Send-Time Optimization
“For each user, will they open an email sent in the next 4-hour window?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
For each user, will they open an email sent in the next 4-hour window?
Email platforms send campaigns at a single time chosen by the marketer — usually 10am Tuesday. But users open emails at vastly different times: early risers check at 6am, commuters at 8am, night owls at 11pm. Sending at the wrong time means the email gets buried under 20 newer messages. Open rates for batch-sent emails average 18-22% when personalized send times can push them to 28-35%. For a brand sending 50M emails monthly, each 1% open rate improvement is worth $1-2M annually in downstream revenue.
Quick answer
Send-time optimization predicts when each user is most likely to open an email by scoring open probability across time windows throughout the day. Instead of sending to everyone at 10am Tuesday, the model identifies that User A opens at 6:30am weekdays while User B opens at 11pm weekends, lifting open rates by 15-25% over batch sending.
Approaches compared
4 ways to solve this problem
1. Batch Send at 'Best Time'
Send all emails at the time with the highest aggregate open rate (usually 10am Tuesday or Thursday). The default for most email marketing platforms.
Best for
Small email lists where the complexity of personalized timing is not worth the lift. Also a reasonable baseline for comparison.
Watch out for
One time is never best for all users. The email arrives at 7am Pacific for East Coast early risers (too late) and 10am Pacific for night owls (too early). The email gets buried under newer messages by the time the user checks their inbox.
2. Timezone-Based Sending
Send at the 'best time' adjusted for each user's timezone. 10am local time for everyone. A simple improvement over batch sending.
Best for
Global email lists with users across many timezones where the timezone effect dominates individual preferences.
Watch out for
Assumes all users in a timezone check email at the same time. An early riser in New York who checks email at 6am and a night owl in New York who checks at 11pm both get the email at 10am. The timezone is a proxy, not a solution.
3. Historical Open-Time Bucketing
Compute each user's most common open hour from historical data and send during that window. Simple analytics, no ML required.
Best for
Users with long, consistent email engagement histories and stable daily routines.
Watch out for
Fails for users with sparse history (new subscribers). Cannot capture weekday vs. weekend patterns, seasonal shifts, or the fact that the optimal time varies by email type (promotional vs. transactional). Also does not handle users whose habits change over time.
4. KumoRFM (Graph Neural Networks on Relational Data)
Predicts open probability for each 4-hour window using binary classification on the user-email-open graph. Captures weekday vs. weekend patterns, campaign type preferences, and cross-user timezone-segment patterns. Enables predictions even for users with sparse history by propagating signals from graph neighbors.
Best for
High-volume email senders (10M+ monthly) where each percentage point of open rate improvement drives significant revenue.
Watch out for
Requires email send and open data with timestamps. If your ESP does not track individual open events (only aggregate open rates), you cannot build user-level send-time models.
Key metric: Personalized send-time optimization lifts open rates by 15-25% over batch sending. For brands sending 50M+ emails monthly, this drives $3-6M in incremental annual revenue.
Why relational data changes the answer
User U001 (America/New_York, early_riser segment) opens 78% of emails between 6-7am on weekdays but shifts to 10am on weekends. A simple historical bucketing would pick 6-7am as the optimal window. But the relational graph adds nuance: on weekdays, U001 opens promotional emails 1.4x faster than transactional ones during the early morning window. Time since last email open is 18 hours, suggesting U001 is 'due' for another inbox check. And 83% of users in the same timezone-segment graph neighborhood open before 7am on weekdays, confirming the pattern even for weeks where U001's own data is sparse.
The cross-user propagation is especially valuable for new subscribers. A new subscriber in the early_riser segment in America/New_York has no personal open history, but the graph neighborhood provides a strong prior: send between 6-7am weekdays. As the subscriber's own open data accumulates, the model blends the graph prior with individual behavior, converging on a personalized optimal window within 2-3 weeks. This cold-start handling is impossible for models that only use individual user history. For brands sending 50M+ emails monthly, personalized send-time optimization lifts open rates by 15-25%, translating to $3-6M in incremental annual revenue from downstream engagement and conversion.
