM(CASE WHEN metricName = 'SECOND_ORDER_REVENUE' THEN metricValue END) AS halo_revenue
FROM modeled_metrics_deduped
GROUP BY 1, 2, 3, 4

Changelog

Version	Date	Description
1.0.0	2026-09	Initial release with 4 streams: modeled_metrics, reported_metrics, models, channel_names

Complete Schema and Summary

Continue through the Schema and Summary steps to choose streams and confirm the connection. Save and run the first sync.

After a successful sync, Airflow creates an organization-scoped DAG for the pipeline, and dbt models feed downstream analytics.

Related documentation

• Data Pipelines overview — Tags-first create flow