Model Comparison reference | Prescient AI Docs

This page lists the parts of the Model Comparison page, which opens from the Compare configs button of a model detail page.

In a label on this page, a word in braces is a value from the account. The target is the outcome that the model measures, for example revenue. The efficiency is the efficiency metric of the target, for example ROAS. The prefix is a letter that comes from the target.

Selectors and timeframes

Control or rule	Value
Menu entry	Model Comparison has no item in the sidebar.
Model A and Model B	Each selector lists first the configs that have a comparison with the config in the other selector. The other configs are under Models in other comparison sets. When you select one of them, the page changes the other selector to a config that it can be compared with.
Timeframes that you cannot select	A timeframe is not available when the 2 models have no comparison for that timeframe.
End of the timeframe	A timeframe ends on the last day that the 2 models both have reported data for.
Name of a model in a selector	The channel name, then the config name, for example the channel name with Default.
Back arrow	The back arrow returns to the page that you opened the comparison from.

Models	Timeframes	Label
Daily models	30, 90, 365, or 730 days	The selector shows Last `{count}` days.
Weekly models	4, 12, 52, or 104 weeks	The selector shows Last `{count}` weeks.

A word in braces stands for a value from your account or your plan.

Sections

Section	What it shows
Comparing Models	The 2 model selectors and the dates of the comparison.
Model Selection Criteria	A Preference Score for each model and the criteria behind the score.
MMM Fit	The modeled totals of the 2 models against the reported total, by date.
MMM Metric Comparison	The modeled total and the modeled efficiency of each model, with the difference.
Historical `{efficiency}`	The modeled efficiency of each model by channel and by timeframe, as a chart and a table.

A word in braces stands for a value from your account or your plan.

Model Selection Criteria

The section shows one panel for each model. Each panel has a gauge with the Preference Score of the model, from 0 to 100. The panel of the preferred model has the Recommended badge.

Rule	Value
Letter before -MMM	The letter comes from the metric of the target. A revenue target shows R.
Run that the comparison uses	The comparison shows the figures of the run in Run compared `{date}`. A later run of the config does not change the comparison. Model Center shows the latest run.
Value NA	A criterion can show NA in place of a number. NA has no color.
Result text	The text under the result title comes from the comparison data of the workspace.

A word in braces stands for a value from your account or your plan.

Additional Metrics

Criterion	What it shows
`{prefix}`-MMM fit	How well the model fits the reported total. A value near 100 is a better fit.
Average expectedness of `{efficiency}` values	How typical the efficiency values of the model are. A value near 1 is typical. A value near 0 is unusual.
`{efficiency}` outliers	At least 1 outlier or No outliers.
`{prefix}`-MMM Paid % within bounds	The paid percentage of the model, as a number. The tooltip gives the range that the product calls reasonable.

A word in braces stands for a value from your account or your plan.

Colors of the values

Value	Colors	Where
Fit	Green at 85 or more. Orange from 75 to less than 85. Red below 75.	The `{prefix}`-MMM fit row.
Expectedness	Green above 0.7. Orange above 0.4, up to 0.7. Red at 0.4 or less.	The Average expectedness of `{efficiency}` values row.
Paid %	Green from 30 to 100. Red for each other value.	The `{prefix}`-MMM Paid % within bounds row.
Preference Score	Green at 60 or more. A different color below 60.	The gauge of each model.

A word in braces stands for a value from your account or your plan.

Result

Result	Meaning
`{model}` Preferred	Model A Preferred or Model B Preferred. The comparison prefers the model with the higher Preference Score.
Equally Preferred	The two scores are equal, or one of the models has no score.

A word in braces stands for a value from your account or your plan.

Tooltips of the criteria

Criterion	Tooltip
Preference Score	The Preference Score (0–100) measures model quality across different criteria.
`{prefix}`-MMM fit	This shows how well Prescient's model fits reported total. Values closer to 100 indicate better fit.
Average expectedness of `{efficiency}` values	It measures how “typical” your `{efficiency}` is compared to the industry average. A score near 1 means your results are standard and expected, while a score near 0 indicates an outlier.
`{efficiency}` outliers	Outlier: A performance value significantly outside the norm (±4.5 STDEV from the channel mean).
`{prefix}`-MMM Paid % within bounds	Reasonable `{prefix}`-MMM Paid % values are roughly above 30% but no higher than 100%.

A word in braces stands for a value from your account or your plan.

Results sections

Series, MMM Fit	What it shows
Reported Total	The total that your sources reported.
Model A	The modeled total of Model A.
Model B	The modeled total of Model B.

Metric card	What it shows
MMM `{target}`	The modeled total that each model gives to paid media.
MMM `{efficiency}`	The modeled efficiency of each model.
Difference	The change from Model A to Model B, as a percentage.

A word in braces stands for a value from your account or your plan.

Rule	Value
Chart legend	Select a series in a chart legend to hide or show the series.
Difference	The percentage is the value of Model B minus the value of Model A, divided by the value of Model A. The product shows 0% when one of the values is zero.
Historical table columns	One group of columns for each timeframe, with a Model A column and a Model B column. A cell without a value shows a hyphen.
Target without a model	The MMM Fit section tells you when one of the models has no model for the selected target. Select a different target.

Empty states

When the config in Model A has no config to be compared with, the page shows the alert No comparison models available for {model}. A different config in Model A can have a comparison.

Message	Text under the message
No comparisons available yet	Model comparisons will appear here once your Prescient team has built comparable model configurations for this organization.
No comparison data available	No comparison data has been found for this model.
No historical data available	No historical data has been found for this specific model config for any timeframe. Please try selecting a different one.

Model Center: the guides and the concepts of the product area.
Compare two configs: the steps to open and read a comparison.
Configs and the Preference Score: what the score and the criteria mean.