After creating the product configuration strategy, you can :

• Preview the recommendations
• Publish or unpublish the strategy
• Edit the strategy as per your needs
• View the model status

Preview the recommendations

Use the preview feature to view the results of the strategies before you publish them on the channel. This helps you see the recommendations that are generated by the algorithm, and how the business rules impact the recommendations. 

You can then verify whether you created the right strategy and entered enough amount of Lookback time frame to view the relevant recommendations on your channel.

The product recommendation models run once a day, at UTC 00:00:00. When you save the strategy for the first time, its preview will be available after the next model run. For the subsequent runs of the strategy, the preview is available immediately but shows the recommendations as per the last run of the strategy. Preview shows the results only after the model runs and comes up with recommendations. Preview is not available if there are no recommendations generated for the strategy.

If you had configured both primary and fallback strategy for the zone, and if the primary strategy does not provide enough recommendations, then recommendations returned from the fallback strategy are appended to the results and displayed.

Preview is not available for Most Recent algorithm as this algorithm generates results on-the-fly based on the latest activity of visitor on the channel.  

Example

Consider a scenario in which you create a product recommendation strategy for a zone, by selecting an algorithm, Lookback time frame, and other relevant parameters. Let us name this model as ProdRecsModel1. Its preview will be available after the models are executed at UTC 00:00:00.

Now, on the next day, if you create another product recommendation strategy for another zone, using the same combination of algorithm, Lookback time frame and other parameters as that of ProdRecsModel1, then the preview for the new model will be immediately available, without the need to wait for another day.

The preview shows the recommendation data as of the time that the last model was successfully run. For example, if you update the algorithm or Lookback time frame that is defined for your strategy, then the models need to be run again to show the updated results in the preview. If you create another strategy for another zone, with same algorithm and time frame, then the model runs once for multiple zones.

Product ID in preview

To preview the recommendations, you need to enter the product ID for which you want to verify the recommendations. In case of View to buy algorithms, you may enter one or more than one product ID(s) for which you want to verify the recommendations.

View to view and Buy to buy algorithms use collaborative filtering that encompasses techniques for matching similar products based on user interactions and makes recommendations on this basis. Hence, you need to enter one product ID to view the recommendations in preview.

Procedure

You must save your strategy to activate the Preview button. When you save the strategy, it is scheduled to be run daily to generate recommendations. Preview will be available after the next model run.

To preview a product recommendation strategy, complete the following steps:

1. In the product recommendation strategy pane, click Preview. 
2. In the preview window, enter the attributes as per the selected algorithm and model input combination. Product ID is case-sensitive. For more details refer the following tables in this topic. 
   
   
   Skip this step if you had selected Most popular as your algorithm, as it does not require any input.
3. Click Run preview.

Top 10 recommendations for the strategy are displayed on the zone.   

When the strategy is published on the channel, the model input and business rule inputs change based on the web page that the user is visiting and the user behavior activity in the current and previous sessions. Hence, the recommendations shown in the Preview may differ than the actual recommendations shown on the zone.

The following tables show the values provided for the Type of input data and Model Input fields, whether a text box is shown to specify further parameters, and the content displayed when you click Run preview.  These combinations provide you the flexibility to create strategies with a variety of possible combinations for input data, thus creating recommendations that are much more relevant and personalized for the channel visitor. 

Preview for Most popular algorithm

Value selected for Type of input data	Value selected for Model input	Value to be entered in preview text box	What is displayed in Preview
None	NA	No text box is shown. Model inputs are not required.	Product IDs that are most popular across all categories
Web page	Category ID/ Most recently viewed category / Most popular viewed category	Category ID	Product IDs that are a part of the specified category ID
Web page	Attributes	Values for attributes selected in model Input	Product IDs that are a part of the specified  attribute combination

 Preview for View to view/Buy to buy algorithms

Value selected for Type of input data	Value selected for Model input	Value to be entered in the preview text box	What is displayed in Preview
None	NA	Product ID	Recommendations based on the Model input for the specified Product ID, category, or attribute list
Web page	Product ID
Web page	Category ID
Web page	Attributes
User behavior	Most recently viewed product ID / Most recently carted product ID / Most recently purchased product ID / Most popular viewed product ID /  Most popular carted product ID / Most popular purchased product ID / Largest Purchase product ID / Most recently viewed category / Most popular viewed category

Preview for View to buy/Abandon to buy algorithms

Value selected for Type of input data	Value selected for Model input	Value to be entered in the preview  text box	What is displayed in Preview
None	NA	Product ID	Recommendations based on the Model input for the specified product ID
Web page	Product ID
User behavior	Most recently viewed product ID / Most recently carted product ID / Most recently purchased product ID / Most popular viewed product ID / Most popular carted product ID / Most popular purchased product ID / Largest Purchase product ID

 

Publish a product recommendation strategy

You must publish a product recommendation strategy to make it available for use on your channel.

To publish a product recommendation strategy, complete the following steps:

1. In the product recommendation strategy pane, click Publish. Confirmation pop up appears.
2. To publish the strategy, click Yes. Your strategy is published successfully and becomes active on the channel.

The product recommendations models are run every 24 hours, at UTC 00:00:00 to generate the recommendations. The models are run automatically, and you do not need to schedule them manually. The recommendations that are generated are displayed on the zone. New recommendation are not shown until the next run completes. If a model fails, the recommendations from its previous successful run are used and displayed on the zone.

Tip: You can track the status of your product recommendation models on the Model status page.

