Skip to main content
Skip to main content
Methodology

The measurement methodology

Bayesian, transparent, and honest about what the model doesn’t know.

Priors bank

Every model Acera runs starts with priors — beliefs about likely channel effectiveness, drawn from a k=5 floor of past engagements in the same vertical. Before your data shapes anything, the model knows: in retail, paid search typically drives 30–40% of revenue. That belief updates the moment your data arrives. It shrinks as evidence accumulates. It never disappears entirely — because one quarter of data is never the whole story.

Privacy

Priors are differentially private. Each prior contribution is noise-injected (ε=1.0) before entering the bank. Your data teaches the bank without being identifiable in it.

Opt-in: enabled by default. You can opt out at any time from Settings.

Credible intervals, not point estimates

Traditional MMM reports: “Paid search drove $2.7M.” Acera reports: “Paid search drove $2.7M (95% CI: $2.2M–$3.1M).” That range is not a caveat — it is the answer. The width of the interval tells you how confident the model is. A narrow interval means strong evidence. A wide interval means more data is needed before you act.

Traditional MMM
“Paid search drove $2.7M”
One number. No uncertainty. False precision.
Acera
“Paid search drove $2.7M
(95% CI: $2.2M–$3.1M)
The range is the answer.

Model warm-start

When a new customer joins, the model does not start from zero. It starts from the priors bank for their vertical. The first month’s model is already calibrated to something reasonable. By month three, your own data dominates. By month six, you have a model built entirely on your evidence — with the priors bank as a distant prior that barely moves the needle.

How priors update over time
01
Prior (broad)
Before your data
±40%credible interval
02
After 1 month
Evidence accumulating
±18%credible interval
03
After 6 months
Your data dominates
±8%credible interval

Bell curve width represents credible interval. Narrower curve = stronger evidence from your data.

Data residency

All customer data is stored in Sydney, Australia (AWS ap-southeast-2). Your data does not leave Australian jurisdiction. Full details on the Trust page

See credible intervals in a live result

The Auditor demo shows real credible intervals from a Bayesian MMM run — channel contributions with uncertainty ranges, not point estimates.

See a live Auditor result