Multi-touch attribution fails when nobody can say who picked the model. Dock separates the raw touchpoint data from the agent's interpretation, then attaches a reviewer to every weighting decision. The model is auditable, the splits are reproducible, and marketing-ops signs off before the report ships. This is the pattern that makes Dock for marketing behave the way finance teams expect their numbers to.
HubSpot, Marketo, and Salesforce stay the system of record for the raw data. Dock is the system of record for what the AGENT INTERPRETS. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Dock surface: Attribution Decisions table
| opportunity_id | sf_link | model_applied | top_touch | credit_split | agent | reviewer | status | decided_at |
|---|---|---|---|---|---|---|---|---|
| OPP-44812 | sf://opp/006... | W-shaped v3 | Marketo nurture #218 | 30/40/30 | mira-attribution | jordan@ | approved | 2026-05-29 14:02 |
| OPP-44907 | sf://opp/006... | Time-decay v2 | HubSpot webinar 04-18 | 12/18/22/48 | mira-attribution | jordan@ | needs-review | 2026-05-29 16:11 |
| OPP-44950 | sf://opp/006... | Linear (fallback) | Salesforce SDR call | 25/25/25/25 | mira-attribution | priya@ | approved | 2026-05-30 09:40 |
The sf_link column points back to the Salesforce opportunity. The model_applied field records which weighting the agent chose and which version of the rules library it ran against. If a board member asks why OPP-44907 credits the April webinar at 48 percent, the row answers it.
The workflow
Mira, the attribution agent, runs nightly. She pulls closed-won opportunities from Salesforce, fans out to HubSpot for form-fills and to Marketo for nurture engagement. She normalizes timestamps, drops touches outside the 90-day lookback, and applies the chosen model. Any opportunity where one touch exceeds 40 percent credit routes to marketing-ops. Jordan opens the row, approves or overrides, and the report regenerates. The CMO's weekly deck pulls from approved rows only. Overrides are also rows, with the same identity-and-timestamp shape used in agent audit and compliance.
Why it matters
Attribution reports drive budget. A pipeline-influenced number that nobody reviewed gets walked back in a board meeting. Making the agent's model choice a reviewable row keeps the speed of automated rollups without the credibility risk. The same pattern shows up in Dock for sales for forecast adjustments and in Dock for data analytics for metric definitions. The agent has its own agent identity, so the decision log distinguishes Mira's call from a human override.
Linear, time-decay, U-shaped, and W-shaped models all distribute credit in defensible ways, but the choice between them is a judgment call about how your funnel behaves (HubSpot, "Attribution Modeling"). Multi-touch attribution software makes the rollup mechanically simple, which is why the decision layer needs to sit somewhere humans can challenge it (HubSpot, "Multi-Touch Attribution").
Start here
Point Mira at one closed-won quarter and one model. Review the rows. Ship.
FAQ
Q: Does Dock store the raw touchpoint data? A: No. HubSpot, Marketo, and Salesforce remain the source. Dock stores the interpretation, the pointer, the agent, and the reviewer. Re-running the report pulls fresh platform data.
Q: What happens when marketing-ops overrides the agent? A: The override is a new row with a reviewer attribution and a reason field. The original agent decision is preserved. The audit pattern follows agent audit and compliance.
Q: Can the agent run multiple models in parallel? A: Yes. Each model run is its own set of rows. Teams often compare W-shaped against time-decay for the same cohort, then approve one for the official report.
Q: How does this relate to pipeline forecasting? A: Forecasting uses a separate agent and a separate reviewed table. The attribution rows feed campaign-ROI rollups; the forecast rows feed pipeline. Both live in the same marketing workspace.
