Product.ai / Join / Projects / SimplyCodes Conversion Attribution Model — replace heuristics with a defensible attribution stack
Project Open to Alpha Team

SimplyCodes Conversion Attribution Model — replace heuristics with a defensible attribution stack

Replace SimplyCodes' current conversion attribution heuristics with a defensible attribution model. Causal-inference-first design. Handles the messy realities — last-touch bias, multi-touch attribution, novelty effects, measurement attenuation, network effects across affiliate partners. Output is a working model in production producing better attribution than the heuristic it replaces, with documented tradeoffs and a quarterly review cadence.
Project Overview
Discipline
AI Systems — Data Scientist / ML Engineer
Duration
2 weeks
Compensation
Your stated freelance rate
Surface
SimplyCodes · Revenue
Kernels
simplycodes · revenue
Outcomes
mlp-convert · attribution · sc-traffic
Tier
Applied
Alpha Team
Open to alpha members who want to take this on
Tooling
Claude Code or Co-work

Why we want this done

Attribution is foundational. Every SimplyCodes growth conversation, every conversion experiment (PRJ-24), every revenue-engine ROI calculation depends on it. Today's heuristics likely have known biases. Last-touch attribution systematically over-credits the closing surface and under-credits earlier touch. Multi-touch heuristics weight by guesswork. The Data Science Phase 3 briefing identifies causal-inference rigor as one of the four-dimension hiring screen. This project tests it on real revenue data with real consequences — the model the candidate ships gets used to make real budget allocation decisions. Substrate-builder vs substrate-rider shows up in attribution work transparently: substrate-builders explicitly handle confounders; substrate-riders run a baseline regression and report R².

Scope

  1. Audit current SimplyCodes attribution — what heuristic, what known biases, what decisions ride on it
  2. Survey the data — affiliate-partner traffic, on-site behavior, code redemption, post-purchase outcomes
  3. Pick a defensible attribution methodology — argue for it (Markov attribution, Shapley value, incrementality testing, or a hybrid)
  4. Build the model — in production-deployable form (not a notebook)
  5. Validate against held-out data and against any existing experimental data the team has
  6. Ship to production behind a feature flag; run in shadow mode first if the team's process supports it
  7. Documented architectural decision record — methodology choice, alternatives rejected, known limitations, quarterly review trigger

What success looks like

  • The model ships to production (shadow mode at minimum)
  • Validation against held-out data is documented; against experimental data if available
  • Known biases are NAMED — not hidden ("this model under-credits brand search because we don't have brand-search instrumentation")
  • The ADR is one page; a stranger can understand the methodology and tradeoffs
  • The quarterly review cadence is set up (calendar entry, owner, re-assessment trigger)
  • The candidate's causal-inference reasoning is visible in writing — confounders identified, correction strategy named

References

references.md
Data Science Phase 3 briefing axiom A3 (Eval-Design Fluency Primary-Among-Several including Causal Inference Rigor), A4 (Substrate-Builder), VERDICT 1 (SimplyCodes-First Sequencing)
Existing SimplyCodes attribution heuristics and associated metrics
attribution.md, mlp-convert.md, sc-traffic.md outcome files
Standard causal inference methodology (Pearl, Imbens/Rubin, Markov attribution, Shapley value attribution literature)
Existing experimental data from any prior A/B tests (the candidate negotiates access on Day 1)

Constraints

  • Claude Code or Co-work as primary substrate
  • Production-deployable form is the deliverable — not a research notebook
  • Shadow mode for at least 2-3 days before any user-visible decision rides on the new model
  • Validation against held-out data is required; experimental validation if any data exists
  • Known biases must be NAMED in the ADR; hiding limitations is a fail
  • IP separation: SimplyCodes attribution data is application-layer; methodology paths are out of scope
  • Quarterly review cadence is non-negotiable (statistical models decay)
Apply
01

Read the Codex (10 min)

The operating principles we work by. If they resonate, the rest of this will land. Open the Codex →

02

12-minute video screen

Hireflix, async. Questions are calibrated to this project specifically.

03

Chemistry call (30-60 min)

Direct call with the CEO. Strategic alignment and mutual fit. No problem-solving exercise.

04

Project begins within 2-3 weeks

1099 contractor agreement, NDA, paid at your stated rate. Day 1 in Santa Monica.

Alpha Team members can take this project without the screen-and-call sequence. Reach out via the Alpha Team channel.

Apply for this trial Browse other projects