After you publish the strategy, the zone on which you had created the strategy provides the following details: 

• Strategy name
• Recommendations applied, which is the number of times the product recommendation strategy was applied on that Zone
• Created date or last published date

When you delete a strategy, the raw recommendations generated for the strategy are not deleted. If you create a new zone and strategy by selecting the same values for algorithm, Lookback time frame, and other parameters as entered in the previous strategy, then the preview will show the same recommendations as generated previously.  

Unpublish a strategy

You may want to unpublish (deactivate) the currently running strategy on a zone, for example because you want to replace it with another strategy, or you want to edit the current strategy. Before editing an active product recommendation strategy, you must unpublish it first. Perform the following steps:

1. Go to the Zones page. Select the relevant channel and the zone. 
2. Click the zone header to open the product recommendation strategy you want to edit.
3. In the strategy page, click Unpublish.

You can now edit the product recommendation strategy.

Edit a product recommendation strategy

Before editing a product recommendation strategy, you must unpublish it first. You can edit a product recommendation strategy to update the recommendation name, lookback time period, algorithm, and business rules of the primary or fallback strategy.

1. Go to the Zones page. Select the channel.
2. For the zone with the recommendation you want to edit, click Actions > Edit recommendation.
   
   
3. When the recommendation opens, update the required fields.
   Click Edit condition to edit business rules.
4. After you have made the required updates, you can do any one of the following actions:
   • If you want to only save the edited strategy without publishing it on your channel, click Save.
   • If you want to save and publish the edited strategy on your channel, click Save and publish. 

You can publish the strategy again after it is edited. After you publish the updated strategy, the new recommendations will be available after the models run at the next scheduled time (00:00:00 UTC).  

View the model status

Use the Model status page to track the latest status of your algorithmic models, with the details such as last run date & time, the algorithm used, its parameters, Lookback time frame value, and so on.

To view the model status, perform the following steps:

1. Go to Settings > Model status.
2. Select the channel for which you want to view the model status. The channels in the list are sorted in alphabetical order. By default, the model status information for the first channel is displayed on the page. You can select the specific channel for which you want to view the model status information. 
   

The Model run status table provides a summary of the model.

Field	Description
Model	Name of the model, based on the combination of algorithm name, parameters, Fallback (if any), and the Lookback time frame value.
Product catalog	Name of the product catalog used by the model. For a new model, these details are available after the first successful model run.
Used in	Zone(s) in which the model is currently being used. Click the link to view the zone names and the strategy names. For a new model, these details are available after the first successful model run. Hence, for a model with status as Scheduled, the Used in link is not active. Also, if you update the zone name or strategy name, the new names are shown in this screen after the next model run.
Last run on	The date and time when the model was last run
Status	Current status of the model. See the section "Understanding the model status" for a detailed description of each status.
Table 2. Model status

To view more details about a model, click its twistie icon.

Model details section appears with the following details.

Status	Description
Algorithm	Algorithm used for the model
Lookback	Value specified for the Lookback time frame (1 day/ 7 days/  30 days)
Type of input data	Option selected for the Type of input data parameter, such as None, Web page, or User behavior
Model input	Option selected for the Model input parameter, depending on the Type of input data
Input attributes	If Type of input data is Web page and Model input is Attributes, then this column shows the attribute names in the order they are processed.
Last successful run	The latest date and time on which the model was successfully run
Table 3. Model status details

The results for Recently viewed algorithm are generated on-the-fly based on the latest activity of visitor on the channel. Since the algorithm does not have a model associated with it, it is not listed on the Model status page.

2.   Zone Name or strategy name will not be available on model status dashboard (Used In link) until ML model is run for a new model.

 

Understanding the model status

The possible values of model status are as follows:

Scheduled

The model is newly created or modified and has not been run yet. The models with this status will be run at the next 00:00:00 UTC. After this, the model status will no longer be displayed as Scheduled, and it will be other possible values shown below.

Even if the status is Scheduled, and if there was a model with same parameters that was successfully run at least once in the past, then the recommendations will be available in preview, based on the older data. When the model is run at the next scheduled time and generates recommendations successfully, the preview will show new recommendations. 

Initiated

When the model run begins, a model shows this status when it is ready but hasn't yet begun processing. If you reload the status page after a few minutes, then this status would change to either Running, Success, or No recommendations generated.

Running

The model is currently being processed and product recommendations are being computed.

Success

The model has completed the processing and recommendations are available. This can be verified using the Preview option in the Zone details page for this model.

No recommendations generated

In the case the model run does not successfully generate any product recommendations, it is indicated by an error icon along with a link that provides more details about the status.

Here are some of the actions you can take to identify the reason for failure:

1. Check if you can see the preview of the respective model on the zone detail page. If the preview displays the recommendations, it means the model has been executed successfully in the past at least once, and the recommendations from its past successful run are shown in the preview (and on website if the strategy is published), so that the zone always has recommendations to show. 
2. Click the View details link. A pop-up displays reason for the failure and possible actions you can take.
3. If the preview does not show any recommendations, then either the model has not been successfully executed even once in the past or there may be an issue in retrieving recommendations. In this case, you can contact Acoustic Support.
4. Most of the times, the reason of failure is the unavailability of the data required for model execution for a given time frame when the model was run, that is, every 24 hours (00:00:00 UTC). If the tooltip suggests increasing time frame and if you have already selected a time frame of 30 days, you can contact Acoustic Support.