Posterior Distribution
In Bayesian inference, the probability distribution over a parameter after combining the prior with observed data via Bayes' theorem.
Last updated
Definition
The posterior IS the answer in Bayesian MMM — every per-channel ROAS, adstock half-life, and saturation curvature lives inside its posterior distribution. Decision tools sit on top: the expected value (mean) is what feeds a budget optimizer; the credible interval is what surfaces in a dashboard; the full posterior is what feeds an exploration-vs-exploitation policy.
How it applies in India
When an Indian brand also sells on marketplaces such as Amazon or Flipkart, the sales its ads drive there are missing from its own order data, so the posterior describes only the sales the model was given.
An example
A campaign has spent ₹50,000 and recorded twelve conversions. A single estimate says it returns 2.4 rupees per rupee, but with so few conversions the honest answer is a spread, perhaps 1.5 to 3.5. Planning on the spread rather than the single number stops one lucky week from doubling the budget.
How WatEase uses it
WatEase’s budget planner treats every spend-and-revenue curve as uncertain. For each plan it draws 200 plausible versions of the curve, with a spread that narrows as conversions build up (or the marketing-mix model’s own spread, where there is one), and plans on all of them: the average gives the target, the spread gives a ± band, and the share of draws that favour raising or trimming the budget shows how settled the decision is. When the spread is very wide, the curve is marked unidentified and its confidence is cut. This is a sampling approximation of a posterior, not a full Bayesian fit.
Frequently asked questions
Does WatEase show a posterior distribution?
Not as a chart. You see what it is used for: a target with a ± band, and how many of the 200 drawn curves favour raising or trimming a budget.
Related terms
- Bayesian MMMMarketing Mix Modeling implemented with Bayesian inference (typically MCMC sampling), producing posterior distributions over channel contributions instead of point estimates.
- Prior DistributionIn Bayesian inference, the probability distribution that encodes belief about a parameter before observing the data.
- Credible IntervalA range that contains the true parameter value with a stated probability (e.g., 95%). The Bayesian counterpart to a frequentist confidence interval.
Try WatEase
Run your WhatsApp commerce on the platform built for India — Cloud API, GST invoices, UPI checkout, opt-in tracking, and 0% markup on WhatsApp messages (competitors add 20–25%). Pricing is on request — quoted by the WatEase team.
Get Started Today