Batch-sending email at 10am is like a newspaper that delivers to every house at the same time. Send-time optimization is like a newspaper that knows Mr. Johnson checks his porch at 5:30am while Ms. Garcia grabs hers at noon, and delivers accordingly. The same content, dramatically better engagement, because the timing matches each reader's natural rhythm.
How KumoRFM solves this
Relational intelligence for true personalization
Kumo predicts whether each user will open an email in the next 4-hour window using binary classification on the user-email-open graph. By scoring each window throughout the day, the system identifies the optimal send time per user. The model captures that User U001 opens emails at 6:30am on weekdays but 10am on weekends, and that users in similar timezone-segment graph neighborhoods share open-time patterns — enabling predictions even for users with sparse history.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
USERS
| user_id | timezone | segment |
|---|---|---|
| U001 | America/New_York | early_riser |
| U002 | America/Los_Angeles | night_owl |
| U003 | Europe/London | commuter |
EMAIL_SENDS
| send_id | user_id | campaign_id | send_hour | timestamp |
|---|---|---|---|---|
| ES001 | U001 | CAMP01 | 6 | 2025-02-18 |
| ES002 | U001 | CAMP02 | 14 | 2025-02-19 |
| ES003 | U002 | CAMP01 | 22 | 2025-02-18 |
EMAIL_OPENS
| open_id | send_id | user_id | timestamp |
|---|---|---|---|
| EO001 | ES001 | U001 | 2025-02-18 06:32 |
| EO002 | ES003 | U002 | 2025-02-18 22:15 |
| EO003 | ES001 | U001 | 2025-02-18 06:45 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT COUNT(EMAIL_OPENS.*, 0, 4, hours) > 0 FOR EACH USERS.USER_ID
Prediction output
Every entity gets a score, updated continuously
| USER_ID | TIMESTAMP | TARGET_PRED | True_PROB |
|---|---|---|---|
| U001 | 2025-03-12 06:00 | True | 0.89 |
| U001 | 2025-03-12 14:00 | False | 0.14 |
| U002 | 2025-03-12 22:00 | True | 0.82 |
Understand why
Every prediction includes feature attributions — no black boxes
User U001 (America/New_York, early_riser segment)
Predicted: Will open email in 06:00-10:00 window — probability 0.89
Top contributing features
Historical open hour distribution
78% of opens between 6-7am
38% attribution
Weekday vs weekend pattern
Weekday — opens 2hrs earlier
22% attribution
Graph neighbors (same timezone + segment)
83% open before 7am weekdays
19% attribution
Time since last email open
18 hours (due for next check)
13% attribution
Campaign type affinity
Opens promotional emails 1.4x faster
8% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about send-time optimization
How much does send-time optimization improve open rates?
Personalized send timing improves email open rates by 15-25% compared to batch sending. The improvement varies by list diversity: lists with users across many timezones and behavior patterns see the largest lift. Lists where most users are in one timezone with similar habits see 8-12% improvement.
Does send-time optimization work for new subscribers?
Yes. Graph models propagate open-time patterns from similar users (same timezone, same segment, same acquisition channel) to new subscribers with no personal history. Within 2-3 weeks, the model has enough individual data to blend with the graph prior for personalized timing.
Should send time vary by email type?
Yes. Users often have different optimal times for different email types. A subscriber might open promotional emails during their morning commute but transactional emails (order confirmations, shipping updates) immediately regardless of time. Graph models learn these type-specific patterns from the email send and open data.
What is the revenue impact of better open rates?
For every 1% improvement in open rate, downstream metrics improve: click-through increases by 0.3-0.5%, and revenue per send increases by $0.02-0.04. For a brand sending 50M emails monthly, a 15% open rate improvement translates to $3-6M in incremental annual revenue from the resulting engagement cascade.
Bottom line: 15-25% improvement in email open rates by sending at each user's optimal time. For brands sending 50M+ emails monthly, this drives $3-6M in incremental annual revenue.
Related use cases
Explore more personalization use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
