Documentation
Get your data in. Read what comes out.
What Bayesin holds, how each fit is built, and the shape your data needs for a first fit.
Your data
What your data needs to look like.
Weekly rows are the source of truth. If you can export this shape from a spreadsheet, you can get a first fit.
weekly_rows.csv
| week | revenue | paid_search | paid_social | display | |
|---|---|---|---|---|---|
| 2026-01-05 | 48,200 | 6,100 | 4,300 | 2,000 | 450 |
| 2026-01-12 | 51,750 | 6,400 | 3,900 | 2,600 | 450 |
| 2026-01-19 | 46,900 | 5,200 | 4,800 | 1,400 | 600 |
| 2026-01-26 | 55,300 | 7,000 | 5,100 | 2,200 | 600 |
01One row per week
Each row is a week, named by its start date. Weeks run in sequence; a missing week is reported, never filled in.
02One revenue column
The number the model explains. Use the revenue you already reconcile to, so the fit and your ledger describe the same business.
03One column per channel
Spend by channel for that week. Names map to your channel registry, so a channel keeps one name across every source.
04New names ask first
A channel name the registry hasn't seen needs a create-or-map decision, so a new spelling never becomes a new channel.
Data thresholds
What a fit needs from your data.
The data check runs on upload, before a fit is queued. A fit that can't run tells you what is missing.
Weekly history
52 weeks
The minimum for a media mix model. 104 is recommended; between the two, the fit runs with wider intervals.
Spend variation
15% CV
A channel whose weekly spend barely moves is flagged. The fit still runs, with a wide interval on that channel.
Channel history
26 weeks
A channel with spend in fewer weeks is flagged. Group it with a similar channel to sharpen it.
Getting started.
01 Your organization
It owns its sources, channel registry and datasets, kept apart from everyone else's.
02 Connect or upload
Connect a source or upload a weekly file. Status shows the data check, not just the upload.
03 The first fit
When the data qualifies, the fit queues. You see when it's waiting, why, and what it found.
What attribution needs.
A touchpoint stream
Every interaction, with its channel and time.
Conversions joined to it
So each journey ends in revenue.
Identity resolution
Identifiers resolved to subjects your organization owns.
A lookback that covers the real one
Long enough for your actual buying cycle.
A lawful basis
For any person-level data you send.
Reference.
Getting started
- Create an organization
- Connect a first source
- Run the first fit
- Invite your team
Models
- Media mix modeling: carryover, saturation, priors
- Attribution runs, credits and learned multi-touch
- Forecast scenarios and intervals
- Budget simulator and proposed splits
Touchpoint schema
- Touchpoints and conversions
- Identity: subjects, identifiers, resolution
- Consent, suppression and erasure
- Lookback and retention
Sources
- Weekly files and sheets
- Google Analytics 4 and Search Console
- Ad platforms
- CRM, commerce and payments
- Warehouses
Vocabulary
- Presets: Generic, Franchise, Retail, B2B
- Overriding individual terms
- Segments, channels, units and groups
Security and privacy.
Separate by organization
Row-level policies keep each organization's sources, registry and datasets apart.
Retention you control
Retention windows are your settings. Deleting an organization removes its data on the documented schedule.
Erasure, per subject
Suppression and erasure requests are enforced before a journey is assembled.
Need something the docs don't cover?
Tell us what you want to model, and a measurement lead will walk through it with you.