Multi-Touch Attribution Model

7-Channel Integration

ConversionOS includes a data-driven multi-touch attribution model that distributes conversion credit across all marketing channels. Replaces last-click with ML-based attribution.

Supported Channels

#	Channel	Data Source	Attribution Weight Method
1	Paid Search	Google Ads API	Click-weighted Shapley
2	Paid Social	Meta/LinkedIn APIs	View-through + click
3	Organic Search	GA4 / Search Console	Session-based
4	Email	SFMC engagement data	Click attribution
5	Direct Mail	Match-back via Adstra	Household-level match
6	Call Center	IVR + agent disposition	Inbound call attribution
7	Organic Social	UTM-tagged traffic	Session-based

Attribution Methodology

Data-Driven Attribution (Shapley Values)

ConversionOS uses Shapley value attribution — a game-theory approach that assigns credit based on each channel's marginal contribution:

import itertools
from collections import defaultdict

def shapley_attribution(conversion_paths, channels):
    """Calculate Shapley values for channel attribution."""
    shapley_values = defaultdict(float)
    n = len(channels)

    for channel in channels:
        for subset_size in range(n):
            for subset in itertools.combinations(
                [c for c in channels if c != channel], subset_size
            ):
                subset_set = set(subset)

                # Conversion rate WITH this channel
                with_channel = conversion_rate(
                    conversion_paths, subset_set | {channel}
                )
                # Conversion rate WITHOUT this channel
                without_channel = conversion_rate(
                    conversion_paths, subset_set
                )

                # Shapley weight
                weight = (
                    factorial(subset_size) *
                    factorial(n - subset_size - 1)
                ) / factorial(n)

                shapley_values[channel] += weight * (
                    with_channel - without_channel
                )

    return dict(shapley_values)

Path Construction

-- Build conversion paths from touchpoint data
WITH touchpoints AS (
  SELECT
    user_id,
    channel,
    touchpoint_timestamp,
    ROW_NUMBER() OVER(
      PARTITION BY user_id ORDER BY touchpoint_timestamp
    ) AS touch_order
  FROM `project.conversionos.unified_touchpoints`
  WHERE touchpoint_timestamp >= DATE_SUB(CURRENT_DATE(), INTERVAL 90 DAY)
),
paths AS (
  SELECT
    user_id,
    STRING_AGG(channel, ' > ' ORDER BY touch_order) AS conversion_path,
    COUNT(*) AS path_length,
    MAX(CASE WHEN converted = 1 THEN 1 ELSE 0 END) AS converted
  FROM touchpoints t
  LEFT JOIN conversions c USING(user_id)
  GROUP BY user_id
)
SELECT * FROM paths;

Incrementality Validation

Attribution models tell you correlation. Incrementality testing tells you causation:

• Geo-lift tests: Holdout geographic regions from specific channels
• PSA holdouts: Serve public service ads to control group
• Platform lift studies: Google/Meta native conversion lift measurement
• Calibration: Use incrementality results to adjust attribution weights

Output Integration

Attribution results flow into:

1. Budget allocation models — Shift spend toward channels with highest incremental ROAS
2. Reporting dashboards — Replace last-click reports with data-driven attribution
3. Model features — Attribution-aware channel engagement features improve propensity models