Prediction Confidence for Incremental Forecasting

What is Prediction Confidence?

When you look at a saturation curve in Incremental Forecasting, you're seeing a model's prediction of how your spend drives revenue. Prediction Confidence shows how well that forecast has held up recently, using automated backtesting.

It gives you a track record before you act — not after. Instead of moving budget and finding out later whether the recommendation was right, you can see how the forecast has performed over the last 30 days first.

Example: A Prediction Confidence score of 80% means that on 24 out of 30 back-tested days, the revenue outcome for that channel landed inside the model's predicted range.

How Automated Backtesting Works

Before surfacing any prediction, Fospha runs an automated backtest in the background. Here's what that process looks like:

Fospha temporarily hides the most recent 30 days of data from the model.
The model predicts what it expects the revenue outcome to have been during that window - including a predicted range for each day.
Fospha checks how often that revenue outcome landed inside the predicted range.
The Prediction Confidence score reflects that result.

This runs automatically for every active channel. You don't need to set anything up - the score updates continuously, so you always have a current read on how the forecast has held up.

What the Score Measures

The score counts how many days out of the last 30 the revenue outcome for that channel fell inside the model's predicted range.

Score	What it means	What to do next
Strong	The forecast has held up well recently. A solid basis for acting on the recommendation.	Treat this as a good candidate for scaling. Still sense-check against channel trends and any known business context (seasonality, a recent campaign change) before committing budget.
Moderate	Reasonable track record. Review alongside other signals before making significant budget changes.	Use the recommendation as a strong input, not the only input. Cross-check against platform reporting or recent performance before sizing up a budget move.
Low	A prompt to test more tentatively for this channel, not a reason to avoid it.	Size the test smaller and more contained rather than committing to the full recommended move. As more recent data comes in, the score updates - revisit before scaling further.

Keep in mind: A strong score doesn't guarantee the next prediction will be exact. It shows the forecast has held up recently. Unusual market conditions or a sudden spend change can affect the score — that's expected, not a fault in the model.

Where to Find Prediction Confidence

Prediction Confidence appears in two places within Incremental Forecasting.

Above each saturation curve

A Prediction Confidence badge sits above every saturation curve. It shows:

The confidence score (e.g. 80%)
The tier (Strong / Moderate / Low)
The label: Based on last 30 days of automated backtesting
A How is this calculated? link, which opens the full methodology explanation

In the summary table

The channel summary table includes a Confidence column, showing the per-channel score and tier so you can compare across your full channel mix at a glance.

What Changes With Prediction Confidence

Before	After
No in-platform evidence of how a forecast had held up — you had to take it on trust or ask your account team	A Prediction Confidence score appears directly above each saturation curve, based on automated backtesting
Comparing forecast performance across channels meant a manual deep-dive	A per-channel Confidence column lets you compare at a glance
Defending a budget move to finance meant saying "the model recommended it"	You can show how the forecast has held up over the last 30 days, and explain how it's calculated

Tips

Use the score as a starting point, not a final answer. A channel with strong confidence is a good candidate for scaling — but review it alongside channel trends and business context.
Check confidence before acting on a saturation curve recommendation. Where confidence is lower, treat the forecast as a guide for a smaller, more contained test rather than a precise target.
Use the Confidence column to prioritise which channels to investigate first when planning budget shifts.

How Prediction Confidence Supports Your Team

A forecast only drives action when the people using it — and the people approving it — feel confident enough to move. Prediction Confidence gives every stakeholder a shared reference point.

For marketing teams

Budget decisions move faster when you can show your work. Rather than presenting a saturation curve and asking for trust, you can point to how the forecast has held up recently. That's especially useful when justifying investment in upper-funnel or demand-generation channels, where the link between spend and revenue is harder to see in platform reporting.

For finance and leadership

When you need to justify a budget move to your CFO or leadership, Prediction Confidence lets you show the forecast's recent track record — rather than saying "the model recommended it." Paired with the methodology explanation in-platform, it gives finance a clear answer to "how do we know this is backed by evidence?"

For the wider organisation

Prediction Confidence gives marketing, finance, and analytics a shared way to understand how a forecast has performed, so budget conversations start from the same evidence rather than different levels of trust in the number.

The goal isn't a perfect score — it's evidence you can act on. A strong score gives you confidence to move. A lower score tells you to test more tentatively before committing further budget. Both help you make decisions grounded in a track record, not a guess.

Troubleshooting

I can't see a Prediction Confidence score on my saturation curve

Prediction Confidence is currently available to a select group of customers. If you don't see a score, your account may not yet have access. Contact your account team to find out when it will be available for you.

My confidence score has changed since I last looked

That's expected. The backtest runs on a rolling 30-day window, so the score updates as new data comes in.

My confidence score seems lower than I'd expect

A low score isn't a red flag on the channel — it's a signal to test more tentatively rather than scale hard. It can reflect unusual market conditions, a recent significant spend change, or limited recent data for that channel. It isn't something you can adjust directly; as more recent data comes in, the score will update. Your account team can help you interpret it.

Frequently Asked Questions

What does the Prediction Confidence score measure?

It shows how many of the last 30 days the revenue outcome for a channel landed inside the model's predicted range. A score of 80% means this happened on 24 of those 30 days.

Does a strong score mean the forecast will be exactly right next time?

Not exactly. It shows the forecast has held up recently — it doesn't guarantee every future prediction will be precise. It's a track record, not a promise.

How is the predicted range calculated?

For each day in the backtest window, the model generates a lower and upper bound for the expected revenue outcome. The confidence score reflects how often that outcome fell within those bounds. Click How is this calculated? on any saturation curve for the full explanation.

Does Prediction Confidence change my underlying forecasts or any numbers I see?

No. The Incremental Forecasting model and saturation curves are unchanged. Prediction Confidence is a read-only signal based on automated backtesting — it doesn't recalculate any forecasted ROAS, conversions, or revenue numbers.

If Fospha checks its own model against itself, how is that not circular?

It's a fair question. The comparison is anchored to your total revenue: every channel's modelled contribution sums to 100% of what your business actually made, so nothing is inflated. Prediction Confidence checks how well we predicted the distribution of that real total across channels — not an independent third-party benchmark, but a consistent, auditable check against your own top-line numbers.

What happens if my score is low? Can I improve it?

A low score is a prompt to test more tentatively, not a sign to avoid the channel. It isn't something you set directly — it reflects how consistently the forecast has matched outcomes, and it refreshes as more recent data comes in. As the picture builds, confidence can improve and support a bigger move.

Why does confidence vary across channels?

Each channel's score reflects that channel's own backtest results. Channels with more consistent recent data tend to score higher; channels with recent spend volatility or limited history may score lower.

Who can see Prediction Confidence?

It's currently available to a select group of customers. Speak to your account team if you'd like to know more about timing for wider availability.