Do all acounts in my segment get model ratings?
Not always. If you built the segment so that all accounts in it are right for the selling situation, then all accounts will be rated bu the Cross-sell model. On the other hand, you can also use Cross-sell ratings to surface the right targets for your selling situation from within a larger segment.
The following accounts in a segment will not get Cross-sell scores or ratings:
- Accounts that do not have any transactions in the system.
- Accounts that do not fit the selling situation, e.g. for the First Purchase mode, any accounts that have already purchased will not get model ratings.
When exporting model ratings, accounts that are not rated will have an empty rating.
When are account ratings updated?
Each time the Data Processing & Analysis job runs, all ratings for all active models are refreshed:
- Target accounts in the segment are updated.
- The updated target accounts are assigned new scores and ratings.
- The new scores and ratings are available in My Data.
- The Model Dashboard will show the updated ratings in the Lift chart.
If the updated ratings show a performance decline, try using the Remodel function to train a new model iteration that is also up to date with your data.
When can I update my rating buckets?
You can update the settings for your rating buckets any time. Keep in mind that the new rating buckets will be applied after the next Data Processing & Analysis job completes.
Update the rating buckets to adjust the number of targets receiving top ratings. This way, you can get the exact campaign membership you need to optimize budget and capacity.
What does Lift mean in Rev.Up?
A Lift for a rating bucket is a measure of how a given rating will perform compared to the average performance against the full set of targets. The average lift is always 1.0x. Lift is displayed for each rating bucket.
- For Likely to Buy models, Lift is the estimated conversion probability factor, compared to all targets.
- For Likely Amount of Spend models, lift is the factor of weighted revenue in the rating bucket, compared to all targets.
How can I use model ratings in workflows?
There are several important ways you can use model ratings:
- Use ratings to create new highly targeted segments by adding rating attributes to segment queries.
- Use ratings when defining the rating rules of a rule-based model to combine predictive dimensions within a campaign.
- Use ratings within a Play to launch a sales campaign through BIS.
For items 1 and 2, note that ratings may have only one level of dependency. Deeper nesting is not supported.
How does Rev.Up compute buying propensity?
Rev.Up uses binary classification modeling to predict buying probability (likely to buy).
How do I use revenue modeling to predict the likely amount of spend?
To combine revenue modeling in with propensity modeling, choose "Likely Amount of Spend" in the Prioritization step when creating a Cross-sell model. This will use buying propensity model combined with a regression model for revenue forecasting. Scores from the two models are combined to generate a weighted revenue forecast for prioritization.
What are the Cross-sell model training options?
When creating a Cross-sell model, the Training step shows several settings that can be used to engineer the training data for your specific situation.
These settings are:
- "Only use successes where spend in the month was...". This setting removes purchases from learning that do not meet the spending criteria. Use this when the Product Bundle pricing varies widely to focus the model on the business you are looking for.
- "Only use successes where quantity bought in the month was...". This setting removes purchases from learning that do not meet the quantity criteria. Use this when purchase quantity for the Product Bundle varies widely to focus the model on the business you are looking for.
- "When modeling, only use historical records from the last...". This Setting only uses the training data with the selected recency. Use this when something significant changed in the past (with account, transaction or product tracking), and you do not want to use older data in model training.
- "Let me model on a similar segment". This setting uses a separate filter for model training, but still rates the original segment. Data scientists use this to have better control over the training data.
"Let me model on similar products". This setting replaces the actual Product Bundle in your selling situation with a different Bundle that is used just for model training. The actual Bundle is still used for selecting target accounts. Data scientists use this when there is insufficient data to train the model based on the actual Bundle, so a similar one with fuller history is used instead.
How many purchases do I need in the training data?
You must have at least 50 First Purchases or Repeat Purchases to learn from. If you proceed with fewer, the model creation will fail.