Optimal Send Times Model Card

Model Release Date: June 2025
Model Version: v3.1
Model Type: Aggregation
Applicable Platform: Sprout Social

What It Does & Why It Matters

The Optimal Send Times (OST) model provides customers with suggested times for posting content on social platforms. Its primary goal is to increase engagement by recommending when a customer's audience is most likely to be online and interacting with the content. This feature aims to suggest "best" times to post content to maximize audience engagement (views, likes, shares, comments, etc.).

Intended Use Cases

• Assist users in determining the best time to make a social media post in order to gain engagement.

• Provide suggested post times per social network and profile, based on previous engagements.

Out-of-scope uses: 

• Guaranteeing specific engagement levels or content virality.
• Offering insights into what content to post, only when to post.
• Adapting to sudden, short-term shifts in audience behavior (such as breaking news events).

Factors & Limitations

• We can only provide the model with the engagement data available to us from each social network. 
• The model considers 16 weeks of historical engagement data for each customer profile. The historical engagement data is refreshed each Sunday of the week. 
• If a customer does not have enough historical engagement data to confidently generate profile-specific scores, they are provided with suggested times for the target network and their specific timezone based on similarly aggregated data.

Example Input: Historical engagement data for a specific social profile over the past 16 weeks, including timestamps of posts and corresponding engagement metrics.

Example Output: A score for every 5-minute increment for the upcoming week, indicating the optimal posting time. In the app, customers are provided with a ranking of the highest scoring 5-minute increments.

Evaluation Metrics

The effectiveness is measured ad hoc by whether customer engagement increases when using the suggested times, which are automatically updated as the model re-runs on a regular cadence.

Training Data

The model is not trained as the term is generally defined, but rather continually cross-references 16 weeks of historical engagement data for each connected profile. This data includes:

• Amount of engagement received from followers.
• Associated day of week and time of day for that engagement.

Evaluation Data

The model uses historical engagement data to generate scores, and there is no separate "evaluation dataset" in the traditional sense. 

Risks & Ethical Considerations

This model aims to empower users to maximize their content's reach and engagement. By leveraging historical audience behavior, it helps users post when their content is most likely to be seen and interacted with. While the model provides a fallback for profiles with insufficient data, it's important to acknowledge that the quality of personalized recommendations depends on the volume and consistency of a profile’s past engagement. It is also important to note that because the model suggests optimal posting times regardless of content, it could potentially be exploited to amplify political messages, propaganda, or other harmful content more effectively.