Skip to main content

​
Overview

In Provalytics, a forecast is not built from a historical average or a spreadsheet multiplier. It is built by asking the model a forward-looking question:
If we allocate budget this way, what outcome does the model expect, and how certain is it?
At a high level, Provalytics makes a forecast in four steps:
  1. it reads each channel’s response curve, not just its blended historical ROAS or CPA
  2. it reallocates spend toward the channels with the strongest marginal return, subject to your constraints
  3. it carries that plan through the rest of the funnel, including carryover and media synergies
  4. it returns a distribution of likely outcomes, not a single overly precise point estimate
That is why a Provalytics forecast is better understood as a modeled probability range than a fixed promise.

​
Why historical averages are not enough

Historical averages describe what your budget produced in the past. They do not tell you what the next dollar is likely to do. That distinction matters because channels do not respond linearly forever. As spend rises, many channels begin to saturate, which means each additional dollar tends to produce less incremental value than the last one. Provalytics works from modeled response curves instead of backward-looking averages. Those curves estimate how spend translates into incremental business outcomes across different spend levels.

​
Step 1: Read the response curve

For each channel, Provalytics estimates a response curve that captures how spend affects incremental outcome across a range of spend levels. This matters because the optimizer is not looking for the channel with the highest blended average return. It is looking for the channel where the next dollar is most productive. The key concept here is marginal response. Marginal response means the expected return of one more dollar given where that channel already sits today. In practical terms:
  • a channel early on its curve may still have room to scale efficiently
  • a channel far along its curve may already be flattening
  • two channels with similar historical ROAS can have very different forward-looking value

​
Step 2: Reallocate within the rules

Once Provalytics understands the marginal response of each channel, it evaluates where budget should move next. The engine shifts budget away from channels whose marginal curves have flattened and toward channels that still have room to absorb additional spend productively. This process continues until the plan reaches a practical equilibrium:
if moving one more dollar from one channel to another would improve the forecast, the allocation is not finished yet
The optimizer does not work against a blank sheet. It works inside the operating rules you set. Those rules can include:
  • minimum spend guardrails
  • maximum spend guardrails
  • locked channels
  • contractual commitments
  • flighting constraints
  • total approved budget
That is why a forecast in Provalytics reflects both model evidence and business reality.

​
Step 3: Propagate the plan through the funnel

Channels do not operate in isolation, and the forecast does not treat them that way. Provalytics uses a Bayesian Seemingly Unrelated Regressions framework, which allows connected KPIs to move together rather than as separate disconnected equations. That means a budget change can propagate through the funnel in ways that simpler planning tools miss. Examples include:
  • upper-funnel media lifting branded search
  • branded search driving sessions
  • sessions driving leads, orders, or revenue
Two especially important effects are preserved in the forecast:
  • Carryover: spend can continue working after the original period of exposure
  • Synergy: channels can become more valuable when they work together than when they are evaluated one by one
This is why the forecast is not just a channel-by-channel calculator. It is a modeled system view of how the spend plan should influence outcomes over time.

​
Step 4: Return a range, not a single number

Every forecast in Provalytics is built from a posterior distribution, not from a single fixed coefficient. That means the output is not just:
  • one projected outcome
It is also:
  • a range of plausible outcomes
  • a most-likely expectation
  • a visible level of uncertainty around that expectation
This matters because not every scenario carries the same level of confidence. For example:
  • a plan that leans heavily on well-measured, stable channels may produce a tighter forecast range
  • a plan that depends on thinner or newer channels may produce a wider range
Provalytics does not hide that uncertainty. It treats it as part of the answer.

​
What this means in plain English

A forecast is the model’s answer to a constrained budget question. It is not saying:
this is exactly what will happen
It is saying:
given the current evidence, current channel response curves, carryover, synergy, and your planning constraints, this is the range of outcomes the model considers most plausible
That is the difference between a planning calculator and a probabilistic forecast.

​
How to use this page with the rest of Provalytics

Use this page alongside: Together, these pages explain:
  • how the model evaluates the next dollar
  • how budgets are reallocated
  • how outcomes are forecast
  • how confidence in those forecasts should be interpreted