ExperimentRecommendation

class pymc_marketing.mmm.experiment_design.recommendation.ExperimentRecommendation(channel, spend_change_frac, spend_change_abs, duration_weeks, expected_lift, expected_lift_hdi, snr, assurance, adstock_ramp_fraction, net_cost, score, rationale)

A recommended experiment design for a single channel.

Each recommendation represents a candidate experiment: a specific channel, spend change, and duration, with posterior-derived metrics for the expected lift, power (assurance), and cost.

Parameters:

channelstr

The marketing channel to test.

spend_change_fracfloat

Fractional per-week spend change (e.g. 0.2 for +20%, -1.0 for go-dark).

spend_change_absfloat

Absolute weekly spend change in model-scale units.

duration_weeksint

Active intervention period in weeks.

expected_liftfloat

Posterior mean of total cumulative lift over the experiment.

expected_lift_hdituple[float, float]

94% HDI of the total cumulative lift distribution.

snrfloat

Signal-to-noise ratio (expected lift / measurement noise).

assurancefloat

Posterior-predictive power (Bayesian assurance).

adstock_ramp_fractionfloat

Mean fraction of steady-state effect captured over the experiment duration (posterior mean across draws).

net_costfloat

Direct additional spend: spend_change_abs * duration_weeks.

scorefloat

Weighted composite score used for ranking.

rationalestr

Auto-generated template-based explanation of the recommendation.

Methods

ExperimentRecommendation.__init__(channel, ...)

Attributes

channel
spend_change_frac
spend_change_abs
duration_weeks
expected_lift
expected_lift_hdi
snr
assurance
adstock_ramp_fraction
net_cost
score
rationale