Skip to main content
Version: 3.20.1

uam

Python​

This example shows how to use uam with aggregate marketing data and, optionally, customer journey paths. The aggregate dataset includes both digital signals and a TV signal. TV is useful in this example because it is available in aggregate data but is usually not observed inside user-level journeys. The aggregate baseline uses prior-weighted and saturated signals before computing attribution.

D_variables defines how raw input variables are mapped to final attribution channels. It must include prior_weight, which converts each raw signal into a more comparable business-weighted signal before saturation and scoring. prior_weight: business weight assigned to each raw signal. A common way to estimate it is conversions / touchpoints, but it can also be set from domain knowledge or historical benchmarks. Several variables can map to the same channel: for example, Facebook impressions and Facebook clicks are both assigned to facebook_ads.

import pandas as pd
from ChannelAttributionPro import uam, markov_model, combine_mta_mmm

token = "yourtoken"

# data_mmm_3.csv includes a TV signal in addition to digital aggregate signals.
df_aggr = pd.read_csv("https://app.channelattribution.io/data/data_mmm_3.csv")

D_variables = pd.DataFrame({
    "variable": [
        "direct_searches",
        "facebook_impressions",
        "facebook_clicks",
        "google_impressions",
        "google_clicks",
        "tv",
    ],
    "channel": [
        "direct",
        "facebook_ads",
        "facebook_ads",
        "google_ads",
        "google_ads",
        "tv",
    ],
    "measure": [
        "direct_searches",
        "impressions",
        "clicks",
        "impressions",
        "clicks",
        "views",
    ],
    "prior_weight": [
        0.50,   # direct search
        0.015,  # Facebook impressions
        0.21,   # Facebook clicks
        0.015,  # Google impressions
        0.25,   # Google clicks
        0.02,   # TV views / GRP-like signal
    ],
})

target = "conversions"

Aggregated-data attribution without customer journey paths​

Use this mode when you only have aggregate time-series data. In this case, uam estimates channel attribution directly from the aggregate signals and the target variable.

res = uam(
    df_aggr=df_aggr,
    D_variables=D_variables,
    target=target,
    df_paths=None,
    baseline_model="linear",  # "reward", "copula", or "linear"
    max_p=12,
    nsim=1000,
    seed=1234567,
    verbose=1,
    password=token,
)

attribution = res["attribution"]
print(attribution.head())

UAM with already aggregated customer journey paths​

Use this mode when customer journeys are already represented as one row per path, with the number of conversions and null outcomes associated with each path.

df_paths_agg = pd.read_csv("https://app.channelattribution.io/data/data_uam_paths_t.csv")

res = uam(
    df_aggr=df_aggr,
    D_variables=D_variables,
    target=target,
    df_paths=df_paths_agg,
    baseline_model="linear",  # "reward", "copula", or "linear"
    var_path="path",
    var_conv="total_conversions",
    var_null="total_null",
    order=1,
    sep=">",
    verbose=1,
    password=token,
)

attribution = res["attribution"]
print(attribution.head())

UAM with event-level customer journey paths​

Use this mode when each row is a single touchpoint event. The conversion event is identified by channel_conv and the path is reconstructed internally.

df_paths_events = pd.read_csv("https://app.channelattribution.io/data/data_uam_paths.csv")

res = uam(
    df_aggr=df_aggr,
    D_variables=D_variables,
    target=target,
    df_paths=df_paths_events,
    baseline_model="linear",
    channel_conv="((CONV))",
    order=1,
    sep=">",
    verbose=1,
    password=token,
)

attribution = res["attribution"]
print(attribution.head())