STEM University — Prove skills by building

Your Work

Attribution is a governed credit-assignment function: AM_A(M, D=x) = Σ_c M(c)·Σ_h α_A(e_h, c)·1[D(e_h)=x]. The dimension here is the product category visited before each order. Build the journey, pick the model, compare credit.

Date range

Preview slice: 20 of 20 events; 2 conversion(s). Gate uses the full ~3,700-event stream.

Attribution panel (preview)

Side-by-side AM by model for category D = Electronics / Apparel / Home, computed live from the preview slice. Truth values come from the full stream.

CategoryLastFirstLinearLast − FirstElectronics$28.73$0.00$14.37+28.73Apparel$0.00$28.73$14.37-28.73Home$0.00$0.00$0.00+0.00

Time-decay synthetic journey

4-touch journey at t = 0, 7d, 14d, 21d; conversion at t = 28d; half-life h = 7d. Raw weights d_h = 2^(−(t_c − t_h)/h); normalized weights sum to 1.

Touch h=1 at 0d

α = 0.0667

Touch h=2 at 7d

α = 0.1333

Touch h=3 at 14d

α = 0.2667

Touch h=4 at 21d

α = 0.5333

closer / starter = 8.00.

Predict the full-stream values

Σ AM_Linear over all categories

Time-decay closer/starter ratio

U-shape middle weight (n=4)

Multi-visit coverage Q_multiVisit

|{u : {Electronics,Apparel} ⊆ Ch(u)}| (integer)

Hist_Electronics(1) — users with exactly 1 Electronics touch

Which category gets the most credit?

Under Last-touch (full stream)

Under First-touch (full stream)

Δ = Last − First (which gains most)

Attribution semantics

Which lookback window is broader?

Which model fits which business question?

AcquisitionClosingFullJourneyDecay

How does Σ_x AM_Participation compare to Σ_c Revenue(c) when journeys are multi-touch?

Evaluation

Practice checks are unlimited and don’t count toward your grade. Final submit is one-time and generates your certificate.

Practice the workspace, ask Dr. Okoye anything, and submit when you're ready. The grade lives in the workspace, not the chat.